version: bump up to v3.0.17+git

Was triggering cancelation errors on outstanding KeepAlives if Grant
had to retry.

.gitignore

@@ -10,3 +10,4 @@
 /hack/insta-discovery/.env
 *.test
 tools/functional-tester/docker/bin
+hack/tls-setup/certs

.travis.yml

@@ -4,9 +4,7 @@ go_import_path: github.com/coreos/etcd
 sudo: false
 
 go:
-  - 1.5
+  - 1.6.4
-  - 1.6
-  - tip
 
 env:
   global:
@@ -22,10 +20,6 @@ matrix:
   allow_failures:
     - go: tip
   exclude:
-  - go: 1.5
-    env: TARGET=arm
-  - go: 1.5
-    env: TARGET=ppc64le
   - go: 1.6
     env: TARGET=arm64
   - go: tip

Dockerfile-release

@@ -1,8 +1,9 @@
 FROM alpine:latest
 
-ADD bin/etcd /usr/local/bin/
+ADD etcd /usr/local/bin/
-ADD bin/etcdctl /usr/local/bin/
+ADD etcdctl /usr/local/bin/
 RUN mkdir -p /var/etcd/
+RUN mkdir -p /var/lib/etcd/
 
 EXPOSE 2379 2380
 

Documentation/dev-guide/api_grpc_gateway.md

@@ -25,13 +25,13 @@ curl -L http://localhost:2379/v3alpha/kv/range \
 
 ## Swagger
 
-Generated [Swapper][swagger] API definitions can be found at [rpc.swagger.json][swagger-doc].
+Generated [Swagger][swagger] API definitions can be found at [rpc.swagger.json][swagger-doc].
 
 [api-ref]: ./api_reference_v3.md
 [go-client]: https://github.com/coreos/etcd/tree/master/clientv3
 [etcdctl]: https://github.com/coreos/etcd/tree/master/etcdctl
 [grpc]: http://www.grpc.io/
-[grpc-gateway]: https://github.com/gengo/grpc-gateway
+[grpc-gateway]: https://github.com/grpc-ecosystem/grpc-gateway
 [json-mapping]: https://developers.google.com/protocol-buffers/docs/proto3#json
 [swagger]: http://swagger.io/
 [swagger-doc]: apispec/swagger/rpc.swagger.json

Documentation/dev-guide/api_reference_v3.md

@@ -427,6 +427,7 @@ Empty field.
 | ----- | ----------- | ---- |
 | key | key is the first key to delete in the range. | bytes |
 | range_end | range_end is the key following the last key to delete for the range [key, range_end). If range_end is not given, the range is defined to contain only the key argument. If range_end is '\0', the range is all keys greater than or equal to the key argument. | bytes |
+| prev_kv | If prev_kv is set, etcd gets the previous key-value pairs before deleting it. The previous key-value pairs will be returned in the delte response. | bool |
 
 
 
@@ -436,6 +437,7 @@ Empty field.
 | ----- | ----------- | ---- |
 | header |  | ResponseHeader |
 | deleted | deleted is the number of keys deleted by the delete range request. | int64 |
+| prev_kvs | if prev_kv is set in the request, the previous key-value pairs will be returned. | (slice of) mvccpb.KeyValue |
 
 
 
@@ -591,6 +593,7 @@ Empty field.
 | key | key is the key, in bytes, to put into the key-value store. | bytes |
 | value | value is the value, in bytes, to associate with the key in the key-value store. | bytes |
 | lease | lease is the lease ID to associate with the key in the key-value store. A lease value of 0 indicates no lease. | int64 |
+| prev_kv | If prev_kv is set, etcd gets the previous key-value pair before changing it. The previous key-value pair will be returned in the put response. | bool |
 
 
 
@@ -599,6 +602,7 @@ Empty field.
 | Field | Description | Type |
 | ----- | ----------- | ---- |
 | header |  | ResponseHeader |
+| prev_kv | if prev_kv is set in the request, the previous key-value pair will be returned. | mvccpb.KeyValue |
 
 
 
@@ -613,6 +617,8 @@ Empty field.
 | sort_order | sort_order is the order for returned sorted results. | SortOrder |
 | sort_target | sort_target is the key-value field to use for sorting. | SortTarget |
 | serializable | serializable sets the range request to use serializable member-local reads. Range requests are linearizable by default; linearizable requests have higher latency and lower throughput than serializable requests but reflect the current consensus of the cluster. For better performance, in exchange for possible stale reads, a serializable range request is served locally without needing to reach consensus with other nodes in the cluster. | bool |
+| keys_only | keys_only when set returns only the keys and not the values. | bool |
+| count_only | count_only when set returns only the count of the keys in the range. | bool |
 
 
 
@@ -621,8 +627,9 @@ Empty field.
 | Field | Description | Type |
 | ----- | ----------- | ---- |
 | header |  | ResponseHeader |
-| kvs | kvs is the list of key-value pairs matched by the range request. | (slice of) mvccpb.KeyValue |
+| kvs | kvs is the list of key-value pairs matched by the range request. kvs is empty when count is requested. | (slice of) mvccpb.KeyValue |
 | more | more indicates if there are more keys to return in the requested range. | bool |
+| count | count is set to the number of keys within the range when requested. | int64 |
 
 
 
@@ -732,6 +739,7 @@ From google paxosdb paper: Our implementation hinges around a powerful primitive
 | range_end | range_end is the end of the range [key, range_end) to watch. If range_end is not given, only the key argument is watched. If range_end is equal to '\0', all keys greater than or equal to the key argument are watched. | bytes |
 | start_revision | start_revision is an optional revision to watch from (inclusive). No start_revision is "now". | int64 |
 | progress_notify | progress_notify is set so that the etcd server will periodically send a WatchResponse with no events to the new watcher if there are no recent events. It is useful when clients wish to recover a disconnected watcher starting from a recent known revision. The etcd server may decide how often it will send notifications based on current load. | bool |
+| prev_kv | If prev_kv is set, created watcher gets the previous KV before the event happens. If the previous KV is already compacted, nothing will be returned. | bool |
 
 
 
@@ -764,6 +772,7 @@ From google paxosdb paper: Our implementation hinges around a powerful primitive
 | ----- | ----------- | ---- |
 | type | type is the kind of event. If type is a PUT, it indicates new data has been stored to the key. If type is a DELETE, it indicates the key was deleted. | EventType |
 | kv | kv holds the KeyValue for the event. A PUT event contains current kv pair. A PUT event with kv.Version=1 indicates the creation of a key. A DELETE/EXPIRE event contains the deleted key with its modification revision set to the revision of deletion. | KeyValue |
+| prev_kv | prev_kv holds the key-value pair before the event happens. | KeyValue |
 
 
 

Documentation/dev-guide/apispec/swagger/rpc.swagger.json

@@ -15,13 +15,553 @@
     "application/json"
   ],
   "paths": {
+    "/v3alpha/auth/authenticate": {
+      "post": {
+        "summary": "Authenticate processes an authenticate request.",
+        "operationId": "Authenticate",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthenticateResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthenticateRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/disable": {
+      "post": {
+        "summary": "AuthDisable disables authentication.",
+        "operationId": "AuthDisable",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthDisableResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthDisableRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/enable": {
+      "post": {
+        "summary": "AuthEnable enables authentication.",
+        "operationId": "AuthEnable",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthEnableResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthEnableRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/add": {
+      "post": {
+        "summary": "RoleAdd adds a new role.",
+        "operationId": "RoleAdd",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleAddResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleAddRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/delete": {
+      "post": {
+        "summary": "RoleDelete deletes a specified role.",
+        "operationId": "RoleDelete",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleDeleteResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleDeleteRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/get": {
+      "post": {
+        "summary": "RoleGet gets detailed role information.",
+        "operationId": "RoleGet",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleGetResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleGetRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/grant": {
+      "post": {
+        "summary": "RoleGrantPermission grants a permission of a specified key or range to a specified role.",
+        "operationId": "RoleGrantPermission",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleGrantPermissionResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleGrantPermissionRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/list": {
+      "post": {
+        "summary": "RoleList gets lists of all roles.",
+        "operationId": "RoleList",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleListResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleListRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/role/revoke": {
+      "post": {
+        "summary": "RoleRevokePermission revokes a key or range permission of a specified role.",
+        "operationId": "RoleRevokePermission",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleRevokePermissionResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthRoleRevokePermissionRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/add": {
+      "post": {
+        "summary": "UserAdd adds a new user.",
+        "operationId": "UserAdd",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserAddResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserAddRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/changepw": {
+      "post": {
+        "summary": "UserChangePassword changes the password of a specified user.",
+        "operationId": "UserChangePassword",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserChangePasswordResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserChangePasswordRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/delete": {
+      "post": {
+        "summary": "UserDelete deletes a specified user.",
+        "operationId": "UserDelete",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserDeleteResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserDeleteRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/get": {
+      "post": {
+        "summary": "UserGet gets detailed user information.",
+        "operationId": "UserGet",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserGetResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserGetRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/grant": {
+      "post": {
+        "summary": "UserGrant grants a role to a specified user.",
+        "operationId": "UserGrantRole",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserGrantRoleResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserGrantRoleRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/list": {
+      "post": {
+        "summary": "UserList gets a list of all users.",
+        "operationId": "UserList",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserListResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserListRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/auth/user/revoke": {
+      "post": {
+        "summary": "UserRevokeRole revokes a role of specified user.",
+        "operationId": "UserRevokeRole",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserRevokeRoleResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAuthUserRevokeRoleRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Auth"
+        ]
+      }
+    },
+    "/v3alpha/cluster/member/add": {
+      "post": {
+        "summary": "MemberAdd adds a member into the cluster.",
+        "operationId": "MemberAdd",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberAddResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberAddRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Cluster"
+        ]
+      }
+    },
+    "/v3alpha/cluster/member/list": {
+      "post": {
+        "summary": "MemberList lists all the members in the cluster.",
+        "operationId": "MemberList",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberListResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberListRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Cluster"
+        ]
+      }
+    },
+    "/v3alpha/cluster/member/remove": {
+      "post": {
+        "summary": "MemberRemove removes an existing member from the cluster.",
+        "operationId": "MemberRemove",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberRemoveResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberRemoveRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Cluster"
+        ]
+      }
+    },
+    "/v3alpha/cluster/member/update": {
+      "post": {
+        "summary": "MemberUpdate updates the member configuration.",
+        "operationId": "MemberUpdate",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberUpdateResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbMemberUpdateRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Cluster"
+        ]
+      }
+    },
     "/v3alpha/kv/compaction": {
       "post": {
-        "summary": "Txn processes multiple requests in a single transaction.\nA txn request increments the revision of the key-value store\nand generates events with the same revision for every completed request.\nIt is not allowed to modify the same key several times within one txn.",
+        "summary": "Compact compacts the event history in the etcd key-value store. The key-value\nstore should be periodically compacted or the event history will continue to grow\nindefinitely.",
         "operationId": "Compact",
         "responses": {
           "200": {
-            "description": "Description",
+            "description": "",
             "schema": {
               "$ref": "#/definitions/etcdserverpbCompactionResponse"
             }
@@ -42,13 +582,67 @@
         ]
       }
     },
+    "/v3alpha/kv/deleterange": {
+      "post": {
+        "summary": "DeleteRange deletes the given range from the key-value store.\nA delete request increments the revision of the key-value store\nand generates a delete event in the event history for every deleted key.",
+        "operationId": "DeleteRange",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbDeleteRangeResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbDeleteRangeRequest"
+            }
+          }
+        ],
+        "tags": [
+          "KV"
+        ]
+      }
+    },
+    "/v3alpha/kv/lease/revoke": {
+      "post": {
+        "summary": "LeaseRevoke revokes a lease. All keys attached to the lease will expire and be deleted.",
+        "operationId": "LeaseRevoke",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseRevokeResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseRevokeRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Lease"
+        ]
+      }
+    },
     "/v3alpha/kv/put": {
       "post": {
         "summary": "Put puts the given key into the key-value store.\nA put request increments the revision of the key-value store\nand generates one event in the event history.",
         "operationId": "Put",
         "responses": {
           "200": {
-            "description": "Description",
+            "description": "",
             "schema": {
               "$ref": "#/definitions/etcdserverpbPutResponse"
             }
@@ -75,7 +669,7 @@
         "operationId": "Range",
         "responses": {
           "200": {
-            "description": "Description",
+            "description": "",
             "schema": {
               "$ref": "#/definitions/etcdserverpbRangeResponse"
             }
@@ -98,11 +692,11 @@
     },
     "/v3alpha/kv/txn": {
       "post": {
-        "summary": "DeleteRange deletes the given range from the key-value store.\nA delete request increments the revision of the key-value store\nand generates a delete event in the event history for every deleted key.",
+        "summary": "Txn processes multiple requests in a single transaction.\nA txn request increments the revision of the key-value store\nand generates events with the same revision for every completed request.\nIt is not allowed to modify the same key several times within one txn.",
         "operationId": "Txn",
         "responses": {
           "200": {
-            "description": "Description",
+            "description": "",
             "schema": {
               "$ref": "#/definitions/etcdserverpbTxnResponse"
             }
@@ -122,6 +716,224 @@
           "KV"
         ]
       }
+    },
+    "/v3alpha/lease/grant": {
+      "post": {
+        "summary": "LeaseGrant creates a lease which expires if the server does not receive a keepAlive\nwithin a given time to live period. All keys attached to the lease will be expired and\ndeleted if the lease expires. Each expired key generates a delete event in the event history.",
+        "operationId": "LeaseGrant",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseGrantResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseGrantRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Lease"
+        ]
+      }
+    },
+    "/v3alpha/lease/keepalive": {
+      "post": {
+        "summary": "LeaseKeepAlive keeps the lease alive by streaming keep alive requests from the client\nto the server and streaming keep alive responses from the server to the client.",
+        "operationId": "LeaseKeepAlive",
+        "responses": {
+          "200": {
+            "description": "(streaming responses)",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseKeepAliveResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "description": "(streaming inputs)",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbLeaseKeepAliveRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Lease"
+        ]
+      }
+    },
+    "/v3alpha/maintenance/alarm": {
+      "post": {
+        "summary": "Alarm activates, deactivates, and queries alarms regarding cluster health.",
+        "operationId": "Alarm",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAlarmResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbAlarmRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Maintenance"
+        ]
+      }
+    },
+    "/v3alpha/maintenance/defragment": {
+      "post": {
+        "summary": "Defragment defragments a member's backend database to recover storage space.",
+        "operationId": "Defragment",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbDefragmentResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbDefragmentRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Maintenance"
+        ]
+      }
+    },
+    "/v3alpha/maintenance/hash": {
+      "post": {
+        "summary": "Hash returns the hash of the local KV state for consistency checking purpose.\nThis is designed for testing; do not use this in production when there\nare ongoing transactions.",
+        "operationId": "Hash",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbHashResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbHashRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Maintenance"
+        ]
+      }
+    },
+    "/v3alpha/maintenance/snapshot": {
+      "post": {
+        "summary": "Snapshot sends a snapshot of the entire backend from a member over a stream to a client.",
+        "operationId": "Snapshot",
+        "responses": {
+          "200": {
+            "description": "(streaming responses)",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbSnapshotResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbSnapshotRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Maintenance"
+        ]
+      }
+    },
+    "/v3alpha/maintenance/status": {
+      "post": {
+        "summary": "Status gets the status of the member.",
+        "operationId": "Status",
+        "responses": {
+          "200": {
+            "description": "",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbStatusResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbStatusRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Maintenance"
+        ]
+      }
+    },
+    "/v3alpha/watch": {
+      "post": {
+        "summary": "Watch watches for events happening or that have happened. Both input and output\nare streams; the input stream is for creating and canceling watchers and the output\nstream sends events. One watch RPC can watch on multiple key ranges, streaming events\nfor several watches at once. The entire event history can be watched starting from the\nlast compaction revision.",
+        "operationId": "Watch",
+        "responses": {
+          "200": {
+            "description": "(streaming responses)",
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbWatchResponse"
+            }
+          }
+        },
+        "parameters": [
+          {
+            "name": "body",
+            "description": "(streaming inputs)",
+            "in": "body",
+            "required": true,
+            "schema": {
+              "$ref": "#/definitions/etcdserverpbWatchRequest"
+            }
+          }
+        ],
+        "tags": [
+          "Watch"
+        ]
+      }
     }
   },
   "definitions": {
@@ -662,6 +1474,11 @@
           "format": "byte",
           "description": "key is the first key to delete in the range."
         },
+        "prev_kv": {
+          "type": "boolean",
+          "format": "boolean",
+          "description": "If prev_kv is set, etcd gets the previous key-value pairs before deleting it.\nThe previous key-value pairs will be returned in the delte response."
+        },
         "range_end": {
           "type": "string",
           "format": "byte",
@@ -679,6 +1496,13 @@
         },
         "header": {
           "$ref": "#/definitions/etcdserverpbResponseHeader"
+        },
+        "prev_kvs": {
+          "type": "array",
+          "items": {
+            "$ref": "#/definitions/mvccpbKeyValue"
+          },
+          "description": "if prev_kv is set in the request, the previous key-value pairs will be returned."
         }
       }
     },
@@ -912,6 +1736,11 @@
           "format": "int64",
           "description": "lease is the lease ID to associate with the key in the key-value store. A lease\nvalue of 0 indicates no lease."
         },
+        "prev_kv": {
+          "type": "boolean",
+          "format": "boolean",
+          "description": "If prev_kv is set, etcd gets the previous key-value pair before changing it.\nThe previous key-value pair will be returned in the put response."
+        },
         "value": {
           "type": "string",
           "format": "byte",
@@ -924,6 +1753,10 @@
       "properties": {
         "header": {
           "$ref": "#/definitions/etcdserverpbResponseHeader"
+        },
+        "prev_kv": {
+          "$ref": "#/definitions/mvccpbKeyValue",
+          "description": "if prev_kv is set in the request, the previous key-value pair will be returned."
         }
       }
     },
@@ -1176,6 +2009,11 @@
           "format": "byte",
           "description": "key is the key to register for watching."
         },
+        "prev_kv": {
+          "type": "boolean",
+          "format": "boolean",
+          "description": "If prev_kv is set, created watcher gets the previous KV before the event happens.\nIf the previous KV is already compacted, nothing will be returned."
+        },
         "progress_notify": {
           "type": "boolean",
           "format": "boolean",
@@ -1245,6 +2083,10 @@
           "$ref": "#/definitions/mvccpbKeyValue",
           "description": "kv holds the KeyValue for the event.\nA PUT event contains current kv pair.\nA PUT event with kv.Version=1 indicates the creation of a key.\nA DELETE/EXPIRE event contains the deleted key with\nits modification revision set to the revision of deletion."
         },
+        "prev_kv": {
+          "$ref": "#/definitions/mvccpbKeyValue",
+          "description": "prev_kv holds the key-value pair before the event happens."
+        },
         "type": {
           "$ref": "#/definitions/EventEventType",
           "description": "type is the kind of event. If type is a PUT, it indicates\nnew data has been stored to the key. If type is a DELETE,\nit indicates the key was deleted."

Documentation/dev-guide/experimental_apis.md

@@ -4,5 +4,5 @@ For the most part, the etcd project is stable, but we are still moving fast! We
 
 ## The current experimental API/features are:
 
-- v3 auth API: expect to be stale in 3.1 release
+- v3 auth API: expect to be stable in 3.1 release
 - etcd gateway: expect to be stable in 3.1 release

Documentation/dl_build.md

@@ -11,7 +11,7 @@ The easiest way to get etcd is to use one of the pre-built release binaries whic
 ## Build the latest version
 
 For those wanting to try the very latest version, build etcd from the `master` branch.
-[Go](https://golang.org/) version 1.5+ is required to build the latest version of etcd.
+[Go](https://golang.org/) version 1.6+ (with HTTP2 support) is required to build the latest version of etcd.
 
 Here are the commands to build an etcd binary from the `master` branch:
 

Documentation/op-guide/clustering.md

@@ -357,6 +357,8 @@ To help clients discover the etcd cluster, the following DNS SRV records are loo
 
 If `_etcd-client-ssl._tcp.example.com` is found, clients will attempt to communicate with the etcd cluster over SSL/TLS.
 
+If etcd is using TLS without a custom certificate authority, the discovery domain (e.g., example.com) must match the SRV record domain (e.g., infra1.example.com). This is to mitigate attacks that forge SRV records to point to a different domain; the domain would have a valid certificate under PKI but be controlled by an unknown third party.
+
 #### Create DNS SRV records
 
 ```

Documentation/op-guide/container.md

@@ -8,7 +8,7 @@ In order to expose the etcd API to clients outside of Docker host, use the host
 
 ```
 # For each machine
-ETCD_VERSION=v3.0.0-beta.0
+ETCD_VERSION=v3.0.0
 TOKEN=my-etcd-token
 CLUSTER_STATE=new
 NAME_1=etcd-node-0

Documentation/upgrades/upgrade_3_0.md

@@ -18,7 +18,7 @@ Also, to ensure a smooth rolling upgrade, the running cluster must be healthy. Y
 
 Before upgrading etcd, always test the services relying on etcd in a staging environment before deploying the upgrade to the production environment.
 
-Before beginning,  [backup the etcd data directory](admin_guide.md#backing-up-the-datastore). Should something go wrong with the upgrade, it is possible to use this backup to [downgrade](#downgrade) back to existing etcd version.
+Before beginning,  [backup the etcd data directory](../v2/admin_guide.md#backing-up-the-datastore). Should something go wrong with the upgrade, it is possible to use this backup to [downgrade](#downgrade) back to existing etcd version.
 
 #### Mixed Versions
 
@@ -34,7 +34,7 @@ For a much larger total data size, 100MB or more , this one-time process might t
 
 If all members have been upgraded to v3.0, the cluster will be upgraded to v3.0, and downgrade from this completed state is **not possible**. If any single member is still v2.3, however, the cluster and its operations remains “v2.3”, and it is possible from this mixed cluster state to return to using a v2.3 etcd binary on all members.
 
-Please [backup the data directory](admin_guide.md#backing-up-the-datastore) of all etcd members to make downgrading the cluster possible even after it has been completely upgraded.
+Please [backup the data directory](../v2/admin_guide.md#backing-up-the-datastore) of all etcd members to make downgrading the cluster possible even after it has been completely upgraded.
 
 ### Upgrade Procedure
 
@@ -64,7 +64,7 @@ When each etcd process is stopped, expected errors will be logged by other clust
 2016-06-27 15:21:48.624175 I | rafthttp: the connection with 8211f1d0f64f3269 became inactive
 ```
 
-It’s a good idea at this point to  [backup the etcd data directory](https://github.com/coreos/etcd/blob/master/Documentation/v2/admin_guide.md#backing-up-the-datastore) to provide a downgrade path should any problems occur:
+It’s a good idea at this point to  [backup the etcd data directory](../v2/admin_guide.md#backing-up-the-datastore) to provide a downgrade path should any problems occur:
 
 ```
 $ etcdctl backup \

README.md

@@ -40,7 +40,7 @@ See [etcdctl][etcdctl] for a simple command line client.
 The easiest way to get etcd is to use one of the pre-built release binaries which are available for OSX, Linux, Windows, AppC (ACI), and Docker. Instructions for using these binaries are on the [GitHub releases page][github-release].
 
 For those wanting to try the very latest version, you can build the latest version of etcd from the `master` branch.
-You will first need [*Go*](https://golang.org/) installed on your machine (version 1.5+ is required).
+You will first need [*Go*](https://golang.org/) installed on your machine (version 1.6+ is required).
 All development occurs on `master`, including new features and bug fixes.
 Bug fixes are first targeted at `master` and subsequently ported to release branches, as described in the [branch management][branch-management] guide.
 

auth/authpb/auth.pb.go

@@ -18,12 +18,12 @@ package authpb
 import (
 	"fmt"
 
-	proto "github.com/gogo/protobuf/proto"
+	proto "github.com/golang/protobuf/proto"
 
 	math "math"
-)
 
-import io "io"
+	io "io"
+)
 
 // Reference imports to suppress errors if they are not otherwise used.
 var _ = proto.Marshal
@@ -32,7 +32,7 @@ var _ = math.Inf
 
 // This is a compile-time assertion to ensure that this generated file
 // is compatible with the proto package it is being compiled against.
-const _ = proto.GoGoProtoPackageIsVersion1
+const _ = proto.ProtoPackageIsVersion1
 
 type Permission_Type int32
 
@@ -798,23 +798,23 @@ var (
 )
 
 var fileDescriptorAuth = []byte{
-	// 276 bytes of a gzipped FileDescriptorProto
+	// 288 bytes of a gzipped FileDescriptorProto
-	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0xe2, 0xe2, 0x4a, 0x2c, 0x2d, 0xc9,
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0x6c, 0x90, 0xc1, 0x4a, 0xc3, 0x30,
-	0xd0, 0x2b, 0x28, 0xca, 0x2f, 0xc9, 0x17, 0x62, 0x03, 0xb1, 0x0b, 0x92, 0xa4, 0x44, 0xd2, 0xf3,
+	0x1c, 0xc6, 0x9b, 0xb6, 0x1b, 0xed, 0x5f, 0x27, 0x25, 0x0c, 0x0c, 0x13, 0x42, 0xe9, 0xa9, 0x78,
-	0xd3, 0xf3, 0xc1, 0x42, 0xfa, 0x20, 0x16, 0x44, 0x56, 0xc9, 0x87, 0x8b, 0x25, 0xb4, 0x38, 0xb5,
+	0xa8, 0xb0, 0x5d, 0xbc, 0x2a, 0xf6, 0x20, 0x78, 0x90, 0x50, 0xf1, 0x28, 0x1d, 0x0d, 0x75, 0x6c,
-	0x48, 0x48, 0x88, 0x8b, 0x25, 0x2f, 0x31, 0x37, 0x55, 0x82, 0x51, 0x81, 0x51, 0x83, 0x27, 0x08,
+	0x6d, 0x4a, 0x32, 0x91, 0xbe, 0x89, 0x07, 0x1f, 0x68, 0xc7, 0x3d, 0x82, 0xab, 0x2f, 0x22, 0x4d,
-	0xcc, 0x16, 0x92, 0xe2, 0xe2, 0x28, 0x48, 0x2c, 0x2e, 0x2e, 0xcf, 0x2f, 0x4a, 0x91, 0x60, 0x02,
+	0x64, 0x43, 0xdc, 0xed, 0xfb, 0xbe, 0xff, 0x97, 0xe4, 0x97, 0x3f, 0x40, 0xfe, 0xb6, 0x7e, 0x4d,
-	0x8b, 0xc3, 0xf9, 0x42, 0x22, 0x5c, 0xac, 0x45, 0xf9, 0x39, 0xa9, 0xc5, 0x12, 0xcc, 0x0a, 0xcc,
+	0x1a, 0x29, 0xd6, 0x02, 0x0f, 0x7b, 0xdd, 0xcc, 0x27, 0xe3, 0x52, 0x94, 0x42, 0x47, 0x57, 0xbd,
-	0x1a, 0x9c, 0x41, 0x10, 0x8e, 0xd2, 0x1c, 0x46, 0x2e, 0xae, 0x80, 0xd4, 0xa2, 0xdc, 0xcc, 0xe2,
+	0x32, 0xd3, 0xe8, 0x01, 0xdc, 0x27, 0xc5, 0x25, 0xc6, 0xe0, 0xd6, 0x79, 0xc5, 0x09, 0x0a, 0x51,
-	0xe2, 0xcc, 0xfc, 0x3c, 0x21, 0x63, 0xa0, 0x01, 0x40, 0x5e, 0x48, 0x65, 0x01, 0xc4, 0x60, 0x3e,
+	0x7c, 0xca, 0xb4, 0xc6, 0x13, 0xf0, 0x9a, 0x5c, 0xa9, 0x77, 0x21, 0x0b, 0x62, 0xeb, 0x7c, 0xef,
-	0x23, 0x71, 0x3d, 0x88, 0x6b, 0xf4, 0x10, 0xaa, 0xf4, 0x40, 0xd2, 0x41, 0x70, 0x85, 0x42, 0x02,
+	0xf1, 0x18, 0x06, 0x52, 0xac, 0xb8, 0x22, 0x4e, 0xe8, 0xc4, 0x3e, 0x33, 0x26, 0xfa, 0x44, 0x00,
-	0x5c, 0xcc, 0xd9, 0xa9, 0x95, 0x50, 0x0b, 0x41, 0x4c, 0x21, 0x69, 0x2e, 0xce, 0xa2, 0xc4, 0xbc,
+	0x8f, 0x5c, 0x56, 0x0b, 0xa5, 0x16, 0xa2, 0xc6, 0x33, 0xf0, 0x1a, 0x2e, 0xab, 0xac, 0x6d, 0xcc,
-	0xf4, 0xd4, 0xf8, 0xd4, 0xbc, 0x14, 0xa0, 0x7d, 0x60, 0x87, 0x80, 0x05, 0x5c, 0xf3, 0x52, 0x94,
+	0xc5, 0x67, 0xd3, 0xf3, 0xc4, 0xd0, 0x24, 0x87, 0x56, 0xd2, 0x8f, 0xd9, 0xbe, 0x88, 0x03, 0x70,
-	0xb4, 0xb8, 0x58, 0xc0, 0xda, 0x38, 0xb8, 0x58, 0x82, 0x5c, 0x1d, 0x5d, 0x04, 0x18, 0x84, 0x38,
+	0x96, 0xbc, 0xfd, 0x7d, 0xb0, 0x97, 0xf8, 0x02, 0x7c, 0x99, 0xd7, 0x25, 0x7f, 0xe1, 0x75, 0x41,
-	0xb9, 0x58, 0xc3, 0x83, 0x3c, 0x43, 0x5c, 0x05, 0x18, 0x85, 0x78, 0xb9, 0x38, 0x41, 0x82, 0x10,
+	0x1c, 0x03, 0xa2, 0x83, 0xb4, 0x2e, 0xa2, 0x4b, 0x70, 0xf5, 0x31, 0x0f, 0x5c, 0x96, 0xde, 0xdc,
-	0x2e, 0x93, 0x52, 0x08, 0x50, 0x0d, 0xd0, 0x9d, 0x58, 0x3d, 0x6b, 0xc1, 0xc5, 0x0b, 0xb4, 0x0b,
+	0x05, 0x16, 0xf6, 0x61, 0xf0, 0xcc, 0xee, 0xb3, 0x34, 0x40, 0x78, 0x04, 0x7e, 0x1f, 0x1a, 0x6b,
-	0xe1, 0x2c, 0xa0, 0x03, 0x98, 0x35, 0xb8, 0x8d, 0x84, 0x30, 0x1d, 0x1c, 0x84, 0xaa, 0xd0, 0x49,
+	0x47, 0x19, 0xb8, 0x4c, 0xac, 0xf8, 0xd1, 0xcf, 0x5e, 0xc3, 0x68, 0xc9, 0xdb, 0x03, 0x16, 0xb1,
-	0xe4, 0xc4, 0x43, 0x39, 0x86, 0x0b, 0x40, 0x7c, 0xe2, 0x91, 0x1c, 0xe3, 0x05, 0x20, 0x7e, 0x00,
+	0x43, 0x27, 0x3e, 0x99, 0xe2, 0xff, 0xc0, 0xec, 0x6f, 0xf1, 0x96, 0x6c, 0x76, 0xd4, 0xda, 0xee,
-	0xc4, 0x49, 0x6c, 0xe0, 0xf0, 0x35, 0x06, 0x04, 0x00, 0x00, 0xff, 0xff, 0x9e, 0x31, 0x53, 0xfd,
+	0xa8, 0xb5, 0xe9, 0x28, 0xda, 0x76, 0x14, 0x7d, 0x75, 0x14, 0x7d, 0x7c, 0x53, 0x6b, 0x3e, 0xd4,
-	0x8b, 0x01, 0x00, 0x00,
+	0x3b, 0x9e, 0xfd, 0x04, 0x00, 0x00, 0xff, 0xff, 0xcc, 0x76, 0x8d, 0x4f, 0x8f, 0x01, 0x00, 0x00,
 }

auth/range_perm_cache.go

@@ -22,7 +22,10 @@ import (
 	"github.com/coreos/etcd/mvcc/backend"
 )
 
-// isSubset returns true if a is a subset of b
+// isSubset returns true if a is a subset of b.
+// If a is a prefix of b, then a is a subset of b.
+// Given intervals [a1,a2) and [b1,b2), is
+// the a interval a subset of b?
 func isSubset(a, b *rangePerm) bool {
 	switch {
 	case len(a.end) == 0 && len(b.end) == 0:
@@ -32,9 +35,11 @@ func isSubset(a, b *rangePerm) bool {
 		// b is a key, a is a range
 		return false
 	case len(a.end) == 0:
-		return 0 <= bytes.Compare(a.begin, b.begin) && bytes.Compare(a.begin, b.end) <= 0
+		// a is a key, b is a range. need b1 <= a1 and a1 < b2
+		return bytes.Compare(b.begin, a.begin) <= 0 && bytes.Compare(a.begin, b.end) < 0
 	default:
-		return 0 <= bytes.Compare(a.begin, b.begin) && bytes.Compare(a.end, b.end) <= 0
+		// both are ranges. need b1 <= a1 and a2 <= b2
+		return bytes.Compare(b.begin, a.begin) <= 0 && bytes.Compare(a.end, b.end) <= 0
 	}
 }
 
@@ -88,12 +93,18 @@ func mergeRangePerms(perms []*rangePerm) []*rangePerm {
 	i := 0
 	for i < len(perms) {
 		begin, next := i, i
-		for next+1 < len(perms) && bytes.Compare(perms[next].end, perms[next+1].begin) != -1 {
+		for next+1 < len(perms) && bytes.Compare(perms[next].end, perms[next+1].begin) >= 0 {
 			next++
 		}
-
+		// don't merge ["a", "b") with ["b", ""), because perms[next+1].end is empty.
-		merged = append(merged, &rangePerm{begin: perms[begin].begin, end: perms[next].end})
+		if next != begin && len(perms[next].end) > 0 {
-
+			merged = append(merged, &rangePerm{begin: perms[begin].begin, end: perms[next].end})
+		} else {
+			merged = append(merged, perms[begin])
+			if next != begin {
+				merged = append(merged, perms[next])
+			}
+		}
 		i = next + 1
 	}
 

auth/range_perm_cache_test.go

@@ -46,6 +46,10 @@ func TestGetMergedPerms(t *testing.T) {
 			[]*rangePerm{{[]byte("a"), []byte("b")}},
 			[]*rangePerm{{[]byte("a"), []byte("b")}},
 		},
+		{
+			[]*rangePerm{{[]byte("a"), []byte("b")}, {[]byte("b"), []byte("")}},
+			[]*rangePerm{{[]byte("a"), []byte("b")}, {[]byte("b"), []byte("")}},
+		},
 		{
 			[]*rangePerm{{[]byte("a"), []byte("b")}, {[]byte("b"), []byte("c")}},
 			[]*rangePerm{{[]byte("a"), []byte("c")}},
@@ -106,7 +110,7 @@ func TestGetMergedPerms(t *testing.T) {
 		},
 		{
 			[]*rangePerm{{[]byte("a"), []byte("")}, {[]byte("b"), []byte("c")}, {[]byte("b"), []byte("")}, {[]byte("c"), []byte("")}, {[]byte("d"), []byte("")}},
-			[]*rangePerm{{[]byte("a"), []byte("")}, {[]byte("b"), []byte("c")}, {[]byte("d"), []byte("")}},
+			[]*rangePerm{{[]byte("a"), []byte("")}, {[]byte("b"), []byte("c")}, {[]byte("c"), []byte("")}, {[]byte("d"), []byte("")}},
 		},
 		// duplicate ranges
 		{

auth/store.go

@@ -603,6 +603,11 @@ func (as *authStore) isOpPermitted(userName string, key, rangeEnd []byte, permTy
 		return false
 	}
 
+	// root role should have permission on all ranges
+	if hasRootRole(user) {
+		return true
+	}
+
 	if as.isRangeOpPermitted(tx, userName, key, rangeEnd, permTyp) {
 		return true
 	}

client/client.go

@@ -37,6 +37,10 @@ var (
 	ErrClusterUnavailable    = errors.New("client: etcd cluster is unavailable or misconfigured")
 	ErrNoLeaderEndpoint      = errors.New("client: no leader endpoint available")
 	errTooManyRedirectChecks = errors.New("client: too many redirect checks")
+
+	// oneShotCtxValue is set on a context using WithValue(&oneShotValue) so
+	// that Do() will not retry a request
+	oneShotCtxValue interface{}
 )
 
 var DefaultRequestTimeout = 5 * time.Second
@@ -335,6 +339,7 @@ func (c *httpClusterClient) Do(ctx context.Context, act httpAction) (*http.Respo
 	var body []byte
 	var err error
 	cerr := &ClusterError{}
+	isOneShot := ctx.Value(&oneShotCtxValue) != nil
 
 	for i := pinned; i < leps+pinned; i++ {
 		k := i % leps
@@ -348,6 +353,9 @@ func (c *httpClusterClient) Do(ctx context.Context, act httpAction) (*http.Respo
 			if err == context.Canceled || err == context.DeadlineExceeded {
 				return nil, nil, err
 			}
+			if isOneShot {
+				return nil, nil, err
+			}
 			continue
 		}
 		if resp.StatusCode/100 == 5 {
@@ -358,6 +366,9 @@ func (c *httpClusterClient) Do(ctx context.Context, act httpAction) (*http.Respo
 			default:
 				cerr.Errors = append(cerr.Errors, fmt.Errorf("client: etcd member %s returns server error [%s]", eps[k].String(), http.StatusText(resp.StatusCode)))
 			}
+			if isOneShot {
+				return nil, nil, cerr.Errors[0]
+			}
 			continue
 		}
 		if k != pinned {

client/integration/client_test.go

@@ -0,0 +1,134 @@
+// Copyright 2016 The etcd Authors
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package integration
+
+import (
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	"golang.org/x/net/context"
+
+	"github.com/coreos/etcd/client"
+	"github.com/coreos/etcd/integration"
+	"github.com/coreos/etcd/pkg/testutil"
+)
+
+// TestV2NoRetryEOF tests destructive api calls won't retry on a disconnection.
+func TestV2NoRetryEOF(t *testing.T) {
+	defer testutil.AfterTest(t)
+	// generate an EOF response; specify address so appears first in sorted ep list
+	lEOF := integration.NewListenerWithAddr(t, fmt.Sprintf("eof:123.%d.sock", os.Getpid()))
+	defer lEOF.Close()
+	tries := uint32(0)
+	go func() {
+		for {
+			conn, err := lEOF.Accept()
+			if err != nil {
+				return
+			}
+			atomic.AddUint32(&tries, 1)
+			conn.Close()
+		}
+	}()
+	eofURL := integration.UrlScheme + "://" + lEOF.Addr().String()
+	cli := integration.MustNewHTTPClient(t, []string{eofURL, eofURL}, nil)
+	kapi := client.NewKeysAPI(cli)
+	for i, f := range noRetryList(kapi) {
+		startTries := atomic.LoadUint32(&tries)
+		if err := f(); err == nil {
+			t.Errorf("#%d: expected EOF error, got nil", i)
+		}
+		endTries := atomic.LoadUint32(&tries)
+		if startTries+1 != endTries {
+			t.Errorf("#%d: expected 1 try, got %d", i, endTries-startTries)
+		}
+	}
+}
+
+// TestV2NoRetryNoLeader tests destructive api calls won't retry if given an error code.
+func TestV2NoRetryNoLeader(t *testing.T) {
+	defer testutil.AfterTest(t)
+
+	lHttp := integration.NewListenerWithAddr(t, fmt.Sprintf("errHttp:123.%d.sock", os.Getpid()))
+	eh := &errHandler{errCode: http.StatusServiceUnavailable}
+	srv := httptest.NewUnstartedServer(eh)
+	defer lHttp.Close()
+	defer srv.Close()
+	srv.Listener = lHttp
+	go srv.Start()
+	lHttpURL := integration.UrlScheme + "://" + lHttp.Addr().String()
+
+	cli := integration.MustNewHTTPClient(t, []string{lHttpURL, lHttpURL}, nil)
+	kapi := client.NewKeysAPI(cli)
+	// test error code
+	for i, f := range noRetryList(kapi) {
+		reqs := eh.reqs
+		if err := f(); err == nil || !strings.Contains(err.Error(), "no leader") {
+			t.Errorf("#%d: expected \"no leader\", got %v", i, err)
+		}
+		if eh.reqs != reqs+1 {
+			t.Errorf("#%d: expected 1 request, got %d", i, eh.reqs-reqs)
+		}
+	}
+}
+
+// TestV2RetryRefuse tests destructive api calls will retry if a connection is refused.
+func TestV2RetryRefuse(t *testing.T) {
+	defer testutil.AfterTest(t)
+	cl := integration.NewCluster(t, 1)
+	cl.Launch(t)
+	defer cl.Terminate(t)
+	// test connection refused; expect no error failover
+	cli := integration.MustNewHTTPClient(t, []string{integration.UrlScheme + "://refuseconn:123", cl.URL(0)}, nil)
+	kapi := client.NewKeysAPI(cli)
+	if _, err := kapi.Set(context.Background(), "/delkey", "def", nil); err != nil {
+		t.Fatal(err)
+	}
+	for i, f := range noRetryList(kapi) {
+		if err := f(); err != nil {
+			t.Errorf("#%d: unexpected retry failure (%v)", i, err)
+		}
+	}
+}
+
+type errHandler struct {
+	errCode int
+	reqs    int
+}
+
+func (eh *errHandler) ServeHTTP(w http.ResponseWriter, req *http.Request) {
+	req.Body.Close()
+	eh.reqs++
+	w.WriteHeader(eh.errCode)
+}
+
+func noRetryList(kapi client.KeysAPI) []func() error {
+	return []func() error{
+		func() error {
+			opts := &client.SetOptions{PrevExist: client.PrevNoExist}
+			_, err := kapi.Set(context.Background(), "/setkey", "bar", opts)
+			return err
+		},
+		func() error {
+			_, err := kapi.Delete(context.Background(), "/delkey", nil)
+			return err
+		},
+	}
+}

client/integration/main_test.go

@@ -0,0 +1,20 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package integration
+
+import (
+	"os"
+	"testing"
+
+	"github.com/coreos/etcd/pkg/testutil"
+)
+
+func TestMain(m *testing.M) {
+	v := m.Run()
+	if v == 0 && testutil.CheckLeakedGoroutine() {
+		os.Exit(1)
+	}
+	os.Exit(v)
+}

client/keys.go

@@ -337,7 +337,11 @@ func (k *httpKeysAPI) Set(ctx context.Context, key, val string, opts *SetOptions
 		act.Dir = opts.Dir
 	}
 
-	resp, body, err := k.client.Do(ctx, act)
+	doCtx := ctx
+	if act.PrevExist == PrevNoExist {
+		doCtx = context.WithValue(doCtx, &oneShotCtxValue, &oneShotCtxValue)
+	}
+	resp, body, err := k.client.Do(doCtx, act)
 	if err != nil {
 		return nil, err
 	}
@@ -385,7 +389,8 @@ func (k *httpKeysAPI) Delete(ctx context.Context, key string, opts *DeleteOption
 		act.Recursive = opts.Recursive
 	}
 
-	resp, body, err := k.client.Do(ctx, act)
+	doCtx := context.WithValue(ctx, &oneShotCtxValue, &oneShotCtxValue)
+	resp, body, err := k.client.Do(doCtx, act)
 	if err != nil {
 		return nil, err
 	}

clientv3/auth.go

@@ -145,12 +145,12 @@ func (auth *auth) UserGrantRole(ctx context.Context, user string, role string) (
 }
 
 func (auth *auth) UserGet(ctx context.Context, name string) (*AuthUserGetResponse, error) {
-	resp, err := auth.remote.UserGet(ctx, &pb.AuthUserGetRequest{Name: name})
+	resp, err := auth.remote.UserGet(ctx, &pb.AuthUserGetRequest{Name: name}, grpc.FailFast(false))
 	return (*AuthUserGetResponse)(resp), toErr(ctx, err)
 }
 
 func (auth *auth) UserList(ctx context.Context) (*AuthUserListResponse, error) {
-	resp, err := auth.remote.UserList(ctx, &pb.AuthUserListRequest{})
+	resp, err := auth.remote.UserList(ctx, &pb.AuthUserListRequest{}, grpc.FailFast(false))
 	return (*AuthUserListResponse)(resp), toErr(ctx, err)
 }
 
@@ -175,12 +175,12 @@ func (auth *auth) RoleGrantPermission(ctx context.Context, name string, key, ran
 }
 
 func (auth *auth) RoleGet(ctx context.Context, role string) (*AuthRoleGetResponse, error) {
-	resp, err := auth.remote.RoleGet(ctx, &pb.AuthRoleGetRequest{Role: role})
+	resp, err := auth.remote.RoleGet(ctx, &pb.AuthRoleGetRequest{Role: role}, grpc.FailFast(false))
 	return (*AuthRoleGetResponse)(resp), toErr(ctx, err)
 }
 
 func (auth *auth) RoleList(ctx context.Context) (*AuthRoleListResponse, error) {
-	resp, err := auth.remote.RoleList(ctx, &pb.AuthRoleListRequest{})
+	resp, err := auth.remote.RoleList(ctx, &pb.AuthRoleListRequest{}, grpc.FailFast(false))
 	return (*AuthRoleListResponse)(resp), toErr(ctx, err)
 }
 
@@ -208,7 +208,7 @@ type authenticator struct {
 }
 
 func (auth *authenticator) authenticate(ctx context.Context, name string, password string) (*AuthenticateResponse, error) {
-	resp, err := auth.remote.Authenticate(ctx, &pb.AuthenticateRequest{Name: name, Password: password})
+	resp, err := auth.remote.Authenticate(ctx, &pb.AuthenticateRequest{Name: name, Password: password}, grpc.FailFast(false))
 	return (*AuthenticateResponse)(resp), toErr(ctx, err)
 }
 

clientv3/balancer.go

@@ -17,7 +17,7 @@ package clientv3
 import (
 	"net/url"
 	"strings"
-	"sync/atomic"
+	"sync"
 
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
@@ -26,32 +26,133 @@ import (
 // simpleBalancer does the bare minimum to expose multiple eps
 // to the grpc reconnection code path
 type simpleBalancer struct {
-	// eps are the client's endpoints stripped of any URL scheme
+	// addrs are the client's endpoints for grpc
-	eps     []string
+	addrs []grpc.Address
-	ch      chan []grpc.Address
+	// notifyCh notifies grpc of the set of addresses for connecting
-	numGets uint32
+	notifyCh chan []grpc.Address
+
+	// readyc closes once the first connection is up
+	readyc    chan struct{}
+	readyOnce sync.Once
+
+	// mu protects upEps, pinAddr, and connectingAddr
+	mu sync.RWMutex
+	// upEps holds the current endpoints that have an active connection
+	upEps map[string]struct{}
+	// upc closes when upEps transitions from empty to non-zero or the balancer closes.
+	upc chan struct{}
+
+	// pinAddr is the currently pinned address; set to the empty string on
+	// intialization and shutdown.
+	pinAddr string
+
+	closed bool
 }
 
-func newSimpleBalancer(eps []string) grpc.Balancer {
+func newSimpleBalancer(eps []string) *simpleBalancer {
-	ch := make(chan []grpc.Address, 1)
+	notifyCh := make(chan []grpc.Address, 1)
 	addrs := make([]grpc.Address, len(eps))
 	for i := range eps {
 		addrs[i].Addr = getHost(eps[i])
 	}
-	ch <- addrs
+	notifyCh <- addrs
-	return &simpleBalancer{eps: eps, ch: ch}
+	sb := &simpleBalancer{
+		addrs:    addrs,
+		notifyCh: notifyCh,
+		readyc:   make(chan struct{}),
+		upEps:    make(map[string]struct{}),
+		upc:      make(chan struct{}),
+	}
+	return sb
 }
 
-func (b *simpleBalancer) Start(target string) error        { return nil }
+func (b *simpleBalancer) Start(target string) error { return nil }
-func (b *simpleBalancer) Up(addr grpc.Address) func(error) { return func(error) {} }
+
-func (b *simpleBalancer) Get(ctx context.Context, opts grpc.BalancerGetOptions) (grpc.Address, func(), error) {
+func (b *simpleBalancer) ConnectNotify() <-chan struct{} {
-	v := atomic.AddUint32(&b.numGets, 1)
+	b.mu.Lock()
-	ep := b.eps[v%uint32(len(b.eps))]
+	defer b.mu.Unlock()
-	return grpc.Address{Addr: getHost(ep)}, func() {}, nil
+	return b.upc
 }
-func (b *simpleBalancer) Notify() <-chan []grpc.Address { return b.ch }
+
+func (b *simpleBalancer) Up(addr grpc.Address) func(error) {
+	b.mu.Lock()
+	defer b.mu.Unlock()
+
+	// gRPC might call Up after it called Close. We add this check
+	// to "fix" it up at application layer. Or our simplerBalancer
+	// might panic since b.upc is closed.
+	if b.closed {
+		return func(err error) {}
+	}
+
+	if len(b.upEps) == 0 {
+		// notify waiting Get()s and pin first connected address
+		close(b.upc)
+		b.pinAddr = addr.Addr
+	}
+	b.upEps[addr.Addr] = struct{}{}
+
+	// notify client that a connection is up
+	b.readyOnce.Do(func() { close(b.readyc) })
+
+	return func(err error) {
+		b.mu.Lock()
+		delete(b.upEps, addr.Addr)
+		if len(b.upEps) == 0 && b.pinAddr != "" {
+			b.upc = make(chan struct{})
+		} else if b.pinAddr == addr.Addr {
+			// choose new random up endpoint
+			for k := range b.upEps {
+				b.pinAddr = k
+				break
+			}
+		}
+		b.mu.Unlock()
+	}
+}
+
+func (b *simpleBalancer) Get(ctx context.Context, opts grpc.BalancerGetOptions) (grpc.Address, func(), error) {
+	var addr string
+	for {
+		b.mu.RLock()
+		ch := b.upc
+		b.mu.RUnlock()
+		select {
+		case <-ch:
+		case <-ctx.Done():
+			return grpc.Address{Addr: ""}, nil, ctx.Err()
+		}
+		b.mu.RLock()
+		addr = b.pinAddr
+		upEps := len(b.upEps)
+		b.mu.RUnlock()
+		if addr == "" {
+			return grpc.Address{Addr: ""}, nil, grpc.ErrClientConnClosing
+		}
+		if upEps > 0 {
+			break
+		}
+	}
+	return grpc.Address{Addr: addr}, func() {}, nil
+}
+
+func (b *simpleBalancer) Notify() <-chan []grpc.Address { return b.notifyCh }
+
 func (b *simpleBalancer) Close() error {
-	close(b.ch)
+	b.mu.Lock()
+	defer b.mu.Unlock()
+	// In case gRPC calls close twice. TODO: remove the checking
+	// when we are sure that gRPC wont call close twice.
+	if b.closed {
+		return nil
+	}
+	b.closed = true
+	close(b.notifyCh)
+	// terminate all waiting Get()s
+	b.pinAddr = ""
+	if len(b.upEps) == 0 {
+		close(b.upc)
+	}
 	return nil
 }
 

clientv3/client.go

@@ -46,9 +46,11 @@ type Client struct {
 	Auth
 	Maintenance
 
-	conn  *grpc.ClientConn
+	conn         *grpc.ClientConn
-	cfg   Config
+	cfg          Config
-	creds *credentials.TransportCredentials
+	creds        *credentials.TransportCredentials
+	balancer     *simpleBalancer
+	retryWrapper retryRpcFunc
 
 	ctx    context.Context
 	cancel context.CancelFunc
@@ -138,11 +140,10 @@ func (c *Client) dialTarget(endpoint string) (proto string, host string, creds *
 	return
 }
 
-// dialSetupOpts gives the dial opts prioer to any authentication
+// dialSetupOpts gives the dial opts prior to any authentication
-func (c *Client) dialSetupOpts(endpoint string, dopts ...grpc.DialOption) []grpc.DialOption {
+func (c *Client) dialSetupOpts(endpoint string, dopts ...grpc.DialOption) (opts []grpc.DialOption) {
-	opts := []grpc.DialOption{
+	if c.cfg.DialTimeout > 0 {
-		grpc.WithBlock(),
+		opts = []grpc.DialOption{grpc.WithTimeout(c.cfg.DialTimeout)}
-		grpc.WithTimeout(c.cfg.DialTimeout),
 	}
 	opts = append(opts, dopts...)
 
@@ -240,12 +241,30 @@ func newClient(cfg *Config) (*Client, error) {
 		client.Password = cfg.Password
 	}
 
-	b := newSimpleBalancer(cfg.Endpoints)
+	client.balancer = newSimpleBalancer(cfg.Endpoints)
-	conn, err := client.dial(cfg.Endpoints[0], grpc.WithBalancer(b))
+	conn, err := client.dial(cfg.Endpoints[0], grpc.WithBalancer(client.balancer))
 	if err != nil {
 		return nil, err
 	}
 	client.conn = conn
+	client.retryWrapper = client.newRetryWrapper()
+
+	// wait for a connection
+	if cfg.DialTimeout > 0 {
+		hasConn := false
+		waitc := time.After(cfg.DialTimeout)
+		select {
+		case <-client.balancer.readyc:
+			hasConn = true
+		case <-ctx.Done():
+		case <-waitc:
+		}
+		if !hasConn {
+			client.cancel()
+			conn.Close()
+			return nil, grpc.ErrClientConnTimeout
+		}
+	}
 
 	client.Cluster = NewCluster(client)
 	client.KV = NewKV(client)
@@ -275,8 +294,17 @@ func isHaltErr(ctx context.Context, err error) bool {
 	if err == nil {
 		return false
 	}
-	return strings.HasPrefix(grpc.ErrorDesc(err), "etcdserver: ") ||
+	eErr := rpctypes.Error(err)
-		strings.Contains(err.Error(), grpc.ErrClientConnClosing.Error())
+	if _, ok := eErr.(rpctypes.EtcdError); ok {
+		return eErr != rpctypes.ErrStopped && eErr != rpctypes.ErrNoLeader
+	}
+	// treat etcdserver errors not recognized by the client as halting
+	return isConnClosing(err) || strings.Contains(err.Error(), "etcdserver:")
+}
+
+// isConnClosing returns true if the error matches a grpc client closing error
+func isConnClosing(err error) bool {
+	return strings.Contains(err.Error(), grpc.ErrClientConnClosing.Error())
 }
 
 func toErr(ctx context.Context, err error) error {
@@ -284,9 +312,12 @@ func toErr(ctx context.Context, err error) error {
 		return nil
 	}
 	err = rpctypes.Error(err)
-	if ctx.Err() != nil && strings.Contains(err.Error(), "context") {
+	switch {
+	case ctx.Err() != nil && strings.Contains(err.Error(), "context"):
 		err = ctx.Err()
-	} else if strings.Contains(err.Error(), grpc.ErrClientConnClosing.Error()) {
+	case strings.Contains(err.Error(), ErrNoAvailableEndpoints.Error()):
+		err = ErrNoAvailableEndpoints
+	case strings.Contains(err.Error(), grpc.ErrClientConnClosing.Error()):
 		err = grpc.ErrClientConnClosing
 	}
 	return err

clientv3/client_test.go

@@ -19,11 +19,15 @@ import (
 	"testing"
 	"time"
 
+	"github.com/coreos/etcd/etcdserver"
+	"github.com/coreos/etcd/pkg/testutil"
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
 )
 
 func TestDialTimeout(t *testing.T) {
+	defer testutil.AfterTest(t)
+
 	donec := make(chan error)
 	go func() {
 		// without timeout, grpc keeps redialing if connection refused
@@ -55,9 +59,24 @@ func TestDialTimeout(t *testing.T) {
 	}
 }
 
+func TestDialNoTimeout(t *testing.T) {
+	cfg := Config{Endpoints: []string{"127.0.0.1:12345"}}
+	c, err := New(cfg)
+	if c == nil || err != nil {
+		t.Fatalf("new client with DialNoWait should succeed, got %v", err)
+	}
+	c.Close()
+}
+
 func TestIsHaltErr(t *testing.T) {
 	if !isHaltErr(nil, fmt.Errorf("etcdserver: some etcdserver error")) {
-		t.Errorf(`error prefixed with "etcdserver: " should be Halted`)
+		t.Errorf(`error prefixed with "etcdserver: " should be Halted by default`)
+	}
+	if isHaltErr(nil, etcdserver.ErrStopped) {
+		t.Errorf("error %v should not halt", etcdserver.ErrStopped)
+	}
+	if isHaltErr(nil, etcdserver.ErrNoLeader) {
+		t.Errorf("error %v should not halt", etcdserver.ErrNoLeader)
 	}
 	ctx, cancel := context.WithCancel(context.TODO())
 	if isHaltErr(ctx, nil) {

clientv3/cluster.go

@@ -17,6 +17,7 @@ package clientv3
 import (
 	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
 	"golang.org/x/net/context"
+	"google.golang.org/grpc"
 )
 
 type (
@@ -46,7 +47,7 @@ type cluster struct {
 }
 
 func NewCluster(c *Client) Cluster {
-	return &cluster{remote: pb.NewClusterClient(c.conn)}
+	return &cluster{remote: RetryClusterClient(c)}
 }
 
 func (c *cluster) MemberAdd(ctx context.Context, peerAddrs []string) (*MemberAddResponse, error) {
@@ -90,7 +91,7 @@ func (c *cluster) MemberUpdate(ctx context.Context, id uint64, peerAddrs []strin
 func (c *cluster) MemberList(ctx context.Context) (*MemberListResponse, error) {
 	// it is safe to retry on list.
 	for {
-		resp, err := c.remote.MemberList(ctx, &pb.MemberListRequest{})
+		resp, err := c.remote.MemberList(ctx, &pb.MemberListRequest{}, grpc.FailFast(false))
 		if err == nil {
 			return (*MemberListResponse)(resp), nil
 		}

clientv3/concurrency/election.go

@@ -40,7 +40,7 @@ type Election struct {
 
 // NewElection returns a new election on a given key prefix.
 func NewElection(client *v3.Client, pfx string) *Election {
-	return &Election{client: client, keyPrefix: pfx}
+	return &Election{client: client, keyPrefix: pfx + "/"}
 }
 
 // Campaign puts a value as eligible for the election. It blocks until
@@ -59,7 +59,6 @@ func (e *Election) Campaign(ctx context.Context, val string) error {
 	if err != nil {
 		return err
 	}
-
 	e.leaderKey, e.leaderRev, e.leaderSession = k, resp.Header.Revision, s
 	if !resp.Succeeded {
 		kv := resp.Responses[0].GetResponseRange().Kvs[0]

clientv3/concurrency/mutex.go

@@ -32,7 +32,7 @@ type Mutex struct {
 }
 
 func NewMutex(client *v3.Client, pfx string) *Mutex {
-	return &Mutex{client, pfx, "", -1}
+	return &Mutex{client, pfx + "/", "", -1}
 }
 
 // Lock locks the mutex with a cancellable context. If the context is cancelled
@@ -43,7 +43,7 @@ func (m *Mutex) Lock(ctx context.Context) error {
 		return serr
 	}
 
-	m.myKey = fmt.Sprintf("%s/%x", m.pfx, s.Lease())
+	m.myKey = fmt.Sprintf("%s%x", m.pfx, s.Lease())
 	cmp := v3.Compare(v3.CreateRevision(m.myKey), "=", 0)
 	// put self in lock waiters via myKey; oldest waiter holds lock
 	put := v3.OpPut(m.myKey, "", v3.WithLease(s.Lease()))

clientv3/example_auth_test.go

@@ -32,35 +32,63 @@ func ExampleAuth() {
 	}
 	defer cli.Close()
 
-	authapi := clientv3.NewAuth(cli)
+	if _, err = cli.RoleAdd(context.TODO(), "root"); err != nil {
-
+		log.Fatal(err)
-	if _, err = authapi.RoleAdd(context.TODO(), "root"); err != nil {
+	}
+	if _, err = cli.UserAdd(context.TODO(), "root", "123"); err != nil {
+		log.Fatal(err)
+	}
+	if _, err = cli.UserGrantRole(context.TODO(), "root", "root"); err != nil {
 		log.Fatal(err)
 	}
 
-	if _, err = authapi.RoleGrantPermission(
+	if _, err = cli.RoleAdd(context.TODO(), "r"); err != nil {
+		log.Fatal(err)
+	}
+
+	if _, err = cli.RoleGrantPermission(
 		context.TODO(),
-		"root", // role name
+		"r",   // role name
-		"foo",  // key
+		"foo", // key
-		"zoo",  // range end
+		"zoo", // range end
 		clientv3.PermissionType(clientv3.PermReadWrite),
 	); err != nil {
 		log.Fatal(err)
 	}
-
+	if _, err = cli.UserAdd(context.TODO(), "u", "123"); err != nil {
-	if _, err = authapi.UserAdd(context.TODO(), "root", "123"); err != nil {
 		log.Fatal(err)
 	}
-
+	if _, err = cli.UserGrantRole(context.TODO(), "u", "r"); err != nil {
-	if _, err = authapi.UserGrantRole(context.TODO(), "root", "root"); err != nil {
 		log.Fatal(err)
 	}
-
+	if _, err = cli.AuthEnable(context.TODO()); err != nil {
-	if _, err = authapi.AuthEnable(context.TODO()); err != nil {
 		log.Fatal(err)
 	}
 
 	cliAuth, err := clientv3.New(clientv3.Config{
+		Endpoints:   endpoints,
+		DialTimeout: dialTimeout,
+		Username:    "u",
+		Password:    "123",
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer cliAuth.Close()
+
+	if _, err = cliAuth.Put(context.TODO(), "foo1", "bar"); err != nil {
+		log.Fatal(err)
+	}
+
+	_, err = cliAuth.Txn(context.TODO()).
+		If(clientv3.Compare(clientv3.Value("zoo1"), ">", "abc")).
+		Then(clientv3.OpPut("zoo1", "XYZ")).
+		Else(clientv3.OpPut("zoo1", "ABC")).
+		Commit()
+	fmt.Println(err)
+
+	// now check the permission with the root account
+	rootCli, err := clientv3.New(clientv3.Config{
 		Endpoints:   endpoints,
 		DialTimeout: dialTimeout,
 		Username:    "root",
@@ -69,31 +97,17 @@ func ExampleAuth() {
 	if err != nil {
 		log.Fatal(err)
 	}
-	defer cliAuth.Close()
+	defer rootCli.Close()
 
-	kv := clientv3.NewKV(cliAuth)
+	resp, err := rootCli.RoleGet(context.TODO(), "r")
-	if _, err = kv.Put(context.TODO(), "foo1", "bar"); err != nil {
-		log.Fatal(err)
-	}
-
-	_, err = kv.Txn(context.TODO()).
-		If(clientv3.Compare(clientv3.Value("zoo1"), ">", "abc")).
-		Then(clientv3.OpPut("zoo1", "XYZ")).
-		Else(clientv3.OpPut("zoo1", "ABC")).
-		Commit()
-	fmt.Println(err)
-
-	// now check the permission
-	authapi2 := clientv3.NewAuth(cliAuth)
-	resp, err := authapi2.RoleGet(context.TODO(), "root")
 	if err != nil {
 		log.Fatal(err)
 	}
-	fmt.Printf("root user permission: key %q, range end %q\n", resp.Perm[0].Key, resp.Perm[0].RangeEnd)
+	fmt.Printf("user u permission: key %q, range end %q\n", resp.Perm[0].Key, resp.Perm[0].RangeEnd)
 
-	if _, err = authapi2.AuthDisable(context.TODO()); err != nil {
+	if _, err = rootCli.AuthDisable(context.TODO()); err != nil {
 		log.Fatal(err)
 	}
 	// Output: etcdserver: permission denied
-	// root user permission: key "foo", range end "zoo"
+	// user u permission: key "foo", range end "zoo"
 }

clientv3/example_kv_test.go

@@ -210,7 +210,7 @@ func ExampleKV_compact() {
 	compRev := resp.Header.Revision // specify compact revision of your choice
 
 	ctx, cancel = context.WithTimeout(context.Background(), requestTimeout)
-	err = cli.Compact(ctx, compRev)
+	_, err = cli.Compact(ctx, compRev)
 	cancel()
 	if err != nil {
 		log.Fatal(err)

clientv3/integration/kv_test.go

@@ -16,6 +16,7 @@ package integration
 
 import (
 	"bytes"
+	"math/rand"
 	"reflect"
 	"strings"
 	"testing"
@@ -470,17 +471,17 @@ func TestKVCompactError(t *testing.T) {
 			t.Fatalf("couldn't put 'foo' (%v)", err)
 		}
 	}
-	err := kv.Compact(ctx, 6)
+	_, err := kv.Compact(ctx, 6)
 	if err != nil {
 		t.Fatalf("couldn't compact 6 (%v)", err)
 	}
 
-	err = kv.Compact(ctx, 6)
+	_, err = kv.Compact(ctx, 6)
 	if err != rpctypes.ErrCompacted {
 		t.Fatalf("expected %v, got %v", rpctypes.ErrCompacted, err)
 	}
 
-	err = kv.Compact(ctx, 100)
+	_, err = kv.Compact(ctx, 100)
 	if err != rpctypes.ErrFutureRev {
 		t.Fatalf("expected %v, got %v", rpctypes.ErrFutureRev, err)
 	}
@@ -501,11 +502,11 @@ func TestKVCompact(t *testing.T) {
 		}
 	}
 
-	err := kv.Compact(ctx, 7)
+	_, err := kv.Compact(ctx, 7)
 	if err != nil {
 		t.Fatalf("couldn't compact kv space (%v)", err)
 	}
-	err = kv.Compact(ctx, 7)
+	_, err = kv.Compact(ctx, 7)
 	if err == nil || err != rpctypes.ErrCompacted {
 		t.Fatalf("error got %v, want %v", err, rpctypes.ErrCompacted)
 	}
@@ -525,7 +526,7 @@ func TestKVCompact(t *testing.T) {
 		t.Fatalf("wchan got %v, expected closed", wr)
 	}
 
-	err = kv.Compact(ctx, 1000)
+	_, err = kv.Compact(ctx, 1000)
 	if err == nil || err != rpctypes.ErrFutureRev {
 		t.Fatalf("error got %v, want %v", err, rpctypes.ErrFutureRev)
 	}
@@ -662,3 +663,75 @@ func TestKVPutStoppedServerAndClose(t *testing.T) {
 		t.Fatal(err)
 	}
 }
+
+// TestKVGetOneEndpointDown ensures a client can connect and get if one endpoint is down
+func TestKVPutOneEndpointDown(t *testing.T) {
+	defer testutil.AfterTest(t)
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 3})
+	defer clus.Terminate(t)
+
+	// get endpoint list
+	eps := make([]string, 3)
+	for i := range eps {
+		eps[i] = clus.Members[i].GRPCAddr()
+	}
+
+	// make a dead node
+	clus.Members[rand.Intn(len(eps))].Stop(t)
+
+	// try to connect with dead node in the endpoint list
+	cfg := clientv3.Config{Endpoints: eps, DialTimeout: 1 * time.Second}
+	cli, err := clientv3.New(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer cli.Close()
+	ctx, cancel := context.WithTimeout(context.TODO(), 3*time.Second)
+	if _, err := cli.Get(ctx, "abc", clientv3.WithSerializable()); err != nil {
+		t.Fatal(err)
+	}
+	cancel()
+}
+
+// TestKVGetResetLoneEndpoint ensures that if an endpoint resets and all other
+// endpoints are down, then it will reconnect.
+func TestKVGetResetLoneEndpoint(t *testing.T) {
+	defer testutil.AfterTest(t)
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 2})
+	defer clus.Terminate(t)
+
+	// get endpoint list
+	eps := make([]string, 2)
+	for i := range eps {
+		eps[i] = clus.Members[i].GRPCAddr()
+	}
+
+	cfg := clientv3.Config{Endpoints: eps, DialTimeout: 500 * time.Millisecond}
+	cli, err := clientv3.New(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer cli.Close()
+
+	// disconnect everything
+	clus.Members[0].Stop(t)
+	clus.Members[1].Stop(t)
+
+	// have Get try to reconnect
+	donec := make(chan struct{})
+	go func() {
+		ctx, cancel := context.WithTimeout(context.TODO(), 5*time.Second)
+		if _, err := cli.Get(ctx, "abc", clientv3.WithSerializable()); err != nil {
+			t.Fatal(err)
+		}
+		cancel()
+		close(donec)
+	}()
+	time.Sleep(500 * time.Millisecond)
+	clus.Members[0].Restart(t)
+	select {
+	case <-time.After(10 * time.Second):
+		t.Fatalf("timed out waiting for Get")
+	case <-donec:
+	}
+}

clientv3/integration/lease_test.go

@@ -359,7 +359,8 @@ func TestLeaseKeepAliveCloseAfterDisconnectRevoke(t *testing.T) {
 	if kerr != nil {
 		t.Fatal(kerr)
 	}
-	if kresp := <-rc; kresp.ID != resp.ID {
+	kresp := <-rc
+	if kresp.ID != resp.ID {
 		t.Fatalf("ID = %x, want %x", kresp.ID, resp.ID)
 	}
 
@@ -374,13 +375,14 @@ func TestLeaseKeepAliveCloseAfterDisconnectRevoke(t *testing.T) {
 
 	clus.Members[0].Restart(t)
 
-	select {
+	// some keep-alives may still be buffered; drain until close
-	case ka, ok := <-rc:
+	timer := time.After(time.Duration(kresp.TTL) * time.Second)
-		if ok {
+	for kresp != nil {
-			t.Fatalf("unexpected keepalive %v", ka)
+		select {
+		case kresp = <-rc:
+		case <-timer:
+			t.Fatalf("keepalive channel did not close")
 		}
-	case <-time.After(5 * time.Second):
-		t.Fatalf("keepalive channel did not close")
 	}
 }
 
@@ -453,3 +455,46 @@ func TestLeaseKeepAliveTTLTimeout(t *testing.T) {
 
 	clus.Members[0].Restart(t)
 }
+
+// TestLeaseRenewLostQuorum ensures keepalives work after losing quorum
+// for a while.
+func TestLeaseRenewLostQuorum(t *testing.T) {
+	defer testutil.AfterTest(t)
+
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 3})
+	defer clus.Terminate(t)
+
+	cli := clus.Client(0)
+	r, err := cli.Grant(context.TODO(), 4)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	kctx, kcancel := context.WithCancel(context.Background())
+	defer kcancel()
+	ka, err := cli.KeepAlive(kctx, r.ID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// consume first keepalive so next message sends when cluster is down
+	<-ka
+
+	// force keepalive stream message to timeout
+	clus.Members[1].Stop(t)
+	clus.Members[2].Stop(t)
+	// Use TTL-1 since the client closes the keepalive channel if no
+	// keepalive arrives before the lease deadline.
+	// The cluster has 1 second to recover and reply to the keepalive.
+	time.Sleep(time.Duration(r.TTL-1) * time.Second)
+	clus.Members[1].Restart(t)
+	clus.Members[2].Restart(t)
+
+	select {
+	case _, ok := <-ka:
+		if !ok {
+			t.Fatalf("keepalive closed")
+		}
+	case <-time.After(time.Duration(r.TTL) * time.Second):
+		t.Fatalf("timed out waiting for keepalive")
+	}
+}

clientv3/integration/mirror_test.go

@@ -15,7 +15,9 @@
 package integration
 
 import (
+	"fmt"
 	"reflect"
+	"sync"
 	"testing"
 	"time"
 
@@ -69,3 +71,55 @@ func TestMirrorSync(t *testing.T) {
 		t.Fatal("failed to receive update in one second")
 	}
 }
+
+func TestMirrorSyncBase(t *testing.T) {
+	cluster := integration.NewClusterV3(nil, &integration.ClusterConfig{Size: 1})
+	defer cluster.Terminate(nil)
+
+	cli := cluster.Client(0)
+	ctx := context.TODO()
+
+	keyCh := make(chan string)
+	var wg sync.WaitGroup
+
+	for i := 0; i < 50; i++ {
+		wg.Add(1)
+
+		go func() {
+			defer wg.Done()
+
+			for key := range keyCh {
+				if _, err := cli.Put(ctx, key, "test"); err != nil {
+					t.Fatal(err)
+				}
+			}
+		}()
+	}
+
+	for i := 0; i < 2000; i++ {
+		keyCh <- fmt.Sprintf("test%d", i)
+	}
+
+	close(keyCh)
+	wg.Wait()
+
+	syncer := mirror.NewSyncer(cli, "test", 0)
+	respCh, errCh := syncer.SyncBase(ctx)
+
+	count := 0
+
+	for resp := range respCh {
+		count = count + len(resp.Kvs)
+		if !resp.More {
+			break
+		}
+	}
+
+	for err := range errCh {
+		t.Fatalf("unexpected error %v", err)
+	}
+
+	if count != 2000 {
+		t.Errorf("unexpected kv count: %d", count)
+	}
+}

clientv3/integration/watch_test.go

@@ -375,7 +375,7 @@ func TestWatchResumeCompacted(t *testing.T) {
 			t.Fatal(err)
 		}
 	}
-	if err := kv.Compact(context.TODO(), 3); err != nil {
+	if _, err := kv.Compact(context.TODO(), 3); err != nil {
 		t.Fatal(err)
 	}
 
@@ -400,7 +400,7 @@ func TestWatchResumeCompacted(t *testing.T) {
 func TestWatchCompactRevision(t *testing.T) {
 	defer testutil.AfterTest(t)
 
-	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 3})
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 1})
 	defer clus.Terminate(t)
 
 	// set some keys
@@ -414,7 +414,7 @@ func TestWatchCompactRevision(t *testing.T) {
 	w := clientv3.NewWatcher(clus.RandClient())
 	defer w.Close()
 
-	if err := kv.Compact(context.TODO(), 4); err != nil {
+	if _, err := kv.Compact(context.TODO(), 4); err != nil {
 		t.Fatal(err)
 	}
 	wch := w.Watch(context.Background(), "foo", clientv3.WithRev(2))
@@ -487,7 +487,7 @@ func testWatchWithProgressNotify(t *testing.T, watchOnPut bool) {
 		} else if len(resp.Events) != 0 { // wait for notification otherwise
 			t.Fatalf("expected no events, but got %+v", resp.Events)
 		}
-	case <-time.After(2 * pi):
+	case <-time.After(time.Duration(1.5 * float64(pi))):
 		t.Fatalf("watch response expected in %v, but timed out", pi)
 	}
 }
@@ -673,3 +673,131 @@ func TestWatchWithRequireLeader(t *testing.T) {
 		t.Fatalf("expected response, got closed channel")
 	}
 }
+
+// TestWatchOverlapContextCancel stresses the watcher stream teardown path by
+// creating/canceling watchers to ensure that new watchers are not taken down
+// by a torn down watch stream. The sort of race that's being detected:
+//     1. create w1 using a cancelable ctx with %v as "ctx"
+//     2. cancel ctx
+//     3. watcher client begins tearing down watcher grpc stream since no more watchers
+//     3. start creating watcher w2 using a new "ctx" (not canceled), attaches to old grpc stream
+//     4. watcher client finishes tearing down stream on "ctx"
+//     5. w2 comes back canceled
+func TestWatchOverlapContextCancel(t *testing.T) {
+	f := func(clus *integration.ClusterV3) {}
+	testWatchOverlapContextCancel(t, f)
+}
+
+func TestWatchOverlapDropConnContextCancel(t *testing.T) {
+	f := func(clus *integration.ClusterV3) {
+		clus.Members[0].DropConnections()
+	}
+	testWatchOverlapContextCancel(t, f)
+}
+
+func testWatchOverlapContextCancel(t *testing.T, f func(*integration.ClusterV3)) {
+	defer testutil.AfterTest(t)
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 1})
+	defer clus.Terminate(t)
+
+	// each unique context "%v" has a unique grpc stream
+	n := 100
+	ctxs, ctxc := make([]context.Context, 5), make([]chan struct{}, 5)
+	for i := range ctxs {
+		// make "%v" unique
+		ctxs[i] = context.WithValue(context.TODO(), "key", i)
+		// limits the maximum number of outstanding watchers per stream
+		ctxc[i] = make(chan struct{}, 2)
+	}
+
+	// issue concurrent watches on "abc" with cancel
+	cli := clus.RandClient()
+	if _, err := cli.Put(context.TODO(), "abc", "def"); err != nil {
+		t.Fatal(err)
+	}
+	ch := make(chan struct{}, n)
+	for i := 0; i < n; i++ {
+		go func() {
+			defer func() { ch <- struct{}{} }()
+			idx := rand.Intn(len(ctxs))
+			ctx, cancel := context.WithCancel(ctxs[idx])
+			ctxc[idx] <- struct{}{}
+			wch := cli.Watch(ctx, "abc", clientv3.WithRev(1))
+			f(clus)
+			select {
+			case _, ok := <-wch:
+				if !ok {
+					t.Fatalf("unexpected closed channel %p", wch)
+				}
+			// may take a second or two to reestablish a watcher because of
+			// grpc backoff policies for disconnects
+			case <-time.After(5 * time.Second):
+				t.Errorf("timed out waiting for watch on %p", wch)
+			}
+			// randomize how cancel overlaps with watch creation
+			if rand.Intn(2) == 0 {
+				<-ctxc[idx]
+				cancel()
+			} else {
+				cancel()
+				<-ctxc[idx]
+			}
+		}()
+	}
+	// join on watches
+	for i := 0; i < n; i++ {
+		select {
+		case <-ch:
+		case <-time.After(5 * time.Second):
+			t.Fatalf("timed out waiting for completed watch")
+		}
+	}
+}
+
+// TestWatchCanelAndCloseClient ensures that canceling a watcher then immediately
+// closing the client does not return a client closing error.
+func TestWatchCancelAndCloseClient(t *testing.T) {
+	defer testutil.AfterTest(t)
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 1})
+	defer clus.Terminate(t)
+	cli := clus.Client(0)
+	ctx, cancel := context.WithCancel(context.Background())
+	wch := cli.Watch(ctx, "abc")
+	donec := make(chan struct{})
+	go func() {
+		defer close(donec)
+		select {
+		case wr, ok := <-wch:
+			if ok {
+				t.Fatalf("expected closed watch after cancel(), got resp=%+v err=%v", wr, wr.Err())
+			}
+		case <-time.After(5 * time.Second):
+			t.Fatal("timed out waiting for closed channel")
+		}
+	}()
+	cancel()
+	if err := cli.Close(); err != nil {
+		t.Fatal(err)
+	}
+	<-donec
+	clus.TakeClient(0)
+}
+
+// TestWatchCancelDisconnected ensures canceling a watcher works when
+// its grpc stream is disconnected / reconnecting.
+func TestWatchCancelDisconnected(t *testing.T) {
+	defer testutil.AfterTest(t)
+	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 1})
+	defer clus.Terminate(t)
+	cli := clus.Client(0)
+	ctx, cancel := context.WithCancel(context.Background())
+	// add more watches than can be resumed before the cancel
+	wch := cli.Watch(ctx, "abc")
+	clus.Members[0].Stop(t)
+	cancel()
+	select {
+	case <-wch:
+	case <-time.After(time.Second):
+		t.Fatal("took too long to cancel disconnected watcher")
+	}
+}

clientv3/kv.go

@@ -17,13 +17,15 @@ package clientv3
 import (
 	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
 	"golang.org/x/net/context"
+	"google.golang.org/grpc"
 )
 
 type (
-	PutResponse    pb.PutResponse
+	CompactResponse pb.CompactionResponse
-	GetResponse    pb.RangeResponse
+	PutResponse     pb.PutResponse
-	DeleteResponse pb.DeleteRangeResponse
+	GetResponse     pb.RangeResponse
-	TxnResponse    pb.TxnResponse
+	DeleteResponse  pb.DeleteRangeResponse
+	TxnResponse     pb.TxnResponse
 )
 
 type KV interface {
@@ -47,7 +49,7 @@ type KV interface {
 	Delete(ctx context.Context, key string, opts ...OpOption) (*DeleteResponse, error)
 
 	// Compact compacts etcd KV history before the given rev.
-	Compact(ctx context.Context, rev int64, opts ...CompactOption) error
+	Compact(ctx context.Context, rev int64, opts ...CompactOption) (*CompactResponse, error)
 
 	// Do applies a single Op on KV without a transaction.
 	// Do is useful when declaring operations to be issued at a later time
@@ -80,7 +82,7 @@ type kv struct {
 }
 
 func NewKV(c *Client) KV {
-	return &kv{remote: pb.NewKVClient(c.conn)}
+	return &kv{remote: RetryKVClient(c)}
 }
 
 func (kv *kv) Put(ctx context.Context, key, val string, opts ...OpOption) (*PutResponse, error) {
@@ -98,11 +100,12 @@ func (kv *kv) Delete(ctx context.Context, key string, opts ...OpOption) (*Delete
 	return r.del, toErr(ctx, err)
 }
 
-func (kv *kv) Compact(ctx context.Context, rev int64, opts ...CompactOption) error {
+func (kv *kv) Compact(ctx context.Context, rev int64, opts ...CompactOption) (*CompactResponse, error) {
-	if _, err := kv.remote.Compact(ctx, OpCompact(rev, opts...).toRequest()); err != nil {
+	resp, err := kv.remote.Compact(ctx, OpCompact(rev, opts...).toRequest(), grpc.FailFast(false))
-		return toErr(ctx, err)
+	if err != nil {
+		return nil, toErr(ctx, err)
 	}
-	return nil
+	return (*CompactResponse)(resp), err
 }
 
 func (kv *kv) Txn(ctx context.Context) Txn {
@@ -148,20 +151,20 @@ func (kv *kv) do(ctx context.Context, op Op) (OpResponse, error) {
 			r.SortTarget = pb.RangeRequest_SortTarget(op.sort.Target)
 		}
 
-		resp, err = kv.remote.Range(ctx, r)
+		resp, err = kv.remote.Range(ctx, r, grpc.FailFast(false))
 		if err == nil {
 			return OpResponse{get: (*GetResponse)(resp)}, nil
 		}
 	case tPut:
 		var resp *pb.PutResponse
-		r := &pb.PutRequest{Key: op.key, Value: op.val, Lease: int64(op.leaseID)}
+		r := &pb.PutRequest{Key: op.key, Value: op.val, Lease: int64(op.leaseID), PrevKv: op.prevKV}
 		resp, err = kv.remote.Put(ctx, r)
 		if err == nil {
 			return OpResponse{put: (*PutResponse)(resp)}, nil
 		}
 	case tDeleteRange:
 		var resp *pb.DeleteRangeResponse
-		r := &pb.DeleteRangeRequest{Key: op.key, RangeEnd: op.end}
+		r := &pb.DeleteRangeRequest{Key: op.key, RangeEnd: op.end, PrevKv: op.prevKV}
 		resp, err = kv.remote.DeleteRange(ctx, r)
 		if err == nil {
 			return OpResponse{del: (*DeleteResponse)(resp)}, nil

clientv3/lease.go

@@ -21,6 +21,7 @@ import (
 	"github.com/coreos/etcd/etcdserver/api/v3rpc/rpctypes"
 	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
 	"golang.org/x/net/context"
+	"google.golang.org/grpc"
 )
 
 type (
@@ -109,7 +110,7 @@ func NewLease(c *Client) Lease {
 	l := &lessor{
 		donec:                 make(chan struct{}),
 		keepAlives:            make(map[LeaseID]*keepAlive),
-		remote:                pb.NewLeaseClient(c.conn),
+		remote:                RetryLeaseClient(c),
 		firstKeepAliveTimeout: c.cfg.DialTimeout + time.Second,
 	}
 	if l.firstKeepAliveTimeout == time.Second {
@@ -142,9 +143,6 @@ func (l *lessor) Grant(ctx context.Context, ttl int64) (*LeaseGrantResponse, err
 		if isHaltErr(cctx, err) {
 			return nil, toErr(ctx, err)
 		}
-		if nerr := l.newStream(); nerr != nil {
-			return nil, nerr
-		}
 	}
 }
 
@@ -163,9 +161,6 @@ func (l *lessor) Revoke(ctx context.Context, id LeaseID) (*LeaseRevokeResponse,
 		if isHaltErr(ctx, err) {
 			return nil, toErr(ctx, err)
 		}
-		if nerr := l.newStream(); nerr != nil {
-			return nil, nerr
-		}
 	}
 }
 
@@ -212,10 +207,6 @@ func (l *lessor) KeepAliveOnce(ctx context.Context, id LeaseID) (*LeaseKeepAlive
 		if isHaltErr(ctx, err) {
 			return nil, toErr(ctx, err)
 		}
-
-		if nerr := l.newStream(); nerr != nil {
-			return nil, nerr
-		}
 	}
 }
 
@@ -261,7 +252,7 @@ func (l *lessor) keepAliveOnce(ctx context.Context, id LeaseID) (*LeaseKeepAlive
 	cctx, cancel := context.WithCancel(ctx)
 	defer cancel()
 
-	stream, err := l.remote.LeaseKeepAlive(cctx)
+	stream, err := l.remote.LeaseKeepAlive(cctx, grpc.FailFast(false))
 	if err != nil {
 		return nil, toErr(ctx, err)
 	}
@@ -311,10 +302,23 @@ func (l *lessor) recvKeepAliveLoop() {
 
 // resetRecv opens a new lease stream and starts sending LeaseKeepAliveRequests
 func (l *lessor) resetRecv() (pb.Lease_LeaseKeepAliveClient, error) {
-	if err := l.newStream(); err != nil {
+	sctx, cancel := context.WithCancel(l.stopCtx)
+	stream, err := l.remote.LeaseKeepAlive(sctx, grpc.FailFast(false))
+	if err = toErr(sctx, err); err != nil {
+		cancel()
 		return nil, err
 	}
-	stream := l.getKeepAliveStream()
+
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	if l.stream != nil && l.streamCancel != nil {
+		l.stream.CloseSend()
+		l.streamCancel()
+	}
+
+	l.streamCancel = cancel
+	l.stream = stream
+
 	go l.sendKeepAliveLoop(stream)
 	return stream, nil
 }
@@ -410,32 +414,6 @@ func (l *lessor) sendKeepAliveLoop(stream pb.Lease_LeaseKeepAliveClient) {
 	}
 }
 
-func (l *lessor) getKeepAliveStream() pb.Lease_LeaseKeepAliveClient {
-	l.mu.Lock()
-	defer l.mu.Unlock()
-	return l.stream
-}
-
-func (l *lessor) newStream() error {
-	sctx, cancel := context.WithCancel(l.stopCtx)
-	stream, err := l.remote.LeaseKeepAlive(sctx)
-	if err != nil {
-		cancel()
-		return toErr(sctx, err)
-	}
-
-	l.mu.Lock()
-	defer l.mu.Unlock()
-	if l.stream != nil && l.streamCancel != nil {
-		l.stream.CloseSend()
-		l.streamCancel()
-	}
-
-	l.streamCancel = cancel
-	l.stream = stream
-	return nil
-}
-
 func (ka *keepAlive) Close() {
 	close(ka.donec)
 	for _, ch := range ka.chs {

clientv3/maintenance.go

@@ -19,6 +19,7 @@ import (
 
 	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
 	"golang.org/x/net/context"
+	"google.golang.org/grpc"
 )
 
 type (
@@ -67,7 +68,7 @@ func (m *maintenance) AlarmList(ctx context.Context) (*AlarmResponse, error) {
 		Alarm:    pb.AlarmType_NONE, // all
 	}
 	for {
-		resp, err := m.remote.Alarm(ctx, req)
+		resp, err := m.remote.Alarm(ctx, req, grpc.FailFast(false))
 		if err == nil {
 			return (*AlarmResponse)(resp), nil
 		}
@@ -100,7 +101,7 @@ func (m *maintenance) AlarmDisarm(ctx context.Context, am *AlarmMember) (*AlarmR
 		return &ret, nil
 	}
 
-	resp, err := m.remote.Alarm(ctx, req)
+	resp, err := m.remote.Alarm(ctx, req, grpc.FailFast(false))
 	if err == nil {
 		return (*AlarmResponse)(resp), nil
 	}
@@ -114,7 +115,7 @@ func (m *maintenance) Defragment(ctx context.Context, endpoint string) (*Defragm
 	}
 	defer conn.Close()
 	remote := pb.NewMaintenanceClient(conn)
-	resp, err := remote.Defragment(ctx, &pb.DefragmentRequest{})
+	resp, err := remote.Defragment(ctx, &pb.DefragmentRequest{}, grpc.FailFast(false))
 	if err != nil {
 		return nil, toErr(ctx, err)
 	}
@@ -128,7 +129,7 @@ func (m *maintenance) Status(ctx context.Context, endpoint string) (*StatusRespo
 	}
 	defer conn.Close()
 	remote := pb.NewMaintenanceClient(conn)
-	resp, err := remote.Status(ctx, &pb.StatusRequest{})
+	resp, err := remote.Status(ctx, &pb.StatusRequest{}, grpc.FailFast(false))
 	if err != nil {
 		return nil, toErr(ctx, err)
 	}
@@ -136,7 +137,7 @@ func (m *maintenance) Status(ctx context.Context, endpoint string) (*StatusRespo
 }
 
 func (m *maintenance) Snapshot(ctx context.Context) (io.ReadCloser, error) {
-	ss, err := m.remote.Snapshot(ctx, &pb.SnapshotRequest{})
+	ss, err := m.remote.Snapshot(ctx, &pb.SnapshotRequest{}, grpc.FailFast(false))
 	if err != nil {
 		return nil, toErr(ctx, err)
 	}

clientv3/mirror/syncer.go

@@ -78,7 +78,7 @@ func (s *syncer) SyncBase(ctx context.Context) (<-chan clientv3.GetResponse, cha
 			// If len(s.prefix) != 0, we will sync key-value space with given prefix.
 			// We then range from the prefix to the next prefix if exists. Or we will
 			// range from the prefix to the end if the next prefix does not exists.
-			opts = append(opts, clientv3.WithPrefix())
+			opts = append(opts, clientv3.WithRange(clientv3.GetPrefixRangeEnd(s.prefix)))
 			key = s.prefix
 		}
 

clientv3/op.go

@@ -47,6 +47,9 @@ type Op struct {
 	// for range, watch
 	rev int64
 
+	// for watch, put, delete
+	prevKV bool
+
 	// progressNotify is for progress updates.
 	progressNotify bool
 
@@ -73,10 +76,10 @@ func (op Op) toRequestOp() *pb.RequestOp {
 		}
 		return &pb.RequestOp{Request: &pb.RequestOp_RequestRange{RequestRange: r}}
 	case tPut:
-		r := &pb.PutRequest{Key: op.key, Value: op.val, Lease: int64(op.leaseID)}
+		r := &pb.PutRequest{Key: op.key, Value: op.val, Lease: int64(op.leaseID), PrevKv: op.prevKV}
 		return &pb.RequestOp{Request: &pb.RequestOp_RequestPut{RequestPut: r}}
 	case tDeleteRange:
-		r := &pb.DeleteRangeRequest{Key: op.key, RangeEnd: op.end}
+		r := &pb.DeleteRangeRequest{Key: op.key, RangeEnd: op.end, PrevKv: op.prevKV}
 		return &pb.RequestOp{Request: &pb.RequestOp_RequestDeleteRange{RequestDeleteRange: r}}
 	default:
 		panic("Unknown Op")
@@ -182,6 +185,12 @@ func WithSort(target SortTarget, order SortOrder) OpOption {
 	}
 }
 
+// GetPrefixRangeEnd gets the range end of the prefix.
+// 'Get(foo, WithPrefix())' is equal to 'Get(foo, WithRange(GetPrefixRangeEnd(foo))'.
+func GetPrefixRangeEnd(prefix string) string {
+	return string(getPrefix([]byte(prefix)))
+}
+
 func getPrefix(key []byte) []byte {
 	end := make([]byte, len(key))
 	copy(end, key)
@@ -206,14 +215,15 @@ func WithPrefix() OpOption {
 	}
 }
 
-// WithRange specifies the range of 'Get' or 'Delete' requests.
+// WithRange specifies the range of 'Get', 'Delete', 'Watch' requests.
 // For example, 'Get' requests with 'WithRange(end)' returns
 // the keys in the range [key, end).
+// endKey must be lexicographically greater than start key.
 func WithRange(endKey string) OpOption {
 	return func(op *Op) { op.end = []byte(endKey) }
 }
 
-// WithFromKey specifies the range of 'Get' or 'Delete' requests
+// WithFromKey specifies the range of 'Get', 'Delete', 'Watch' requests
 // to be equal or greater than the key in the argument.
 func WithFromKey() OpOption { return WithRange("\x00") }
 
@@ -265,3 +275,11 @@ func WithProgressNotify() OpOption {
 		op.progressNotify = true
 	}
 }
+
+// WithPrevKV gets the previous key-value pair before the event happens. If the previous KV is already compacted,
+// nothing will be returned.
+func WithPrevKV() OpOption {
+	return func(op *Op) {
+		op.prevKV = true
+	}
+}

clientv3/retry.go

@@ -0,0 +1,243 @@
+// Copyright 2016 The etcd Authors
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package clientv3
+
+import (
+	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
+	"golang.org/x/net/context"
+	"google.golang.org/grpc"
+)
+
+type rpcFunc func(ctx context.Context) error
+type retryRpcFunc func(context.Context, rpcFunc)
+
+func (c *Client) newRetryWrapper() retryRpcFunc {
+	return func(rpcCtx context.Context, f rpcFunc) {
+		for {
+			err := f(rpcCtx)
+			// ignore grpc conn closing on fail-fast calls; they are transient errors
+			if err == nil || !isConnClosing(err) {
+				return
+			}
+			select {
+			case <-c.balancer.ConnectNotify():
+			case <-rpcCtx.Done():
+			case <-c.ctx.Done():
+				return
+			}
+		}
+	}
+}
+
+type retryKVClient struct {
+	pb.KVClient
+	retryf retryRpcFunc
+}
+
+// RetryKVClient implements a KVClient that uses the client's FailFast retry policy.
+func RetryKVClient(c *Client) pb.KVClient {
+	return &retryKVClient{pb.NewKVClient(c.conn), c.retryWrapper}
+}
+
+func (rkv *retryKVClient) Put(ctx context.Context, in *pb.PutRequest, opts ...grpc.CallOption) (resp *pb.PutResponse, err error) {
+	rkv.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rkv.KVClient.Put(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rkv *retryKVClient) DeleteRange(ctx context.Context, in *pb.DeleteRangeRequest, opts ...grpc.CallOption) (resp *pb.DeleteRangeResponse, err error) {
+	rkv.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rkv.KVClient.DeleteRange(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rkv *retryKVClient) Txn(ctx context.Context, in *pb.TxnRequest, opts ...grpc.CallOption) (resp *pb.TxnResponse, err error) {
+	rkv.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rkv.KVClient.Txn(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rkv *retryKVClient) Compact(ctx context.Context, in *pb.CompactionRequest, opts ...grpc.CallOption) (resp *pb.CompactionResponse, err error) {
+	rkv.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rkv.KVClient.Compact(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+type retryLeaseClient struct {
+	pb.LeaseClient
+	retryf retryRpcFunc
+}
+
+// RetryLeaseClient implements a LeaseClient that uses the client's FailFast retry policy.
+func RetryLeaseClient(c *Client) pb.LeaseClient {
+	return &retryLeaseClient{pb.NewLeaseClient(c.conn), c.retryWrapper}
+}
+
+func (rlc *retryLeaseClient) LeaseGrant(ctx context.Context, in *pb.LeaseGrantRequest, opts ...grpc.CallOption) (resp *pb.LeaseGrantResponse, err error) {
+	rlc.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rlc.LeaseClient.LeaseGrant(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+
+}
+
+func (rlc *retryLeaseClient) LeaseRevoke(ctx context.Context, in *pb.LeaseRevokeRequest, opts ...grpc.CallOption) (resp *pb.LeaseRevokeResponse, err error) {
+	rlc.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rlc.LeaseClient.LeaseRevoke(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+type retryClusterClient struct {
+	pb.ClusterClient
+	retryf retryRpcFunc
+}
+
+// RetryClusterClient implements a ClusterClient that uses the client's FailFast retry policy.
+func RetryClusterClient(c *Client) pb.ClusterClient {
+	return &retryClusterClient{pb.NewClusterClient(c.conn), c.retryWrapper}
+}
+
+func (rcc *retryClusterClient) MemberAdd(ctx context.Context, in *pb.MemberAddRequest, opts ...grpc.CallOption) (resp *pb.MemberAddResponse, err error) {
+	rcc.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rcc.ClusterClient.MemberAdd(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rcc *retryClusterClient) MemberRemove(ctx context.Context, in *pb.MemberRemoveRequest, opts ...grpc.CallOption) (resp *pb.MemberRemoveResponse, err error) {
+	rcc.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rcc.ClusterClient.MemberRemove(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rcc *retryClusterClient) MemberUpdate(ctx context.Context, in *pb.MemberUpdateRequest, opts ...grpc.CallOption) (resp *pb.MemberUpdateResponse, err error) {
+	rcc.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rcc.ClusterClient.MemberUpdate(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+type retryAuthClient struct {
+	pb.AuthClient
+	retryf retryRpcFunc
+}
+
+// RetryAuthClient implements a AuthClient that uses the client's FailFast retry policy.
+func RetryAuthClient(c *Client) pb.AuthClient {
+	return &retryAuthClient{pb.NewAuthClient(c.conn), c.retryWrapper}
+}
+
+func (rac *retryAuthClient) AuthEnable(ctx context.Context, in *pb.AuthEnableRequest, opts ...grpc.CallOption) (resp *pb.AuthEnableResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.AuthEnable(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) AuthDisable(ctx context.Context, in *pb.AuthDisableRequest, opts ...grpc.CallOption) (resp *pb.AuthDisableResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.AuthDisable(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) UserAdd(ctx context.Context, in *pb.AuthUserAddRequest, opts ...grpc.CallOption) (resp *pb.AuthUserAddResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.UserAdd(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) UserDelete(ctx context.Context, in *pb.AuthUserDeleteRequest, opts ...grpc.CallOption) (resp *pb.AuthUserDeleteResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.UserDelete(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) UserChangePassword(ctx context.Context, in *pb.AuthUserChangePasswordRequest, opts ...grpc.CallOption) (resp *pb.AuthUserChangePasswordResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.UserChangePassword(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) UserGrantRole(ctx context.Context, in *pb.AuthUserGrantRoleRequest, opts ...grpc.CallOption) (resp *pb.AuthUserGrantRoleResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.UserGrantRole(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) UserRevokeRole(ctx context.Context, in *pb.AuthUserRevokeRoleRequest, opts ...grpc.CallOption) (resp *pb.AuthUserRevokeRoleResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.UserRevokeRole(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) RoleAdd(ctx context.Context, in *pb.AuthRoleAddRequest, opts ...grpc.CallOption) (resp *pb.AuthRoleAddResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.RoleAdd(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) RoleDelete(ctx context.Context, in *pb.AuthRoleDeleteRequest, opts ...grpc.CallOption) (resp *pb.AuthRoleDeleteResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.RoleDelete(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) RoleGrantPermission(ctx context.Context, in *pb.AuthRoleGrantPermissionRequest, opts ...grpc.CallOption) (resp *pb.AuthRoleGrantPermissionResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.RoleGrantPermission(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}
+
+func (rac *retryAuthClient) RoleRevokePermission(ctx context.Context, in *pb.AuthRoleRevokePermissionRequest, opts ...grpc.CallOption) (resp *pb.AuthRoleRevokePermissionResponse, err error) {
+	rac.retryf(ctx, func(rctx context.Context) error {
+		resp, err = rac.AuthClient.RoleRevokePermission(rctx, in, opts...)
+		return err
+	})
+	return resp, err
+}

clientv3/watch.go

@@ -23,6 +23,7 @@ import (
 	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
 	mvccpb "github.com/coreos/etcd/mvcc/mvccpb"
 	"golang.org/x/net/context"
+	"google.golang.org/grpc"
 )
 
 const (
@@ -60,6 +61,9 @@ type WatchResponse struct {
 	// the channel sends a final response that has Canceled set to true with a non-nil Err().
 	Canceled bool
 
+	// created is used to indicate the creation of the watcher.
+	created bool
+
 	closeErr error
 }
 
@@ -88,7 +92,7 @@ func (wr *WatchResponse) Err() error {
 
 // IsProgressNotify returns true if the WatchResponse is progress notification.
 func (wr *WatchResponse) IsProgressNotify() bool {
-	return len(wr.Events) == 0 && !wr.Canceled
+	return len(wr.Events) == 0 && !wr.Canceled && !wr.created && wr.CompactRevision == 0 && wr.Header.Revision != 0
 }
 
 // watcher implements the Watcher interface
@@ -101,6 +105,7 @@ type watcher struct {
 	streams map[string]*watchGrpcStream
 }
 
+// watchGrpcStream tracks all watch resources attached to a single grpc stream.
 type watchGrpcStream struct {
 	owner  *watcher
 	remote pb.WatchClient
@@ -111,23 +116,25 @@ type watchGrpcStream struct {
 	ctxKey string
 	cancel context.CancelFunc
 
-	// mu protects the streams map
+	// substreams holds all active watchers on this grpc stream
-	mu sync.RWMutex
+	substreams map[int64]*watcherStream
-	// streams holds all active watchers
+	// resuming holds all resuming watchers on this grpc stream
-	streams map[int64]*watcherStream
+	resuming []*watcherStream
 
 	// reqc sends a watch request from Watch() to the main goroutine
 	reqc chan *watchRequest
 	// respc receives data from the watch client
 	respc chan *pb.WatchResponse
-	// stopc is sent to the main goroutine to stop all processing
-	stopc chan struct{}
 	// donec closes to broadcast shutdown
 	donec chan struct{}
 	// errc transmits errors from grpc Recv to the watch stream reconn logic
 	errc chan error
+	// closingc gets the watcherStream of closing watchers
+	closingc chan *watcherStream
 
-	// the error that closed the watch stream
+	// resumec closes to signal that all substreams should begin resuming
+	resumec chan struct{}
+	// closeErr is the error that closed the watch stream
 	closeErr error
 }
 
@@ -139,6 +146,8 @@ type watchRequest struct {
 	rev int64
 	// progressNotify is for progress updates.
 	progressNotify bool
+	// get the previous key-value pair before the event happens
+	prevKV bool
 	// retc receives a chan WatchResponse once the watcher is established
 	retc chan chan WatchResponse
 }
@@ -149,15 +158,18 @@ type watcherStream struct {
 	initReq watchRequest
 
 	// outc publishes watch responses to subscriber
-	outc chan<- WatchResponse
+	outc chan WatchResponse
 	// recvc buffers watch responses before publishing
 	recvc chan *WatchResponse
-	id    int64
+	// donec closes when the watcherStream goroutine stops.
+	donec chan struct{}
+	// closing is set to true when stream should be scheduled to shutdown.
+	closing bool
+	// id is the registered watch id on the grpc stream
+	id int64
 
-	// lastRev is revision last successfully sent over outc
+	// buf holds all events received from etcd but not yet consumed by the client
-	lastRev int64
+	buf []*WatchResponse
-	// resumec indicates the stream must recover at a given revision
-	resumec chan int64
 }
 
 func NewWatcher(c *Client) Watcher {
@@ -181,18 +193,19 @@ func (vc *valCtx) Err() error                  { return nil }
 func (w *watcher) newWatcherGrpcStream(inctx context.Context) *watchGrpcStream {
 	ctx, cancel := context.WithCancel(&valCtx{inctx})
 	wgs := &watchGrpcStream{
-		owner:   w,
+		owner:      w,
-		remote:  w.remote,
+		remote:     w.remote,
-		ctx:     ctx,
+		ctx:        ctx,
-		ctxKey:  fmt.Sprintf("%v", inctx),
+		ctxKey:     fmt.Sprintf("%v", inctx),
-		cancel:  cancel,
+		cancel:     cancel,
-		streams: make(map[int64]*watcherStream),
+		substreams: make(map[int64]*watcherStream),
 
-		respc: make(chan *pb.WatchResponse),
+		respc:    make(chan *pb.WatchResponse),
-		reqc:  make(chan *watchRequest),
+		reqc:     make(chan *watchRequest),
-		stopc: make(chan struct{}),
+		donec:    make(chan struct{}),
-		donec: make(chan struct{}),
+		errc:     make(chan error, 1),
-		errc:  make(chan error, 1),
+		closingc: make(chan *watcherStream),
+		resumec:  make(chan struct{}),
 	}
 	go wgs.run()
 	return wgs
@@ -202,14 +215,14 @@ func (w *watcher) newWatcherGrpcStream(inctx context.Context) *watchGrpcStream {
 func (w *watcher) Watch(ctx context.Context, key string, opts ...OpOption) WatchChan {
 	ow := opWatch(key, opts...)
 
-	retc := make(chan chan WatchResponse, 1)
 	wr := &watchRequest{
 		ctx:            ctx,
 		key:            string(ow.key),
 		end:            string(ow.end),
 		rev:            ow.rev,
 		progressNotify: ow.progressNotify,
-		retc:           retc,
+		prevKV:         ow.prevKV,
+		retc:           make(chan chan WatchResponse, 1),
 	}
 
 	ok := false
@@ -253,7 +266,7 @@ func (w *watcher) Watch(ctx context.Context, key string, opts ...OpOption) Watch
 	// receive channel
 	if ok {
 		select {
-		case ret := <-retc:
+		case ret := <-wr.retc:
 			return ret
 		case <-ctx.Done():
 		case <-donec:
@@ -284,7 +297,7 @@ func (w *watcher) Close() (err error) {
 }
 
 func (w *watchGrpcStream) Close() (err error) {
-	close(w.stopc)
+	w.cancel()
 	<-w.donec
 	select {
 	case err = <-w.errc:
@@ -293,65 +306,57 @@ func (w *watchGrpcStream) Close() (err error) {
 	return toErr(w.ctx, err)
 }
 
-func (w *watchGrpcStream) addStream(resp *pb.WatchResponse, pendingReq *watchRequest) {
+func (w *watcher) closeStream(wgs *watchGrpcStream) {
-	if pendingReq == nil {
-		// no pending request; ignore
-		return
-	}
-	if resp.Canceled || resp.CompactRevision != 0 {
-		// a cancel at id creation time means the start revision has
-		// been compacted out of the store
-		ret := make(chan WatchResponse, 1)
-		ret <- WatchResponse{
-			Header:          *resp.Header,
-			CompactRevision: resp.CompactRevision,
-			Canceled:        true}
-		close(ret)
-		pendingReq.retc <- ret
-		return
-	}
-
-	ret := make(chan WatchResponse)
-	if resp.WatchId == -1 {
-		// failed; no channel
-		close(ret)
-		pendingReq.retc <- ret
-		return
-	}
-
-	ws := &watcherStream{
-		initReq: *pendingReq,
-		id:      resp.WatchId,
-		outc:    ret,
-		// buffered so unlikely to block on sending while holding mu
-		recvc:   make(chan *WatchResponse, 4),
-		resumec: make(chan int64),
-	}
-
-	if pendingReq.rev == 0 {
-		// note the header revision so that a put following a current watcher
-		// disconnect will arrive on the watcher channel after reconnect
-		ws.initReq.rev = resp.Header.Revision
-	}
-
 	w.mu.Lock()
-	w.streams[ws.id] = ws
+	close(wgs.donec)
+	wgs.cancel()
+	if w.streams != nil {
+		delete(w.streams, wgs.ctxKey)
+	}
 	w.mu.Unlock()
-
-	// pass back the subscriber channel for the watcher
-	pendingReq.retc <- ret
-
-	// send messages to subscriber
-	go w.serveStream(ws)
 }
 
-// closeStream closes the watcher resources and removes it
+func (w *watchGrpcStream) addSubstream(resp *pb.WatchResponse, ws *watcherStream) {
-func (w *watchGrpcStream) closeStream(ws *watcherStream) {
+	if resp.WatchId == -1 {
-	// cancels request stream; subscriber receives nil channel
+		// failed; no channel
-	close(ws.initReq.retc)
+		close(ws.recvc)
-	// close subscriber's channel
+		return
+	}
+	ws.id = resp.WatchId
+	w.substreams[ws.id] = ws
+}
+
+func (w *watchGrpcStream) sendCloseSubstream(ws *watcherStream, resp *WatchResponse) {
+	select {
+	case ws.outc <- *resp:
+	case <-ws.initReq.ctx.Done():
+	case <-time.After(closeSendErrTimeout):
+	}
 	close(ws.outc)
-	delete(w.streams, ws.id)
+}
+
+func (w *watchGrpcStream) closeSubstream(ws *watcherStream) {
+	// send channel response in case stream was never established
+	select {
+	case ws.initReq.retc <- ws.outc:
+	default:
+	}
+	// close subscriber's channel
+	if closeErr := w.closeErr; closeErr != nil && ws.initReq.ctx.Err() == nil {
+		go w.sendCloseSubstream(ws, &WatchResponse{closeErr: w.closeErr})
+	} else if ws.outc != nil {
+		close(ws.outc)
+	}
+	if ws.id != -1 {
+		delete(w.substreams, ws.id)
+		return
+	}
+	for i := range w.resuming {
+		if w.resuming[i] == ws {
+			w.resuming[i] = nil
+			return
+		}
+	}
 }
 
 // run is the root of the goroutines for managing a watcher client
@@ -359,15 +364,29 @@ func (w *watchGrpcStream) run() {
 	var wc pb.Watch_WatchClient
 	var closeErr error
 
+	// substreams marked to close but goroutine still running; needed for
+	// avoiding double-closing recvc on grpc stream teardown
+	closing := make(map[*watcherStream]struct{})
+
 	defer func() {
-		w.owner.mu.Lock()
 		w.closeErr = closeErr
-		if w.owner.streams != nil {
+		// shutdown substreams and resuming substreams
-			delete(w.owner.streams, w.ctxKey)
+		for _, ws := range w.substreams {
+			if _, ok := closing[ws]; !ok {
+				close(ws.recvc)
+			}
 		}
-		close(w.donec)
+		for _, ws := range w.resuming {
-		w.owner.mu.Unlock()
+			if _, ok := closing[ws]; ws != nil && !ok {
-		w.cancel()
+				close(ws.recvc)
+			}
+		}
+		w.joinSubstreams()
+		for toClose := len(w.substreams) + len(w.resuming); toClose > 0; toClose-- {
+			w.closeSubstream(<-w.closingc)
+		}
+
+		w.owner.closeStream(w)
 	}()
 
 	// start a stream with the etcd grpc server
@@ -375,42 +394,49 @@ func (w *watchGrpcStream) run() {
 		return
 	}
 
-	var pendingReq, failedReq *watchRequest
-	curReqC := w.reqc
 	cancelSet := make(map[int64]struct{})
 
 	for {
 		select {
 		// Watch() requested
-		case pendingReq = <-curReqC:
+		case wreq := <-w.reqc:
-			// no more watch requests until there's a response
+			outc := make(chan WatchResponse, 1)
-			curReqC = nil
+			ws := &watcherStream{
-			if err := wc.Send(pendingReq.toPB()); err == nil {
+				initReq: *wreq,
-				// pendingReq now waits on w.respc
+				id:      -1,
-				break
+				outc:    outc,
+				// unbufffered so resumes won't cause repeat events
+				recvc: make(chan *WatchResponse),
+			}
+
+			ws.donec = make(chan struct{})
+			go w.serveSubstream(ws, w.resumec)
+
+			// queue up for watcher creation/resume
+			w.resuming = append(w.resuming, ws)
+			if len(w.resuming) == 1 {
+				// head of resume queue, can register a new watcher
+				wc.Send(ws.initReq.toPB())
 			}
-			failedReq = pendingReq
 		// New events from the watch client
 		case pbresp := <-w.respc:
 			switch {
 			case pbresp.Created:
-				// response to pending req, try to add
+				// response to head of queue creation
-				w.addStream(pbresp, pendingReq)
+				if ws := w.resuming[0]; ws != nil {
-				pendingReq = nil
+					w.addSubstream(pbresp, ws)
-				curReqC = w.reqc
+					w.dispatchEvent(pbresp)
+					w.resuming[0] = nil
+				}
+				if ws := w.nextResume(); ws != nil {
+					wc.Send(ws.initReq.toPB())
+				}
 			case pbresp.Canceled:
 				delete(cancelSet, pbresp.WatchId)
-				// shutdown serveStream, if any
+				if ws, ok := w.substreams[pbresp.WatchId]; ok {
-				w.mu.Lock()
+					// signal to stream goroutine to update closingc
-				if ws, ok := w.streams[pbresp.WatchId]; ok {
 					close(ws.recvc)
-					delete(w.streams, ws.id)
+					closing[ws] = struct{}{}
-				}
-				numStreams := len(w.streams)
-				w.mu.Unlock()
-				if numStreams == 0 {
-					// don't leak watcher streams
-					return
 				}
 			default:
 				// dispatch to appropriate watch stream
@@ -431,7 +457,6 @@ func (w *watchGrpcStream) run() {
 				wc.Send(req)
 			}
 		// watch client failed to recv; spawn another if possible
-		// TODO report watch client errors from errc?
 		case err := <-w.errc:
 			if toErr(w.ctx, err) == v3rpc.ErrNoLeader {
 				closeErr = err
@@ -440,48 +465,58 @@ func (w *watchGrpcStream) run() {
 			if wc, closeErr = w.newWatchClient(); closeErr != nil {
 				return
 			}
-			curReqC = w.reqc
+			if ws := w.nextResume(); ws != nil {
-			if pendingReq != nil {
+				wc.Send(ws.initReq.toPB())
-				failedReq = pendingReq
 			}
 			cancelSet = make(map[int64]struct{})
-		case <-w.stopc:
+		case <-w.ctx.Done():
 			return
-		}
+		case ws := <-w.closingc:
-
+			w.closeSubstream(ws)
-		// send failed; queue for retry
+			delete(closing, ws)
-		if failedReq != nil {
+			if len(w.substreams)+len(w.resuming) == 0 {
-			go func(wr *watchRequest) {
+				// no more watchers on this stream, shutdown
-				select {
+				return
-				case w.reqc <- wr:
+			}
-				case <-wr.ctx.Done():
-				case <-w.donec:
-				}
-			}(pendingReq)
-			failedReq = nil
-			pendingReq = nil
 		}
 	}
 }
 
+// nextResume chooses the next resuming to register with the grpc stream. Abandoned
+// streams are marked as nil in the queue since the head must wait for its inflight registration.
+func (w *watchGrpcStream) nextResume() *watcherStream {
+	for len(w.resuming) != 0 {
+		if w.resuming[0] != nil {
+			return w.resuming[0]
+		}
+		w.resuming = w.resuming[1:len(w.resuming)]
+	}
+	return nil
+}
+
 // dispatchEvent sends a WatchResponse to the appropriate watcher stream
 func (w *watchGrpcStream) dispatchEvent(pbresp *pb.WatchResponse) bool {
-	w.mu.RLock()
+	ws, ok := w.substreams[pbresp.WatchId]
-	defer w.mu.RUnlock()
+	if !ok {
-	ws, ok := w.streams[pbresp.WatchId]
+		return false
+	}
 	events := make([]*Event, len(pbresp.Events))
 	for i, ev := range pbresp.Events {
 		events[i] = (*Event)(ev)
 	}
-	if ok {
+	wr := &WatchResponse{
-		wr := &WatchResponse{
+		Header:          *pbresp.Header,
-			Header:          *pbresp.Header,
+		Events:          events,
-			Events:          events,
+		CompactRevision: pbresp.CompactRevision,
-			CompactRevision: pbresp.CompactRevision,
+		created:         pbresp.Created,
-			Canceled:        pbresp.Canceled}
+		Canceled:        pbresp.Canceled,
-		ws.recvc <- wr
 	}
-	return ok
+	select {
+	case ws.recvc <- wr:
+	case <-ws.donec:
+		return false
+	}
+	return true
 }
 
 // serveWatchClient forwards messages from the grpc stream to run()
@@ -503,125 +538,171 @@ func (w *watchGrpcStream) serveWatchClient(wc pb.Watch_WatchClient) {
 	}
 }
 
-// serveStream forwards watch responses from run() to the subscriber
+// serveSubstream forwards watch responses from run() to the subscriber
-func (w *watchGrpcStream) serveStream(ws *watcherStream) {
+func (w *watchGrpcStream) serveSubstream(ws *watcherStream, resumec chan struct{}) {
-	emptyWr := &WatchResponse{}
+	if ws.closing {
-	wrs := []*WatchResponse{}
+		panic("created substream goroutine but substream is closing")
+	}
+
+	// nextRev is the minimum expected next revision
+	nextRev := ws.initReq.rev
 	resuming := false
-	closing := false
+	defer func() {
-	for !closing {
+		if !resuming {
+			ws.closing = true
+		}
+		close(ws.donec)
+		if !resuming {
+			w.closingc <- ws
+		}
+	}()
+
+	emptyWr := &WatchResponse{}
+	for {
 		curWr := emptyWr
 		outc := ws.outc
-		if len(wrs) > 0 {
+
-			curWr = wrs[0]
+		if len(ws.buf) > 0 && ws.buf[0].created {
+			select {
+			case ws.initReq.retc <- ws.outc:
+			default:
+			}
+			ws.buf = ws.buf[1:]
+		}
+
+		if len(ws.buf) > 0 {
+			curWr = ws.buf[0]
 		} else {
 			outc = nil
 		}
 		select {
 		case outc <- *curWr:
-			if wrs[0].Err() != nil {
+			if ws.buf[0].Err() != nil {
-				closing = true
-				break
-			}
-			var newRev int64
-			if len(wrs[0].Events) > 0 {
-				newRev = wrs[0].Events[len(wrs[0].Events)-1].Kv.ModRevision
-			} else {
-				newRev = wrs[0].Header.Revision
-			}
-			if newRev != ws.lastRev {
-				ws.lastRev = newRev
-			}
-			wrs[0] = nil
-			wrs = wrs[1:]
-		case wr, ok := <-ws.recvc:
-			if !ok {
-				// shutdown from closeStream
 				return
 			}
-			// resume up to last seen event if disconnected
+			ws.buf[0] = nil
-			if resuming && wr.Err() == nil {
+			ws.buf = ws.buf[1:]
-				resuming = false
+		case wr, ok := <-ws.recvc:
-				// trim events already seen
+			if !ok {
-				for i := 0; i < len(wr.Events); i++ {
+				// shutdown from closeSubstream
-					if wr.Events[i].Kv.ModRevision > ws.lastRev {
+				return
-						wr.Events = wr.Events[i:]
-						break
-					}
-				}
-				// only forward new events
-				if wr.Events[0].Kv.ModRevision == ws.lastRev {
-					break
-				}
 			}
-			resuming = false
+			// TODO pause channel if buffer gets too large
-			// TODO don't keep buffering if subscriber stops reading
+			ws.buf = append(ws.buf, wr)
-			wrs = append(wrs, wr)
+			nextRev = wr.Header.Revision
-		case resumeRev := <-ws.resumec:
+			if len(wr.Events) > 0 {
-			wrs = nil
+				nextRev = wr.Events[len(wr.Events)-1].Kv.ModRevision + 1
-			resuming = true
-			if resumeRev == -1 {
-				// pause serving stream while resume gets set up
-				break
 			}
-			if resumeRev != ws.lastRev {
+			ws.initReq.rev = nextRev
-				panic("unexpected resume revision")
+		case <-w.ctx.Done():
-			}
+			return
-		case <-w.donec:
-			closing = true
 		case <-ws.initReq.ctx.Done():
-			closing = true
+			return
+		case <-resumec:
+			resuming = true
+			return
 		}
 	}
-
-	// try to send off close error
-	if w.closeErr != nil {
-		select {
-		case ws.outc <- WatchResponse{closeErr: w.closeErr}:
-		case <-w.donec:
-		case <-time.After(closeSendErrTimeout):
-		}
-	}
-
-	w.mu.Lock()
-	w.closeStream(ws)
-	w.mu.Unlock()
 	// lazily send cancel message if events on missing id
 }
 
 func (w *watchGrpcStream) newWatchClient() (pb.Watch_WatchClient, error) {
-	ws, rerr := w.resume()
+	// mark all substreams as resuming
-	if rerr != nil {
+	close(w.resumec)
-		return nil, rerr
+	w.resumec = make(chan struct{})
+	w.joinSubstreams()
+	for _, ws := range w.substreams {
+		ws.id = -1
+		w.resuming = append(w.resuming, ws)
 	}
-	go w.serveWatchClient(ws)
+	// strip out nils, if any
-	return ws, nil
+	var resuming []*watcherStream
-}
+	for _, ws := range w.resuming {
-
+		if ws != nil {
-// resume creates a new WatchClient with all current watchers reestablished
+			resuming = append(resuming, ws)
-func (w *watchGrpcStream) resume() (ws pb.Watch_WatchClient, err error) {
+		}
-	for {
+	}
-		if ws, err = w.openWatchClient(); err != nil {
+	w.resuming = resuming
-			break
+	w.substreams = make(map[int64]*watcherStream)
-		} else if err = w.resumeWatchers(ws); err == nil {
+
-			break
+	// connect to grpc stream while accepting watcher cancelation
+	stopc := make(chan struct{})
+	donec := w.waitCancelSubstreams(stopc)
+	wc, err := w.openWatchClient()
+	close(stopc)
+	<-donec
+
+	// serve all non-closing streams, even if there's a client error
+	// so that the teardown path can shutdown the streams as expected.
+	for _, ws := range w.resuming {
+		if ws.closing {
+			continue
+		}
+		ws.donec = make(chan struct{})
+		go w.serveSubstream(ws, w.resumec)
+	}
+
+	if err != nil {
+		return nil, v3rpc.Error(err)
+	}
+
+	// receive data from new grpc stream
+	go w.serveWatchClient(wc)
+	return wc, nil
+}
+
+func (w *watchGrpcStream) waitCancelSubstreams(stopc <-chan struct{}) <-chan struct{} {
+	var wg sync.WaitGroup
+	wg.Add(len(w.resuming))
+	donec := make(chan struct{})
+	for i := range w.resuming {
+		go func(ws *watcherStream) {
+			defer wg.Done()
+			if ws.closing {
+				return
+			}
+			select {
+			case <-ws.initReq.ctx.Done():
+				// closed ws will be removed from resuming
+				ws.closing = true
+				close(ws.outc)
+				ws.outc = nil
+				go func() { w.closingc <- ws }()
+			case <-stopc:
+			}
+		}(w.resuming[i])
+	}
+	go func() {
+		defer close(donec)
+		wg.Wait()
+	}()
+	return donec
+}
+
+// joinSubstream waits for all substream goroutines to complete
+func (w *watchGrpcStream) joinSubstreams() {
+	for _, ws := range w.substreams {
+		<-ws.donec
+	}
+	for _, ws := range w.resuming {
+		if ws != nil {
+			<-ws.donec
 		}
 	}
-	return ws, v3rpc.Error(err)
 }
 
 // openWatchClient retries opening a watchclient until retryConnection fails
 func (w *watchGrpcStream) openWatchClient() (ws pb.Watch_WatchClient, err error) {
 	for {
 		select {
-		case <-w.stopc:
+		case <-w.ctx.Done():
 			if err == nil {
-				err = context.Canceled
+				return nil, w.ctx.Err()
 			}
 			return nil, err
 		default:
 		}
-		if ws, err = w.remote.Watch(w.ctx); ws != nil && err == nil {
+		if ws, err = w.remote.Watch(w.ctx, grpc.FailFast(false)); ws != nil && err == nil {
 			break
 		}
 		if isHaltErr(w.ctx, err) {
@@ -631,48 +712,6 @@ func (w *watchGrpcStream) openWatchClient() (ws pb.Watch_WatchClient, err error)
 	return ws, nil
 }
 
-// resumeWatchers rebuilds every registered watcher on a new client
-func (w *watchGrpcStream) resumeWatchers(wc pb.Watch_WatchClient) error {
-	w.mu.RLock()
-	streams := make([]*watcherStream, 0, len(w.streams))
-	for _, ws := range w.streams {
-		streams = append(streams, ws)
-	}
-	w.mu.RUnlock()
-
-	for _, ws := range streams {
-		// pause serveStream
-		ws.resumec <- -1
-
-		// reconstruct watcher from initial request
-		if ws.lastRev != 0 {
-			ws.initReq.rev = ws.lastRev
-		}
-		if err := wc.Send(ws.initReq.toPB()); err != nil {
-			return err
-		}
-
-		// wait for request ack
-		resp, err := wc.Recv()
-		if err != nil {
-			return err
-		} else if len(resp.Events) != 0 || !resp.Created {
-			return fmt.Errorf("watcher: unexpected response (%+v)", resp)
-		}
-
-		// id may be different since new remote watcher; update map
-		w.mu.Lock()
-		delete(w.streams, ws.id)
-		ws.id = resp.WatchId
-		w.streams[ws.id] = ws
-		w.mu.Unlock()
-
-		// unpause serveStream
-		ws.resumec <- ws.lastRev
-	}
-	return nil
-}
-
 // toPB converts an internal watch request structure to its protobuf messagefunc (wr *watchRequest)
 func (wr *watchRequest) toPB() *pb.WatchRequest {
 	req := &pb.WatchCreateRequest{
@@ -680,6 +719,7 @@ func (wr *watchRequest) toPB() *pb.WatchRequest {
 		Key:            []byte(wr.key),
 		RangeEnd:       []byte(wr.end),
 		ProgressNotify: wr.progressNotify,
+		PrevKv:         wr.prevKV,
 	}
 	cr := &pb.WatchRequest_CreateRequest{CreateRequest: req}
 	return &pb.WatchRequest{RequestUnion: cr}

cmd/Godeps/Godeps.json

@@ -1,6 +1,6 @@
 {
 	"ImportPath": "github.com/coreos/etcd",
-	"GoVersion": "go1.6",
+	"GoVersion": "go1.7",
 	"GodepVersion": "v74",
 	"Packages": [
 		"./..."
@@ -11,10 +11,6 @@
 			"Comment": "null-5",
 			"Rev": "'75cd24fc2f2c2a2088577d12123ddee5f54e0675'"
 		},
-		{
-			"ImportPath": "github.com/akrennmair/gopcap",
-			"Rev": "00e11033259acb75598ba416495bb708d864a010"
-		},
 		{
 			"ImportPath": "github.com/beorn7/perks/quantile",
 			"Rev": "b965b613227fddccbfffe13eae360ed3fa822f8d"
@@ -25,8 +21,8 @@
 		},
 		{
 			"ImportPath": "github.com/boltdb/bolt",
-			"Comment": "v1.2.1",
+			"Comment": "v1.3.0",
-			"Rev": "dfb21201d9270c1082d5fb0f07f500311ff72f18"
+			"Rev": "583e8937c61f1af6513608ccc75c97b6abdf4ff9"
 		},
 		{
 			"ImportPath": "github.com/cockroachdb/cmux",
@@ -38,18 +34,18 @@
 		},
 		{
 			"ImportPath": "github.com/coreos/go-systemd/daemon",
-			"Comment": "v3-6-gcea488b",
+			"Comment": "v10-13-gd6c05a1d",
-			"Rev": "cea488b4e6855fee89b6c22a811e3c5baca861b6"
+			"Rev": "d6c05a1dcbb5ac02b7653da4d99e5db340c20778"
 		},
 		{
 			"ImportPath": "github.com/coreos/go-systemd/journal",
-			"Comment": "v3-6-gcea488b",
+			"Comment": "v10-13-gd6c05a1d",
-			"Rev": "cea488b4e6855fee89b6c22a811e3c5baca861b6"
+			"Rev": "d6c05a1dcbb5ac02b7653da4d99e5db340c20778"
 		},
 		{
 			"ImportPath": "github.com/coreos/go-systemd/util",
-			"Comment": "v3-6-gcea488b",
+			"Comment": "v10-13-gd6c05a1d",
-			"Rev": "cea488b4e6855fee89b6c22a811e3c5baca861b6"
+			"Rev": "d6c05a1dcbb5ac02b7653da4d99e5db340c20778"
 		},
 		{
 			"ImportPath": "github.com/coreos/pkg/capnslog",
@@ -65,26 +61,14 @@
 			"ImportPath": "github.com/dustin/go-humanize",
 			"Rev": "8929fe90cee4b2cb9deb468b51fb34eba64d1bf0"
 		},
-		{
-			"ImportPath": "github.com/gengo/grpc-gateway/runtime",
-			"Rev": "dcb844349dc5d2cb0300fdc4d2d374839d0d2e13"
-		},
-		{
-			"ImportPath": "github.com/gengo/grpc-gateway/runtime/internal",
-			"Rev": "dcb844349dc5d2cb0300fdc4d2d374839d0d2e13"
-		},
-		{
-			"ImportPath": "github.com/gengo/grpc-gateway/utilities",
-			"Rev": "dcb844349dc5d2cb0300fdc4d2d374839d0d2e13"
-		},
 		{
 			"ImportPath": "github.com/ghodss/yaml",
 			"Rev": "73d445a93680fa1a78ae23a5839bad48f32ba1ee"
 		},
 		{
 			"ImportPath": "github.com/gogo/protobuf/proto",
-			"Comment": "v0.2-13-gc3995ae",
+			"Comment": "v0.2-33-ge18d7aa",
-			"Rev": "c3995ae437bb78d1189f4f147dfe5f87ad3596e4"
+			"Rev": "e18d7aa8f8c624c915db340349aad4c49b10d173"
 		},
 		{
 			"ImportPath": "github.com/golang/glog",
@@ -106,6 +90,21 @@
 			"ImportPath": "github.com/google/btree",
 			"Rev": "7d79101e329e5a3adf994758c578dab82b90c017"
 		},
+		{
+			"ImportPath": "github.com/grpc-ecosystem/grpc-gateway/runtime",
+			"Comment": "v1.0.0-8-gf52d055",
+			"Rev": "f52d055dc48aec25854ed7d31862f78913cf17d1"
+		},
+		{
+			"ImportPath": "github.com/grpc-ecosystem/grpc-gateway/runtime/internal",
+			"Comment": "v1.0.0-8-gf52d055",
+			"Rev": "f52d055dc48aec25854ed7d31862f78913cf17d1"
+		},
+		{
+			"ImportPath": "github.com/grpc-ecosystem/grpc-gateway/utilities",
+			"Comment": "v1.0.0-8-gf52d055",
+			"Rev": "f52d055dc48aec25854ed7d31862f78913cf17d1"
+		},
 		{
 			"ImportPath": "github.com/inconshreveable/mousetrap",
 			"Rev": "76626ae9c91c4f2a10f34cad8ce83ea42c93bb75"
@@ -234,39 +233,48 @@
 		},
 		{
 			"ImportPath": "google.golang.org/grpc",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/codes",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/credentials",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/grpclog",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/internal",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/metadata",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/naming",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/peer",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "google.golang.org/grpc/transport",
-			"Rev": "e78224b060cf3215247b7be455f80ea22e469b66"
+			"Comment": "v1.0.0-183-g231b4cf",
+			"Rev": "231b4cfea0e79843053a33f5fe90bd4d84b23cd3"
 		},
 		{
 			"ImportPath": "gopkg.in/cheggaaa/pb.v1",

cmd/vendor/github.com/akrennmair/gopcap/.gitignore

@@ -1,5 +0,0 @@
-#*
-*~
-/tools/pass/pass
-/tools/pcaptest/pcaptest
-/tools/tcpdump/tcpdump

cmd/vendor/github.com/akrennmair/gopcap/LICENSE

@@ -1,27 +0,0 @@
-Copyright (c) 2009-2011 Andreas Krennmair. All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are
-met:
-
-   * Redistributions of source code must retain the above copyright
-notice, this list of conditions and the following disclaimer.
-   * Redistributions in binary form must reproduce the above
-copyright notice, this list of conditions and the following disclaimer
-in the documentation and/or other materials provided with the
-distribution.
-   * Neither the name of Andreas Krennmair nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

cmd/vendor/github.com/akrennmair/gopcap/README.mkd

@@ -1,11 +0,0 @@
-# PCAP
-
-This is a simple wrapper around libpcap for Go.  Originally written by Andreas
-Krennmair <ak@synflood.at> and only minorly touched up by Mark Smith <mark@qq.is>.
-
-Please see the included pcaptest.go and tcpdump.go programs for instructions on
-how to use this library.
-
-Miek Gieben <miek@miek.nl> has created a more Go-like package and replaced functionality
-with standard functions from the standard library. The package has also been renamed to
-pcap.

cmd/vendor/github.com/akrennmair/gopcap/decode.go

@@ -1,527 +0,0 @@
-package pcap
-
-import (
-	"encoding/binary"
-	"fmt"
-	"net"
-	"reflect"
-	"strings"
-)
-
-const (
-	TYPE_IP   = 0x0800
-	TYPE_ARP  = 0x0806
-	TYPE_IP6  = 0x86DD
-	TYPE_VLAN = 0x8100
-
-	IP_ICMP = 1
-	IP_INIP = 4
-	IP_TCP  = 6
-	IP_UDP  = 17
-)
-
-const (
-	ERRBUF_SIZE = 256
-
-	// According to pcap-linktype(7).
-	LINKTYPE_NULL             = 0
-	LINKTYPE_ETHERNET         = 1
-	LINKTYPE_TOKEN_RING       = 6
-	LINKTYPE_ARCNET           = 7
-	LINKTYPE_SLIP             = 8
-	LINKTYPE_PPP              = 9
-	LINKTYPE_FDDI             = 10
-	LINKTYPE_ATM_RFC1483      = 100
-	LINKTYPE_RAW              = 101
-	LINKTYPE_PPP_HDLC         = 50
-	LINKTYPE_PPP_ETHER        = 51
-	LINKTYPE_C_HDLC           = 104
-	LINKTYPE_IEEE802_11       = 105
-	LINKTYPE_FRELAY           = 107
-	LINKTYPE_LOOP             = 108
-	LINKTYPE_LINUX_SLL        = 113
-	LINKTYPE_LTALK            = 104
-	LINKTYPE_PFLOG            = 117
-	LINKTYPE_PRISM_HEADER     = 119
-	LINKTYPE_IP_OVER_FC       = 122
-	LINKTYPE_SUNATM           = 123
-	LINKTYPE_IEEE802_11_RADIO = 127
-	LINKTYPE_ARCNET_LINUX     = 129
-	LINKTYPE_LINUX_IRDA       = 144
-	LINKTYPE_LINUX_LAPD       = 177
-)
-
-type addrHdr interface {
-	SrcAddr() string
-	DestAddr() string
-	Len() int
-}
-
-type addrStringer interface {
-	String(addr addrHdr) string
-}
-
-func decodemac(pkt []byte) uint64 {
-	mac := uint64(0)
-	for i := uint(0); i < 6; i++ {
-		mac = (mac << 8) + uint64(pkt[i])
-	}
-	return mac
-}
-
-// Decode decodes the headers of a Packet.
-func (p *Packet) Decode() {
-	if len(p.Data) <= 14 {
-		return
-	}
-
-	p.Type = int(binary.BigEndian.Uint16(p.Data[12:14]))
-	p.DestMac = decodemac(p.Data[0:6])
-	p.SrcMac = decodemac(p.Data[6:12])
-
-	if len(p.Data) >= 15 {
-		p.Payload = p.Data[14:]
-	}
-
-	switch p.Type {
-	case TYPE_IP:
-		p.decodeIp()
-	case TYPE_IP6:
-		p.decodeIp6()
-	case TYPE_ARP:
-		p.decodeArp()
-	case TYPE_VLAN:
-		p.decodeVlan()
-	}
-}
-
-func (p *Packet) headerString(headers []interface{}) string {
-	// If there's just one header, return that.
-	if len(headers) == 1 {
-		if hdr, ok := headers[0].(fmt.Stringer); ok {
-			return hdr.String()
-		}
-	}
-	// If there are two headers (IPv4/IPv6 -> TCP/UDP/IP..)
-	if len(headers) == 2 {
-		// Commonly the first header is an address.
-		if addr, ok := p.Headers[0].(addrHdr); ok {
-			if hdr, ok := p.Headers[1].(addrStringer); ok {
-				return fmt.Sprintf("%s %s", p.Time, hdr.String(addr))
-			}
-		}
-	}
-	// For IP in IP, we do a recursive call.
-	if len(headers) >= 2 {
-		if addr, ok := headers[0].(addrHdr); ok {
-			if _, ok := headers[1].(addrHdr); ok {
-				return fmt.Sprintf("%s > %s IP in IP: ",
-					addr.SrcAddr(), addr.DestAddr(), p.headerString(headers[1:]))
-			}
-		}
-	}
-
-	var typeNames []string
-	for _, hdr := range headers {
-		typeNames = append(typeNames, reflect.TypeOf(hdr).String())
-	}
-
-	return fmt.Sprintf("unknown [%s]", strings.Join(typeNames, ","))
-}
-
-// String prints a one-line representation of the packet header.
-// The output is suitable for use in a tcpdump program.
-func (p *Packet) String() string {
-	// If there are no headers, print "unsupported protocol".
-	if len(p.Headers) == 0 {
-		return fmt.Sprintf("%s unsupported protocol %d", p.Time, int(p.Type))
-	}
-	return fmt.Sprintf("%s %s", p.Time, p.headerString(p.Headers))
-}
-
-// Arphdr is a ARP packet header.
-type Arphdr struct {
-	Addrtype          uint16
-	Protocol          uint16
-	HwAddressSize     uint8
-	ProtAddressSize   uint8
-	Operation         uint16
-	SourceHwAddress   []byte
-	SourceProtAddress []byte
-	DestHwAddress     []byte
-	DestProtAddress   []byte
-}
-
-func (arp *Arphdr) String() (s string) {
-	switch arp.Operation {
-	case 1:
-		s = "ARP request"
-	case 2:
-		s = "ARP Reply"
-	}
-	if arp.Addrtype == LINKTYPE_ETHERNET && arp.Protocol == TYPE_IP {
-		s = fmt.Sprintf("%012x (%s) > %012x (%s)",
-			decodemac(arp.SourceHwAddress), arp.SourceProtAddress,
-			decodemac(arp.DestHwAddress), arp.DestProtAddress)
-	} else {
-		s = fmt.Sprintf("addrtype = %d protocol = %d", arp.Addrtype, arp.Protocol)
-	}
-	return
-}
-
-func (p *Packet) decodeArp() {
-	if len(p.Payload) < 8 {
-		return
-	}
-
-	pkt := p.Payload
-	arp := new(Arphdr)
-	arp.Addrtype = binary.BigEndian.Uint16(pkt[0:2])
-	arp.Protocol = binary.BigEndian.Uint16(pkt[2:4])
-	arp.HwAddressSize = pkt[4]
-	arp.ProtAddressSize = pkt[5]
-	arp.Operation = binary.BigEndian.Uint16(pkt[6:8])
-
-	if len(pkt) < int(8+2*arp.HwAddressSize+2*arp.ProtAddressSize) {
-		return
-	}
-	arp.SourceHwAddress = pkt[8 : 8+arp.HwAddressSize]
-	arp.SourceProtAddress = pkt[8+arp.HwAddressSize : 8+arp.HwAddressSize+arp.ProtAddressSize]
-	arp.DestHwAddress = pkt[8+arp.HwAddressSize+arp.ProtAddressSize : 8+2*arp.HwAddressSize+arp.ProtAddressSize]
-	arp.DestProtAddress = pkt[8+2*arp.HwAddressSize+arp.ProtAddressSize : 8+2*arp.HwAddressSize+2*arp.ProtAddressSize]
-
-	p.Headers = append(p.Headers, arp)
-
-	if len(pkt) >= int(8+2*arp.HwAddressSize+2*arp.ProtAddressSize) {
-		p.Payload = p.Payload[8+2*arp.HwAddressSize+2*arp.ProtAddressSize:]
-	}
-}
-
-// IPadr is the header of an IP packet.
-type Iphdr struct {
-	Version    uint8
-	Ihl        uint8
-	Tos        uint8
-	Length     uint16
-	Id         uint16
-	Flags      uint8
-	FragOffset uint16
-	Ttl        uint8
-	Protocol   uint8
-	Checksum   uint16
-	SrcIp      []byte
-	DestIp     []byte
-}
-
-func (p *Packet) decodeIp() {
-	if len(p.Payload) < 20 {
-		return
-	}
-
-	pkt := p.Payload
-	ip := new(Iphdr)
-
-	ip.Version = uint8(pkt[0]) >> 4
-	ip.Ihl = uint8(pkt[0]) & 0x0F
-	ip.Tos = pkt[1]
-	ip.Length = binary.BigEndian.Uint16(pkt[2:4])
-	ip.Id = binary.BigEndian.Uint16(pkt[4:6])
-	flagsfrags := binary.BigEndian.Uint16(pkt[6:8])
-	ip.Flags = uint8(flagsfrags >> 13)
-	ip.FragOffset = flagsfrags & 0x1FFF
-	ip.Ttl = pkt[8]
-	ip.Protocol = pkt[9]
-	ip.Checksum = binary.BigEndian.Uint16(pkt[10:12])
-	ip.SrcIp = pkt[12:16]
-	ip.DestIp = pkt[16:20]
-
-	pEnd := int(ip.Length)
-	if pEnd > len(pkt) {
-		pEnd = len(pkt)
-	}
-
-	if len(pkt) >= pEnd && int(ip.Ihl*4) < pEnd {
-		p.Payload = pkt[ip.Ihl*4 : pEnd]
-	} else {
-		p.Payload = []byte{}
-	}
-
-	p.Headers = append(p.Headers, ip)
-	p.IP = ip
-
-	switch ip.Protocol {
-	case IP_TCP:
-		p.decodeTcp()
-	case IP_UDP:
-		p.decodeUdp()
-	case IP_ICMP:
-		p.decodeIcmp()
-	case IP_INIP:
-		p.decodeIp()
-	}
-}
-
-func (ip *Iphdr) SrcAddr() string  { return net.IP(ip.SrcIp).String() }
-func (ip *Iphdr) DestAddr() string { return net.IP(ip.DestIp).String() }
-func (ip *Iphdr) Len() int         { return int(ip.Length) }
-
-type Vlanhdr struct {
-	Priority       byte
-	DropEligible   bool
-	VlanIdentifier int
-	Type           int // Not actually part of the vlan header, but the type of the actual packet
-}
-
-func (v *Vlanhdr) String() {
-	fmt.Sprintf("VLAN Priority:%d Drop:%v Tag:%d", v.Priority, v.DropEligible, v.VlanIdentifier)
-}
-
-func (p *Packet) decodeVlan() {
-	pkt := p.Payload
-	vlan := new(Vlanhdr)
-	if len(pkt) < 4 {
-		return
-	}
-
-	vlan.Priority = (pkt[2] & 0xE0) >> 13
-	vlan.DropEligible = pkt[2]&0x10 != 0
-	vlan.VlanIdentifier = int(binary.BigEndian.Uint16(pkt[:2])) & 0x0FFF
-	vlan.Type = int(binary.BigEndian.Uint16(p.Payload[2:4]))
-	p.Headers = append(p.Headers, vlan)
-
-	if len(pkt) >= 5 {
-		p.Payload = p.Payload[4:]
-	}
-
-	switch vlan.Type {
-	case TYPE_IP:
-		p.decodeIp()
-	case TYPE_IP6:
-		p.decodeIp6()
-	case TYPE_ARP:
-		p.decodeArp()
-	}
-}
-
-type Tcphdr struct {
-	SrcPort    uint16
-	DestPort   uint16
-	Seq        uint32
-	Ack        uint32
-	DataOffset uint8
-	Flags      uint16
-	Window     uint16
-	Checksum   uint16
-	Urgent     uint16
-	Data       []byte
-}
-
-const (
-	TCP_FIN = 1 << iota
-	TCP_SYN
-	TCP_RST
-	TCP_PSH
-	TCP_ACK
-	TCP_URG
-	TCP_ECE
-	TCP_CWR
-	TCP_NS
-)
-
-func (p *Packet) decodeTcp() {
-	if len(p.Payload) < 20 {
-		return
-	}
-
-	pkt := p.Payload
-	tcp := new(Tcphdr)
-	tcp.SrcPort = binary.BigEndian.Uint16(pkt[0:2])
-	tcp.DestPort = binary.BigEndian.Uint16(pkt[2:4])
-	tcp.Seq = binary.BigEndian.Uint32(pkt[4:8])
-	tcp.Ack = binary.BigEndian.Uint32(pkt[8:12])
-	tcp.DataOffset = (pkt[12] & 0xF0) >> 4
-	tcp.Flags = binary.BigEndian.Uint16(pkt[12:14]) & 0x1FF
-	tcp.Window = binary.BigEndian.Uint16(pkt[14:16])
-	tcp.Checksum = binary.BigEndian.Uint16(pkt[16:18])
-	tcp.Urgent = binary.BigEndian.Uint16(pkt[18:20])
-	if len(pkt) >= int(tcp.DataOffset*4) {
-		p.Payload = pkt[tcp.DataOffset*4:]
-	}
-	p.Headers = append(p.Headers, tcp)
-	p.TCP = tcp
-}
-
-func (tcp *Tcphdr) String(hdr addrHdr) string {
-	return fmt.Sprintf("TCP %s:%d > %s:%d %s SEQ=%d ACK=%d LEN=%d",
-		hdr.SrcAddr(), int(tcp.SrcPort), hdr.DestAddr(), int(tcp.DestPort),
-		tcp.FlagsString(), int64(tcp.Seq), int64(tcp.Ack), hdr.Len())
-}
-
-func (tcp *Tcphdr) FlagsString() string {
-	var sflags []string
-	if 0 != (tcp.Flags & TCP_SYN) {
-		sflags = append(sflags, "syn")
-	}
-	if 0 != (tcp.Flags & TCP_FIN) {
-		sflags = append(sflags, "fin")
-	}
-	if 0 != (tcp.Flags & TCP_ACK) {
-		sflags = append(sflags, "ack")
-	}
-	if 0 != (tcp.Flags & TCP_PSH) {
-		sflags = append(sflags, "psh")
-	}
-	if 0 != (tcp.Flags & TCP_RST) {
-		sflags = append(sflags, "rst")
-	}
-	if 0 != (tcp.Flags & TCP_URG) {
-		sflags = append(sflags, "urg")
-	}
-	if 0 != (tcp.Flags & TCP_NS) {
-		sflags = append(sflags, "ns")
-	}
-	if 0 != (tcp.Flags & TCP_CWR) {
-		sflags = append(sflags, "cwr")
-	}
-	if 0 != (tcp.Flags & TCP_ECE) {
-		sflags = append(sflags, "ece")
-	}
-	return fmt.Sprintf("[%s]", strings.Join(sflags, " "))
-}
-
-type Udphdr struct {
-	SrcPort  uint16
-	DestPort uint16
-	Length   uint16
-	Checksum uint16
-}
-
-func (p *Packet) decodeUdp() {
-	if len(p.Payload) < 8 {
-		return
-	}
-
-	pkt := p.Payload
-	udp := new(Udphdr)
-	udp.SrcPort = binary.BigEndian.Uint16(pkt[0:2])
-	udp.DestPort = binary.BigEndian.Uint16(pkt[2:4])
-	udp.Length = binary.BigEndian.Uint16(pkt[4:6])
-	udp.Checksum = binary.BigEndian.Uint16(pkt[6:8])
-	p.Headers = append(p.Headers, udp)
-	p.UDP = udp
-	if len(p.Payload) >= 8 {
-		p.Payload = pkt[8:]
-	}
-}
-
-func (udp *Udphdr) String(hdr addrHdr) string {
-	return fmt.Sprintf("UDP %s:%d > %s:%d LEN=%d CHKSUM=%d",
-		hdr.SrcAddr(), int(udp.SrcPort), hdr.DestAddr(), int(udp.DestPort),
-		int(udp.Length), int(udp.Checksum))
-}
-
-type Icmphdr struct {
-	Type     uint8
-	Code     uint8
-	Checksum uint16
-	Id       uint16
-	Seq      uint16
-	Data     []byte
-}
-
-func (p *Packet) decodeIcmp() *Icmphdr {
-	if len(p.Payload) < 8 {
-		return nil
-	}
-
-	pkt := p.Payload
-	icmp := new(Icmphdr)
-	icmp.Type = pkt[0]
-	icmp.Code = pkt[1]
-	icmp.Checksum = binary.BigEndian.Uint16(pkt[2:4])
-	icmp.Id = binary.BigEndian.Uint16(pkt[4:6])
-	icmp.Seq = binary.BigEndian.Uint16(pkt[6:8])
-	p.Payload = pkt[8:]
-	p.Headers = append(p.Headers, icmp)
-	return icmp
-}
-
-func (icmp *Icmphdr) String(hdr addrHdr) string {
-	return fmt.Sprintf("ICMP %s > %s Type = %d Code = %d ",
-		hdr.SrcAddr(), hdr.DestAddr(), icmp.Type, icmp.Code)
-}
-
-func (icmp *Icmphdr) TypeString() (result string) {
-	switch icmp.Type {
-	case 0:
-		result = fmt.Sprintf("Echo reply seq=%d", icmp.Seq)
-	case 3:
-		switch icmp.Code {
-		case 0:
-			result = "Network unreachable"
-		case 1:
-			result = "Host unreachable"
-		case 2:
-			result = "Protocol unreachable"
-		case 3:
-			result = "Port unreachable"
-		default:
-			result = "Destination unreachable"
-		}
-	case 8:
-		result = fmt.Sprintf("Echo request seq=%d", icmp.Seq)
-	case 30:
-		result = "Traceroute"
-	}
-	return
-}
-
-type Ip6hdr struct {
-	// http://www.networksorcery.com/enp/protocol/ipv6.htm
-	Version      uint8  // 4 bits
-	TrafficClass uint8  // 8 bits
-	FlowLabel    uint32 // 20 bits
-	Length       uint16 // 16 bits
-	NextHeader   uint8  // 8 bits, same as Protocol in Iphdr
-	HopLimit     uint8  // 8 bits
-	SrcIp        []byte // 16 bytes
-	DestIp       []byte // 16 bytes
-}
-
-func (p *Packet) decodeIp6() {
-	if len(p.Payload) < 40 {
-		return
-	}
-
-	pkt := p.Payload
-	ip6 := new(Ip6hdr)
-	ip6.Version = uint8(pkt[0]) >> 4
-	ip6.TrafficClass = uint8((binary.BigEndian.Uint16(pkt[0:2]) >> 4) & 0x00FF)
-	ip6.FlowLabel = binary.BigEndian.Uint32(pkt[0:4]) & 0x000FFFFF
-	ip6.Length = binary.BigEndian.Uint16(pkt[4:6])
-	ip6.NextHeader = pkt[6]
-	ip6.HopLimit = pkt[7]
-	ip6.SrcIp = pkt[8:24]
-	ip6.DestIp = pkt[24:40]
-
-	if len(p.Payload) >= 40 {
-		p.Payload = pkt[40:]
-	}
-
-	p.Headers = append(p.Headers, ip6)
-
-	switch ip6.NextHeader {
-	case IP_TCP:
-		p.decodeTcp()
-	case IP_UDP:
-		p.decodeUdp()
-	case IP_ICMP:
-		p.decodeIcmp()
-	case IP_INIP:
-		p.decodeIp()
-	}
-}
-
-func (ip6 *Ip6hdr) SrcAddr() string  { return net.IP(ip6.SrcIp).String() }
-func (ip6 *Ip6hdr) DestAddr() string { return net.IP(ip6.DestIp).String() }
-func (ip6 *Ip6hdr) Len() int         { return int(ip6.Length) }

cmd/vendor/github.com/akrennmair/gopcap/io.go

@@ -1,206 +0,0 @@
-package pcap
-
-import (
-	"encoding/binary"
-	"fmt"
-	"io"
-	"time"
-)
-
-// FileHeader is the parsed header of a pcap file.
-// http://wiki.wireshark.org/Development/LibpcapFileFormat
-type FileHeader struct {
-	MagicNumber  uint32
-	VersionMajor uint16
-	VersionMinor uint16
-	TimeZone     int32
-	SigFigs      uint32
-	SnapLen      uint32
-	Network      uint32
-}
-
-type PacketTime struct {
-	Sec  int32
-	Usec int32
-}
-
-// Convert the PacketTime to a go Time struct.
-func (p *PacketTime) Time() time.Time {
-	return time.Unix(int64(p.Sec), int64(p.Usec)*1000)
-}
-
-// Packet is a single packet parsed from a pcap file.
-//
-// Convenient access to IP, TCP, and UDP headers is provided after Decode()
-// is called if the packet is of the appropriate type.
-type Packet struct {
-	Time   time.Time // packet send/receive time
-	Caplen uint32    // bytes stored in the file (caplen <= len)
-	Len    uint32    // bytes sent/received
-	Data   []byte    // packet data
-
-	Type    int // protocol type, see LINKTYPE_*
-	DestMac uint64
-	SrcMac  uint64
-
-	Headers []interface{} // decoded headers, in order
-	Payload []byte        // remaining non-header bytes
-
-	IP  *Iphdr  // IP header (for IP packets, after decoding)
-	TCP *Tcphdr // TCP header (for TCP packets, after decoding)
-	UDP *Udphdr // UDP header (for UDP packets after decoding)
-}
-
-// Reader parses pcap files.
-type Reader struct {
-	flip         bool
-	buf          io.Reader
-	err          error
-	fourBytes    []byte
-	twoBytes     []byte
-	sixteenBytes []byte
-	Header       FileHeader
-}
-
-// NewReader reads pcap data from an io.Reader.
-func NewReader(reader io.Reader) (*Reader, error) {
-	r := &Reader{
-		buf:          reader,
-		fourBytes:    make([]byte, 4),
-		twoBytes:     make([]byte, 2),
-		sixteenBytes: make([]byte, 16),
-	}
-	switch magic := r.readUint32(); magic {
-	case 0xa1b2c3d4:
-		r.flip = false
-	case 0xd4c3b2a1:
-		r.flip = true
-	default:
-		return nil, fmt.Errorf("pcap: bad magic number: %0x", magic)
-	}
-	r.Header = FileHeader{
-		MagicNumber:  0xa1b2c3d4,
-		VersionMajor: r.readUint16(),
-		VersionMinor: r.readUint16(),
-		TimeZone:     r.readInt32(),
-		SigFigs:      r.readUint32(),
-		SnapLen:      r.readUint32(),
-		Network:      r.readUint32(),
-	}
-	return r, nil
-}
-
-// Next returns the next packet or nil if no more packets can be read.
-func (r *Reader) Next() *Packet {
-	d := r.sixteenBytes
-	r.err = r.read(d)
-	if r.err != nil {
-		return nil
-	}
-	timeSec := asUint32(d[0:4], r.flip)
-	timeUsec := asUint32(d[4:8], r.flip)
-	capLen := asUint32(d[8:12], r.flip)
-	origLen := asUint32(d[12:16], r.flip)
-
-	data := make([]byte, capLen)
-	if r.err = r.read(data); r.err != nil {
-		return nil
-	}
-	return &Packet{
-		Time:   time.Unix(int64(timeSec), int64(timeUsec)),
-		Caplen: capLen,
-		Len:    origLen,
-		Data:   data,
-	}
-}
-
-func (r *Reader) read(data []byte) error {
-	var err error
-	n, err := r.buf.Read(data)
-	for err == nil && n != len(data) {
-		var chunk int
-		chunk, err = r.buf.Read(data[n:])
-		n += chunk
-	}
-	if len(data) == n {
-		return nil
-	}
-	return err
-}
-
-func (r *Reader) readUint32() uint32 {
-	data := r.fourBytes
-	if r.err = r.read(data); r.err != nil {
-		return 0
-	}
-	return asUint32(data, r.flip)
-}
-
-func (r *Reader) readInt32() int32 {
-	data := r.fourBytes
-	if r.err = r.read(data); r.err != nil {
-		return 0
-	}
-	return int32(asUint32(data, r.flip))
-}
-
-func (r *Reader) readUint16() uint16 {
-	data := r.twoBytes
-	if r.err = r.read(data); r.err != nil {
-		return 0
-	}
-	return asUint16(data, r.flip)
-}
-
-// Writer writes a pcap file.
-type Writer struct {
-	writer io.Writer
-	buf    []byte
-}
-
-// NewWriter creates a Writer that stores output in an io.Writer.
-// The FileHeader is written immediately.
-func NewWriter(writer io.Writer, header *FileHeader) (*Writer, error) {
-	w := &Writer{
-		writer: writer,
-		buf:    make([]byte, 24),
-	}
-	binary.LittleEndian.PutUint32(w.buf, header.MagicNumber)
-	binary.LittleEndian.PutUint16(w.buf[4:], header.VersionMajor)
-	binary.LittleEndian.PutUint16(w.buf[6:], header.VersionMinor)
-	binary.LittleEndian.PutUint32(w.buf[8:], uint32(header.TimeZone))
-	binary.LittleEndian.PutUint32(w.buf[12:], header.SigFigs)
-	binary.LittleEndian.PutUint32(w.buf[16:], header.SnapLen)
-	binary.LittleEndian.PutUint32(w.buf[20:], header.Network)
-	if _, err := writer.Write(w.buf); err != nil {
-		return nil, err
-	}
-	return w, nil
-}
-
-// Writer writes a packet to the underlying writer.
-func (w *Writer) Write(pkt *Packet) error {
-	binary.LittleEndian.PutUint32(w.buf, uint32(pkt.Time.Unix()))
-	binary.LittleEndian.PutUint32(w.buf[4:], uint32(pkt.Time.Nanosecond()))
-	binary.LittleEndian.PutUint32(w.buf[8:], uint32(pkt.Time.Unix()))
-	binary.LittleEndian.PutUint32(w.buf[12:], pkt.Len)
-	if _, err := w.writer.Write(w.buf[:16]); err != nil {
-		return err
-	}
-	_, err := w.writer.Write(pkt.Data)
-	return err
-}
-
-func asUint32(data []byte, flip bool) uint32 {
-	if flip {
-		return binary.BigEndian.Uint32(data)
-	}
-	return binary.LittleEndian.Uint32(data)
-}
-
-func asUint16(data []byte, flip bool) uint16 {
-	if flip {
-		return binary.BigEndian.Uint16(data)
-	}
-	return binary.LittleEndian.Uint16(data)
-}

cmd/vendor/github.com/akrennmair/gopcap/pcap.go

@@ -1,266 +0,0 @@
-// Interface to both live and offline pcap parsing.
-package pcap
-
-/*
-#cgo linux LDFLAGS: -lpcap
-#cgo freebsd LDFLAGS: -lpcap
-#cgo darwin LDFLAGS: -lpcap
-#cgo windows CFLAGS: -I C:/WpdPack/Include
-#cgo windows,386 LDFLAGS: -L C:/WpdPack/Lib -lwpcap
-#cgo windows,amd64 LDFLAGS: -L C:/WpdPack/Lib/x64 -lwpcap
-#include <stdlib.h>
-#include <pcap.h>
-
-// Workaround for not knowing how to cast to const u_char**
-int hack_pcap_next_ex(pcap_t *p, struct pcap_pkthdr **pkt_header,
-                      u_char **pkt_data) {
-    return pcap_next_ex(p, pkt_header, (const u_char **)pkt_data);
-}
-*/
-import "C"
-import (
-	"errors"
-	"net"
-	"syscall"
-	"time"
-	"unsafe"
-)
-
-type Pcap struct {
-	cptr *C.pcap_t
-}
-
-type Stat struct {
-	PacketsReceived  uint32
-	PacketsDropped   uint32
-	PacketsIfDropped uint32
-}
-
-type Interface struct {
-	Name        string
-	Description string
-	Addresses   []IFAddress
-	// TODO: add more elements
-}
-
-type IFAddress struct {
-	IP      net.IP
-	Netmask net.IPMask
-	// TODO: add broadcast + PtP dst ?
-}
-
-func (p *Pcap) Next() (pkt *Packet) {
-	rv, _ := p.NextEx()
-	return rv
-}
-
-// Openlive opens a device and returns a *Pcap handler
-func Openlive(device string, snaplen int32, promisc bool, timeout_ms int32) (handle *Pcap, err error) {
-	var buf *C.char
-	buf = (*C.char)(C.calloc(ERRBUF_SIZE, 1))
-	h := new(Pcap)
-	var pro int32
-	if promisc {
-		pro = 1
-	}
-
-	dev := C.CString(device)
-	defer C.free(unsafe.Pointer(dev))
-
-	h.cptr = C.pcap_open_live(dev, C.int(snaplen), C.int(pro), C.int(timeout_ms), buf)
-	if nil == h.cptr {
-		handle = nil
-		err = errors.New(C.GoString(buf))
-	} else {
-		handle = h
-	}
-	C.free(unsafe.Pointer(buf))
-	return
-}
-
-func Openoffline(file string) (handle *Pcap, err error) {
-	var buf *C.char
-	buf = (*C.char)(C.calloc(ERRBUF_SIZE, 1))
-	h := new(Pcap)
-
-	cf := C.CString(file)
-	defer C.free(unsafe.Pointer(cf))
-
-	h.cptr = C.pcap_open_offline(cf, buf)
-	if nil == h.cptr {
-		handle = nil
-		err = errors.New(C.GoString(buf))
-	} else {
-		handle = h
-	}
-	C.free(unsafe.Pointer(buf))
-	return
-}
-
-func (p *Pcap) NextEx() (pkt *Packet, result int32) {
-	var pkthdr *C.struct_pcap_pkthdr
-
-	var buf_ptr *C.u_char
-	var buf unsafe.Pointer
-	result = int32(C.hack_pcap_next_ex(p.cptr, &pkthdr, &buf_ptr))
-
-	buf = unsafe.Pointer(buf_ptr)
-	if nil == buf {
-		return
-	}
-
-	pkt = new(Packet)
-	pkt.Time = time.Unix(int64(pkthdr.ts.tv_sec), int64(pkthdr.ts.tv_usec)*1000)
-	pkt.Caplen = uint32(pkthdr.caplen)
-	pkt.Len = uint32(pkthdr.len)
-	pkt.Data = C.GoBytes(buf, C.int(pkthdr.caplen))
-	return
-}
-
-func (p *Pcap) Close() {
-	C.pcap_close(p.cptr)
-}
-
-func (p *Pcap) Geterror() error {
-	return errors.New(C.GoString(C.pcap_geterr(p.cptr)))
-}
-
-func (p *Pcap) Getstats() (stat *Stat, err error) {
-	var cstats _Ctype_struct_pcap_stat
-	if -1 == C.pcap_stats(p.cptr, &cstats) {
-		return nil, p.Geterror()
-	}
-	stats := new(Stat)
-	stats.PacketsReceived = uint32(cstats.ps_recv)
-	stats.PacketsDropped = uint32(cstats.ps_drop)
-	stats.PacketsIfDropped = uint32(cstats.ps_ifdrop)
-
-	return stats, nil
-}
-
-func (p *Pcap) Setfilter(expr string) (err error) {
-	var bpf _Ctype_struct_bpf_program
-	cexpr := C.CString(expr)
-	defer C.free(unsafe.Pointer(cexpr))
-
-	if -1 == C.pcap_compile(p.cptr, &bpf, cexpr, 1, 0) {
-		return p.Geterror()
-	}
-
-	if -1 == C.pcap_setfilter(p.cptr, &bpf) {
-		C.pcap_freecode(&bpf)
-		return p.Geterror()
-	}
-	C.pcap_freecode(&bpf)
-	return nil
-}
-
-func Version() string {
-	return C.GoString(C.pcap_lib_version())
-}
-
-func (p *Pcap) Datalink() int {
-	return int(C.pcap_datalink(p.cptr))
-}
-
-func (p *Pcap) Setdatalink(dlt int) error {
-	if -1 == C.pcap_set_datalink(p.cptr, C.int(dlt)) {
-		return p.Geterror()
-	}
-	return nil
-}
-
-func DatalinkValueToName(dlt int) string {
-	if name := C.pcap_datalink_val_to_name(C.int(dlt)); name != nil {
-		return C.GoString(name)
-	}
-	return ""
-}
-
-func DatalinkValueToDescription(dlt int) string {
-	if desc := C.pcap_datalink_val_to_description(C.int(dlt)); desc != nil {
-		return C.GoString(desc)
-	}
-	return ""
-}
-
-func Findalldevs() (ifs []Interface, err error) {
-	var buf *C.char
-	buf = (*C.char)(C.calloc(ERRBUF_SIZE, 1))
-	defer C.free(unsafe.Pointer(buf))
-	var alldevsp *C.pcap_if_t
-
-	if -1 == C.pcap_findalldevs((**C.pcap_if_t)(&alldevsp), buf) {
-		return nil, errors.New(C.GoString(buf))
-	}
-	defer C.pcap_freealldevs((*C.pcap_if_t)(alldevsp))
-	dev := alldevsp
-	var i uint32
-	for i = 0; dev != nil; dev = (*C.pcap_if_t)(dev.next) {
-		i++
-	}
-	ifs = make([]Interface, i)
-	dev = alldevsp
-	for j := uint32(0); dev != nil; dev = (*C.pcap_if_t)(dev.next) {
-		var iface Interface
-		iface.Name = C.GoString(dev.name)
-		iface.Description = C.GoString(dev.description)
-		iface.Addresses = findalladdresses(dev.addresses)
-		// TODO: add more elements
-		ifs[j] = iface
-		j++
-	}
-	return
-}
-
-func findalladdresses(addresses *_Ctype_struct_pcap_addr) (retval []IFAddress) {
-	// TODO - make it support more than IPv4 and IPv6?
-	retval = make([]IFAddress, 0, 1)
-	for curaddr := addresses; curaddr != nil; curaddr = (*_Ctype_struct_pcap_addr)(curaddr.next) {
-		var a IFAddress
-		var err error
-		if a.IP, err = sockaddr_to_IP((*syscall.RawSockaddr)(unsafe.Pointer(curaddr.addr))); err != nil {
-			continue
-		}
-		if a.Netmask, err = sockaddr_to_IP((*syscall.RawSockaddr)(unsafe.Pointer(curaddr.addr))); err != nil {
-			continue
-		}
-		retval = append(retval, a)
-	}
-	return
-}
-
-func sockaddr_to_IP(rsa *syscall.RawSockaddr) (IP []byte, err error) {
-	switch rsa.Family {
-	case syscall.AF_INET:
-		pp := (*syscall.RawSockaddrInet4)(unsafe.Pointer(rsa))
-		IP = make([]byte, 4)
-		for i := 0; i < len(IP); i++ {
-			IP[i] = pp.Addr[i]
-		}
-		return
-	case syscall.AF_INET6:
-		pp := (*syscall.RawSockaddrInet6)(unsafe.Pointer(rsa))
-		IP = make([]byte, 16)
-		for i := 0; i < len(IP); i++ {
-			IP[i] = pp.Addr[i]
-		}
-		return
-	}
-	err = errors.New("Unsupported address type")
-	return
-}
-
-func (p *Pcap) Inject(data []byte) (err error) {
-	buf := (*C.char)(C.malloc((C.size_t)(len(data))))
-
-	for i := 0; i < len(data); i++ {
-		*(*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(buf)) + uintptr(i))) = data[i]
-	}
-
-	if -1 == C.pcap_sendpacket(p.cptr, (*C.u_char)(unsafe.Pointer(buf)), (C.int)(len(data))) {
-		err = p.Geterror()
-	}
-	C.free(unsafe.Pointer(buf))
-	return
-}

cmd/vendor/github.com/boltdb/bolt/README.md

@@ -1,4 +1,4 @@
-Bolt [![Coverage Status](https://coveralls.io/repos/boltdb/bolt/badge.svg?branch=master)](https://coveralls.io/r/boltdb/bolt?branch=master) [![GoDoc](https://godoc.org/github.com/boltdb/bolt?status.svg)](https://godoc.org/github.com/boltdb/bolt) ![Version](https://img.shields.io/badge/version-1.0-green.svg)
+Bolt [![Coverage Status](https://coveralls.io/repos/boltdb/bolt/badge.svg?branch=master)](https://coveralls.io/r/boltdb/bolt?branch=master) [![GoDoc](https://godoc.org/github.com/boltdb/bolt?status.svg)](https://godoc.org/github.com/boltdb/bolt) ![Version](https://img.shields.io/badge/version-1.2.1-green.svg)
 ====
 
 Bolt is a pure Go key/value store inspired by [Howard Chu's][hyc_symas]
@@ -313,7 +313,7 @@ func (s *Store) CreateUser(u *User) error {
         // Generate ID for the user.
         // This returns an error only if the Tx is closed or not writeable.
         // That can't happen in an Update() call so I ignore the error check.
-        id, _ = b.NextSequence()
+        id, _ := b.NextSequence()
         u.ID = int(id)
 
         // Marshal user data into bytes.
@@ -557,7 +557,7 @@ if err != nil {
 Bolt is able to run on mobile devices by leveraging the binding feature of the
 [gomobile](https://github.com/golang/mobile) tool. Create a struct that will
 contain your database logic and a reference to a `*bolt.DB` with a initializing
-contstructor that takes in a filepath where the database file will be stored.
+constructor that takes in a filepath where the database file will be stored.
 Neither Android nor iOS require extra permissions or cleanup from using this method.
 
 ```go
@@ -807,6 +807,7 @@ them via pull request.
 
 Below is a list of public, open source projects that use Bolt:
 
+* [BoltDbWeb](https://github.com/evnix/boltdbweb) - A web based GUI for BoltDB files.
 * [Operation Go: A Routine Mission](http://gocode.io) - An online programming game for Golang using Bolt for user accounts and a leaderboard.
 * [Bazil](https://bazil.org/) - A file system that lets your data reside where it is most convenient for it to reside.
 * [DVID](https://github.com/janelia-flyem/dvid) - Added Bolt as optional storage engine and testing it against Basho-tuned leveldb.
@@ -825,7 +826,6 @@ Below is a list of public, open source projects that use Bolt:
 * [cayley](https://github.com/google/cayley) - Cayley is an open-source graph database using Bolt as optional backend.
 * [bleve](http://www.blevesearch.com/) - A pure Go search engine similar to ElasticSearch that uses Bolt as the default storage backend.
 * [tentacool](https://github.com/optiflows/tentacool) - REST api server to manage system stuff (IP, DNS, Gateway...) on a linux server.
-* [SkyDB](https://github.com/skydb/sky) - Behavioral analytics database.
 * [Seaweed File System](https://github.com/chrislusf/seaweedfs) - Highly scalable distributed key~file system with O(1) disk read.
 * [InfluxDB](https://influxdata.com) - Scalable datastore for metrics, events, and real-time analytics.
 * [Freehold](http://tshannon.bitbucket.org/freehold/) - An open, secure, and lightweight platform for your files and data.
@@ -842,9 +842,11 @@ Below is a list of public, open source projects that use Bolt:
 * [Go Report Card](https://goreportcard.com/) - Go code quality report cards as a (free and open source) service.
 * [Boltdb Boilerplate](https://github.com/bobintornado/boltdb-boilerplate) - Boilerplate wrapper around bolt aiming to make simple calls one-liners.
 * [lru](https://github.com/crowdriff/lru) - Easy to use Bolt-backed Least-Recently-Used (LRU) read-through cache with chainable remote stores.
-* [Storm](https://github.com/asdine/storm) - A simple ORM around BoltDB.
+* [Storm](https://github.com/asdine/storm) - Simple and powerful ORM for BoltDB.
 * [GoWebApp](https://github.com/josephspurrier/gowebapp) - A basic MVC web application in Go using BoltDB.
 * [SimpleBolt](https://github.com/xyproto/simplebolt) - A simple way to use BoltDB. Deals mainly with strings.
 * [Algernon](https://github.com/xyproto/algernon) - A HTTP/2 web server with built-in support for Lua. Uses BoltDB as the default database backend.
+* [MuLiFS](https://github.com/dankomiocevic/mulifs) - Music Library Filesystem creates a filesystem to organise your music files.
+* [GoShort](https://github.com/pankajkhairnar/goShort) - GoShort is a URL shortener written in Golang and BoltDB for persistent key/value storage and for routing it's using high performent HTTPRouter.
 
 If you are using Bolt in a project please send a pull request to add it to the list.

cmd/vendor/github.com/boltdb/bolt/freelist.go

@@ -166,12 +166,16 @@ func (f *freelist) read(p *page) {
 	}
 
 	// Copy the list of page ids from the freelist.
-	ids := ((*[maxAllocSize]pgid)(unsafe.Pointer(&p.ptr)))[idx:count]
+	if count == 0 {
-	f.ids = make([]pgid, len(ids))
+		f.ids = nil
-	copy(f.ids, ids)
+	} else {
+		ids := ((*[maxAllocSize]pgid)(unsafe.Pointer(&p.ptr)))[idx:count]
+		f.ids = make([]pgid, len(ids))
+		copy(f.ids, ids)
 
-	// Make sure they're sorted.
+		// Make sure they're sorted.
-	sort.Sort(pgids(f.ids))
+		sort.Sort(pgids(f.ids))
+	}
 
 	// Rebuild the page cache.
 	f.reindex()
@@ -189,7 +193,9 @@ func (f *freelist) write(p *page) error {
 
 	// The page.count can only hold up to 64k elements so if we overflow that
 	// number then we handle it by putting the size in the first element.
-	if len(ids) < 0xFFFF {
+	if len(ids) == 0 {
+		p.count = uint16(len(ids))
+	} else if len(ids) < 0xFFFF {
 		p.count = uint16(len(ids))
 		copy(((*[maxAllocSize]pgid)(unsafe.Pointer(&p.ptr)))[:], ids)
 	} else {

cmd/vendor/github.com/boltdb/bolt/node.go

@@ -201,6 +201,11 @@ func (n *node) write(p *page) {
 	}
 	p.count = uint16(len(n.inodes))
 
+	// Stop here if there are no items to write.
+	if p.count == 0 {
+		return
+	}
+
 	// Loop over each item and write it to the page.
 	b := (*[maxAllocSize]byte)(unsafe.Pointer(&p.ptr))[n.pageElementSize()*len(n.inodes):]
 	for i, item := range n.inodes {

cmd/vendor/github.com/boltdb/bolt/page.go

@@ -62,6 +62,9 @@ func (p *page) leafPageElement(index uint16) *leafPageElement {
 
 // leafPageElements retrieves a list of leaf nodes.
 func (p *page) leafPageElements() []leafPageElement {
+	if p.count == 0 {
+		return nil
+	}
 	return ((*[0x7FFFFFF]leafPageElement)(unsafe.Pointer(&p.ptr)))[:]
 }
 
@@ -72,6 +75,9 @@ func (p *page) branchPageElement(index uint16) *branchPageElement {
 
 // branchPageElements retrieves a list of branch nodes.
 func (p *page) branchPageElements() []branchPageElement {
+	if p.count == 0 {
+		return nil
+	}
 	return ((*[0x7FFFFFF]branchPageElement)(unsafe.Pointer(&p.ptr)))[:]
 }
 

cmd/vendor/github.com/coreos/go-systemd/daemon/sdnotify.go

@@ -2,30 +2,37 @@
 package daemon
 
 import (
-	"errors"
 	"net"
 	"os"
 )
 
-var SdNotifyNoSocket = errors.New("No socket")
-
 // SdNotify sends a message to the init daemon. It is common to ignore the error.
-func SdNotify(state string) error {
+// It returns one of the following:
+// (false, nil) - notification not supported (i.e. NOTIFY_SOCKET is unset)
+// (false, err) - notification supported, but failure happened (e.g. error connecting to NOTIFY_SOCKET or while sending data)
+// (true, nil) - notification supported, data has been sent
+func SdNotify(state string) (sent bool, err error) {
 	socketAddr := &net.UnixAddr{
 		Name: os.Getenv("NOTIFY_SOCKET"),
 		Net:  "unixgram",
 	}
 
+	// NOTIFY_SOCKET not set
 	if socketAddr.Name == "" {
-		return SdNotifyNoSocket
+		return false, nil
 	}
 
 	conn, err := net.DialUnix(socketAddr.Net, nil, socketAddr)
+	// Error connecting to NOTIFY_SOCKET
 	if err != nil {
-		return err
+		return false, err
 	}
 	defer conn.Close()
 
 	_, err = conn.Write([]byte(state))
-	return err
+	// Error sending the message
+	if err != nil {
+		return false, err
+	}
+	return true, nil
 }

cmd/vendor/github.com/coreos/go-systemd/journal/journal.go

@@ -12,7 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// Package journal provides write bindings to the systemd journal
+// Package journal provides write bindings to the local systemd journal.
+// It is implemented in pure Go and connects to the journal directly over its
+// unix socket.
+//
+// To read from the journal, see the "sdjournal" package, which wraps the
+// sd-journal a C API.
+//
+// http://www.freedesktop.org/software/systemd/man/systemd-journald.service.html
 package journal
 
 import (
@@ -53,14 +60,14 @@ func init() {
 	}
 }
 
-// Enabled returns true iff the systemd journal is available for logging
+// Enabled returns true if the local systemd journal is available for logging
 func Enabled() bool {
 	return conn != nil
 }
 
-// Send a message to the systemd journal. vars is a map of journald fields to
+// Send a message to the local systemd journal. vars is a map of journald
-// values.  Fields must be composed of uppercase letters, numbers, and
+// fields to values.  Fields must be composed of uppercase letters, numbers,
-// underscores, but must not start with an underscore. Within these
+// and underscores, but must not start with an underscore. Within these
 // restrictions, any arbitrary field name may be used.  Some names have special
 // significance: see the journalctl documentation
 // (http://www.freedesktop.org/software/systemd/man/systemd.journal-fields.html)
@@ -83,6 +90,7 @@ func Send(message string, priority Priority, vars map[string]string) error {
 		if err != nil {
 			return journalError(err.Error())
 		}
+		defer file.Close()
 		_, err = io.Copy(file, data)
 		if err != nil {
 			return journalError(err.Error())
@@ -102,6 +110,11 @@ func Send(message string, priority Priority, vars map[string]string) error {
 	return nil
 }
 
+// Print prints a message to the local systemd journal using Send().
+func Print(priority Priority, format string, a ...interface{}) error {
+	return Send(fmt.Sprintf(format, a...), priority, nil)
+}
+
 func appendVariable(w io.Writer, name, value string) {
 	if !validVarName(name) {
 		journalError("variable name contains invalid character, ignoring")

cmd/vendor/github.com/coreos/go-systemd/util/util.go

@@ -13,15 +13,66 @@
 // limitations under the License.
 
 // Package util contains utility functions related to systemd that applications
-// can use to check things like whether systemd is running.
+// can use to check things like whether systemd is running.  Note that some of
+// these functions attempt to manually load systemd libraries at runtime rather
+// than linking against them.
 package util
 
+// #include <stdlib.h>
+// #include <sys/types.h>
+// #include <unistd.h>
+//
+// int
+// my_sd_pid_get_owner_uid(void *f, pid_t pid, uid_t *uid)
+// {
+//   int (*sd_pid_get_owner_uid)(pid_t, uid_t *);
+//
+//   sd_pid_get_owner_uid = (int (*)(pid_t, uid_t *))f;
+//   return sd_pid_get_owner_uid(pid, uid);
+// }
+//
+// int
+// my_sd_pid_get_unit(void *f, pid_t pid, char **unit)
+// {
+//   int (*sd_pid_get_unit)(pid_t, char **);
+//
+//   sd_pid_get_unit = (int (*)(pid_t, char **))f;
+//   return sd_pid_get_unit(pid, unit);
+// }
+//
+// int
+// my_sd_pid_get_slice(void *f, pid_t pid, char **slice)
+// {
+//   int (*sd_pid_get_slice)(pid_t, char **);
+//
+//   sd_pid_get_slice = (int (*)(pid_t, char **))f;
+//   return sd_pid_get_slice(pid, slice);
+// }
+//
+// int
+// am_session_leader()
+// {
+//   return (getsid(0) == getpid());
+// }
 import (
+	"fmt"
+	"io/ioutil"
 	"os"
+	"strings"
 )
 
+var libsystemdNames = []string{
+	// systemd < 209
+	"libsystemd-login.so.0",
+	"libsystemd-login.so",
+
+	// systemd >= 209 merged libsystemd-login into libsystemd proper
+	"libsystemd.so.0",
+	"libsystemd.so",
+}
+
 // IsRunningSystemd checks whether the host was booted with systemd as its init
-// system. This functions similar to systemd's `sd_booted(3)`: internally, it
+// system. This functions similarly to systemd's `sd_booted(3)`: internally, it
 // checks whether /run/systemd/system/ exists and is a directory.
 // http://www.freedesktop.org/software/systemd/man/sd_booted.html
 func IsRunningSystemd() bool {
@@ -31,3 +82,15 @@ func IsRunningSystemd() bool {
 	}
 	return fi.IsDir()
 }
+
+// GetMachineID returns a host's 128-bit machine ID as a string. This functions
+// similarly to systemd's `sd_id128_get_machine`: internally, it simply reads
+// the contents of /etc/machine-id
+// http://www.freedesktop.org/software/systemd/man/sd_id128_get_machine.html
+func GetMachineID() (string, error) {
+	machineID, err := ioutil.ReadFile("/etc/machine-id")
+	if err != nil {
+		return "", fmt.Errorf("failed to read /etc/machine-id: %v", err)
+	}
+	return strings.TrimSpace(string(machineID)), nil
+}

cmd/vendor/github.com/gogo/protobuf/proto/decode.go

@@ -773,10 +773,11 @@ func (o *Buffer) dec_new_map(p *Properties, base structPointer) error {
 		}
 	}
 	keyelem, valelem := keyptr.Elem(), valptr.Elem()
-	if !keyelem.IsValid() || !valelem.IsValid() {
+	if !keyelem.IsValid() {
-		// We did not decode the key or the value in the map entry.
+		keyelem = reflect.Zero(p.mtype.Key())
-		// Either way, it's an invalid map entry.
+	}
-		return fmt.Errorf("proto: bad map data: missing key/val")
+	if !valelem.IsValid() {
+		valelem = reflect.Zero(p.mtype.Elem())
 	}
 
 	v.SetMapIndex(keyelem, valelem)

cmd/vendor/github.com/gogo/protobuf/proto/encode.go

@@ -64,6 +64,10 @@ var (
 	// a struct with a repeated field containing a nil element.
 	errRepeatedHasNil = errors.New("proto: repeated field has nil element")
 
+	// errOneofHasNil is the error returned if Marshal is called with
+	// a struct with a oneof field containing a nil element.
+	errOneofHasNil = errors.New("proto: oneof field has nil value")
+
 	// ErrNil is the error returned if Marshal is called with nil.
 	ErrNil = errors.New("proto: Marshal called with nil")
 )
@@ -1222,7 +1226,9 @@ func (o *Buffer) enc_struct(prop *StructProperties, base structPointer) error {
 	// Do oneof fields.
 	if prop.oneofMarshaler != nil {
 		m := structPointer_Interface(base, prop.stype).(Message)
-		if err := prop.oneofMarshaler(m, o); err != nil {
+		if err := prop.oneofMarshaler(m, o); err == ErrNil {
+			return errOneofHasNil
+		} else if err != nil {
 			return err
 		}
 	}

cmd/vendor/github.com/gogo/protobuf/proto/extensions_gogo.go

@@ -56,6 +56,10 @@ func (this *Extension) Equal(that *Extension) bool {
 	return bytes.Equal(this.enc, that.enc)
 }
 
+func (this *Extension) Compare(that *Extension) int {
+	return bytes.Compare(this.enc, that.enc)
+}
+
 func SizeOfExtensionMap(m map[int32]Extension) (n int) {
 	return sizeExtensionMap(m)
 }

cmd/vendor/github.com/gogo/protobuf/proto/text.go

@@ -335,7 +335,8 @@ func writeStruct(w *textWriter, sv reflect.Value) error {
 				}
 				inner := fv.Elem().Elem() // interface -> *T -> T
 				tag := inner.Type().Field(0).Tag.Get("protobuf")
-				props.Parse(tag) // Overwrite the outer props.
+				props = new(Properties) // Overwrite the outer props var, but not its pointee.
+				props.Parse(tag)
 				// Write the value in the oneof, not the oneof itself.
 				fv = inner.Field(0)
 
@@ -727,7 +728,14 @@ func (w *textWriter) writeIndent() {
 	w.complete = false
 }
 
-func marshalText(w io.Writer, pb Message, compact bool) error {
+// TextMarshaler is a configurable text format marshaler.
+type TextMarshaler struct {
+	Compact bool // use compact text format (one line).
+}
+
+// Marshal writes a given protocol buffer in text format.
+// The only errors returned are from w.
+func (m *TextMarshaler) Marshal(w io.Writer, pb Message) error {
 	val := reflect.ValueOf(pb)
 	if pb == nil || val.IsNil() {
 		w.Write([]byte("<nil>"))
@@ -742,7 +750,7 @@ func marshalText(w io.Writer, pb Message, compact bool) error {
 	aw := &textWriter{
 		w:        ww,
 		complete: true,
-		compact:  compact,
+		compact:  m.Compact,
 	}
 
 	if tm, ok := pb.(encoding.TextMarshaler); ok {
@@ -769,25 +777,29 @@ func marshalText(w io.Writer, pb Message, compact bool) error {
 	return nil
 }
 
+// Text is the same as Marshal, but returns the string directly.
+func (m *TextMarshaler) Text(pb Message) string {
+	var buf bytes.Buffer
+	m.Marshal(&buf, pb)
+	return buf.String()
+}
+
+var (
+	defaultTextMarshaler = TextMarshaler{}
+	compactTextMarshaler = TextMarshaler{Compact: true}
+)
+
+// TODO: consider removing some of the Marshal functions below.
+
 // MarshalText writes a given protocol buffer in text format.
 // The only errors returned are from w.
-func MarshalText(w io.Writer, pb Message) error {
+func MarshalText(w io.Writer, pb Message) error { return defaultTextMarshaler.Marshal(w, pb) }
-	return marshalText(w, pb, false)
-}
 
 // MarshalTextString is the same as MarshalText, but returns the string directly.
-func MarshalTextString(pb Message) string {
+func MarshalTextString(pb Message) string { return defaultTextMarshaler.Text(pb) }
-	var buf bytes.Buffer
-	marshalText(&buf, pb, false)
-	return buf.String()
-}
 
 // CompactText writes a given protocol buffer in compact text format (one line).
-func CompactText(w io.Writer, pb Message) error { return marshalText(w, pb, true) }
+func CompactText(w io.Writer, pb Message) error { return compactTextMarshaler.Marshal(w, pb) }
 
 // CompactTextString is the same as CompactText, but returns the string directly.
-func CompactTextString(pb Message) string {
+func CompactTextString(pb Message) string { return compactTextMarshaler.Text(pb) }
-	var buf bytes.Buffer
-	marshalText(&buf, pb, true)
-	return buf.String()
-}

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/errors.go

@@ -104,6 +104,7 @@ func DefaultHTTPError(ctx context.Context, marshaler Marshaler, w http.ResponseW
 	}
 
 	handleForwardResponseServerMetadata(w, md)
+	handleForwardResponseTrailerHeader(w, md)
 	st := HTTPStatusFromCode(grpc.Code(err))
 	w.WriteHeader(st)
 	if _, err := w.Write(buf); err != nil {

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/handler.go

@@ -6,8 +6,8 @@ import (
 	"net/http"
 	"net/textproto"
 
-	"github.com/gengo/grpc-gateway/runtime/internal"
 	"github.com/golang/protobuf/proto"
+	"github.com/grpc-ecosystem/grpc-gateway/runtime/internal"
 	"golang.org/x/net/context"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/grpclog"

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/internal/stream_chunk.pb.go

@@ -43,23 +43,23 @@ func (*StreamError) ProtoMessage()               {}
 func (*StreamError) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{0} }
 
 func init() {
-	proto.RegisterType((*StreamError)(nil), "gengo.grpc.gateway.runtime.StreamError")
+	proto.RegisterType((*StreamError)(nil), "grpc.gateway.runtime.StreamError")
 }
 
 func init() { proto.RegisterFile("runtime/internal/stream_chunk.proto", fileDescriptor0) }
 
 var fileDescriptor0 = []byte{
-	// 182 bytes of a gzipped FileDescriptorProto
+	// 180 bytes of a gzipped FileDescriptorProto
-	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0x34, 0x8e, 0x3d, 0xef, 0x82, 0x30,
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0xe2, 0x52, 0x2e, 0x2a, 0xcd, 0x2b,
-	0x10, 0x87, 0xc3, 0xff, 0x15, 0x8e, 0x8d, 0xa9, 0xd1, 0x41, 0xa3, 0x8b, 0x53, 0x19, 0xfc, 0x06,
+	0xc9, 0xcc, 0x4d, 0xd5, 0xcf, 0xcc, 0x2b, 0x49, 0x2d, 0xca, 0x4b, 0xcc, 0xd1, 0x2f, 0x2e, 0x29,
-	0x1a, 0xbf, 0x00, 0x6c, 0x2e, 0xa4, 0xe2, 0xa5, 0x10, 0xa5, 0x25, 0xd7, 0x23, 0xc6, 0xd5, 0x4f,
+	0x4a, 0x4d, 0xcc, 0x8d, 0x4f, 0xce, 0x28, 0xcd, 0xcb, 0xd6, 0x2b, 0x28, 0xca, 0x2f, 0xc9, 0x17,
-	0x2e, 0x2d, 0x32, 0xde, 0xf3, 0xdc, 0x93, 0xfc, 0x60, 0x4b, 0x83, 0xe1, 0xb6, 0xc3, 0xbc, 0x35,
+	0x12, 0x49, 0x2f, 0x2a, 0x48, 0xd6, 0x4b, 0x4f, 0x2c, 0x49, 0x2d, 0x4f, 0xac, 0xd4, 0x83, 0xea,
-	0x8c, 0x64, 0xd4, 0x3d, 0x77, 0x4c, 0xa8, 0xba, 0xaa, 0x6e, 0x06, 0x73, 0x93, 0x3d, 0x59, 0xb6,
+	0x50, 0x6a, 0x62, 0xe4, 0xe2, 0x0e, 0x06, 0x2b, 0x76, 0x2d, 0x2a, 0xca, 0x2f, 0x12, 0x92, 0xe6,
-	0xd9, 0x42, 0xa3, 0xd1, 0x56, 0x6a, 0xea, 0x6b, 0xa9, 0x15, 0xe3, 0x43, 0x3d, 0xe5, 0xa7, 0xdb,
+	0xe2, 0x04, 0xa9, 0x8b, 0x4f, 0xce, 0x4f, 0x49, 0x95, 0x60, 0x54, 0x60, 0xd4, 0x60, 0x0d, 0xe2,
-	0xbc, 0x22, 0x48, 0xcb, 0x90, 0x9c, 0x88, 0x2c, 0x65, 0x4b, 0x48, 0xfc, 0x5f, 0x55, 0xdb, 0x2b,
+	0x00, 0x09, 0x38, 0x03, 0xf9, 0x20, 0xc9, 0x8c, 0x92, 0x92, 0x02, 0x88, 0x24, 0x13, 0x44, 0x12,
-	0x8a, 0x68, 0x1d, 0xed, 0x7e, 0x8b, 0xd8, 0x83, 0xe3, 0x78, 0x7b, 0xd9, 0x30, 0xf7, 0x93, 0xfc,
+	0x24, 0x00, 0x96, 0x94, 0xe0, 0x62, 0xcf, 0x4d, 0x2d, 0x2e, 0x4e, 0x4c, 0x4f, 0x95, 0x60, 0x06,
-	0x9a, 0xa4, 0x07, 0x41, 0x0a, 0xf8, 0xef, 0xd0, 0x39, 0xa5, 0x51, 0x7c, 0x8f, 0x2a, 0x29, 0xe6,
+	0x4a, 0x71, 0x06, 0xc1, 0xb8, 0x42, 0xf2, 0x5c, 0xdc, 0x60, 0x6d, 0xc5, 0x25, 0x89, 0x25, 0xa5,
-	0x33, 0x5b, 0x41, 0x1a, 0x32, 0xc7, 0x8a, 0x07, 0x27, 0x7e, 0x82, 0x05, 0x8f, 0xca, 0x40, 0x0e,
+	0xc5, 0x12, 0x2c, 0x60, 0x59, 0x2e, 0x90, 0x50, 0x30, 0x58, 0xc4, 0x89, 0x2b, 0x8a, 0x03, 0xe6,
-	0x70, 0x8e, 0xe7, 0xfd, 0x97, 0xbf, 0xb0, 0x79, 0xff, 0x0e, 0x00, 0x00, 0xff, 0xff, 0xf8, 0x7f,
+	0xf2, 0x24, 0x36, 0xb0, 0x6b, 0x8d, 0x01, 0x01, 0x00, 0x00, 0xff, 0xff, 0xa9, 0x07, 0x92, 0xb6,
-	0x7d, 0x56, 0xda, 0x00, 0x00, 0x00,
+	0xd4, 0x00, 0x00, 0x00,
 }

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/internal/stream_chunk.proto

@@ -1,5 +1,5 @@
 syntax = "proto3";
-package gengo.grpc.gateway.runtime;
+package grpc.gateway.runtime;
 option go_package = "internal";
 
 // StreamError is a response type which is returned when

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/pattern.go

@@ -5,7 +5,7 @@ import (
 	"fmt"
 	"strings"
 
-	"github.com/gengo/grpc-gateway/utilities"
+	"github.com/grpc-ecosystem/grpc-gateway/utilities"
 	"google.golang.org/grpc/grpclog"
 )
 

cmd/vendor/github.com/grpc-ecosystem/grpc-gateway/runtime/query.go

@@ -6,8 +6,8 @@ import (
 	"reflect"
 	"strings"
 
-	"github.com/gengo/grpc-gateway/utilities"
 	"github.com/golang/protobuf/proto"
+	"github.com/grpc-ecosystem/grpc-gateway/utilities"
 	"google.golang.org/grpc/grpclog"
 )
 

cmd/vendor/google.golang.org/grpc/.travis.yml

@@ -1,17 +1,21 @@
 language: go
 
 go:
-  - 1.5.3
+  - 1.5.4
-  - 1.6
+  - 1.6.3
+
+go_import_path: google.golang.org/grpc
 
 before_install:
+  - go get golang.org/x/tools/cmd/goimports
+  - go get github.com/golang/lint/golint
   - go get github.com/axw/gocov/gocov
   - go get github.com/mattn/goveralls
   - go get golang.org/x/tools/cmd/cover
 
-install:
-  - mkdir -p "$GOPATH/src/google.golang.org"
-  - mv "$TRAVIS_BUILD_DIR" "$GOPATH/src/google.golang.org/grpc"
-
 script:
+  - '! gofmt -s -d -l . 2>&1 | read'
+  - '! goimports -l . | read'
+  - '! golint ./... | grep -vE "(_string|\.pb)\.go:"'
+  - '! go tool vet -all . 2>&1 | grep -vE "constant [0-9]+ not a string in call to Errorf"'
   - make test testrace

cmd/vendor/google.golang.org/grpc/balancer.go

@@ -40,7 +40,6 @@ import (
 	"golang.org/x/net/context"
 	"google.golang.org/grpc/grpclog"
 	"google.golang.org/grpc/naming"
-	"google.golang.org/grpc/transport"
 )
 
 // Address represents a server the client connects to.
@@ -94,10 +93,10 @@ type Balancer interface {
 	// instead of blocking.
 	//
 	// The function returns put which is called once the rpc has completed or failed.
-	// put can collect and report RPC stats to a remote load balancer. gRPC internals
+	// put can collect and report RPC stats to a remote load balancer.
-	// will try to call this again if err is non-nil (unless err is ErrClientConnClosing).
 	//
-	// TODO: Add other non-recoverable errors?
+	// This function should only return the errors Balancer cannot recover by itself.
+	// gRPC internals will fail the RPC if an error is returned.
 	Get(ctx context.Context, opts BalancerGetOptions) (addr Address, put func(), err error)
 	// Notify returns a channel that is used by gRPC internals to watch the addresses
 	// gRPC needs to connect. The addresses might be from a name resolver or remote
@@ -139,35 +138,40 @@ func RoundRobin(r naming.Resolver) Balancer {
 	return &roundRobin{r: r}
 }
 
+type addrInfo struct {
+	addr      Address
+	connected bool
+}
+
 type roundRobin struct {
-	r         naming.Resolver
+	r      naming.Resolver
-	w         naming.Watcher
+	w      naming.Watcher
-	open      []Address // all the addresses the client should potentially connect
+	addrs  []*addrInfo // all the addresses the client should potentially connect
-	mu        sync.Mutex
+	mu     sync.Mutex
-	addrCh    chan []Address // the channel to notify gRPC internals the list of addresses the client should connect to.
+	addrCh chan []Address // the channel to notify gRPC internals the list of addresses the client should connect to.
-	connected []Address      // all the connected addresses
+	next   int            // index of the next address to return for Get()
-	next      int            // index of the next address to return for Get()
+	waitCh chan struct{}  // the channel to block when there is no connected address available
-	waitCh    chan struct{}  // the channel to block when there is no connected address available
+	done   bool           // The Balancer is closed.
-	done      bool           // The Balancer is closed.
 }
 
 func (rr *roundRobin) watchAddrUpdates() error {
 	updates, err := rr.w.Next()
 	if err != nil {
-		grpclog.Println("grpc: the naming watcher stops working due to %v.", err)
+		grpclog.Printf("grpc: the naming watcher stops working due to %v.\n", err)
 		return err
 	}
 	rr.mu.Lock()
 	defer rr.mu.Unlock()
 	for _, update := range updates {
 		addr := Address{
-			Addr: update.Addr,
+			Addr:     update.Addr,
+			Metadata: update.Metadata,
 		}
 		switch update.Op {
 		case naming.Add:
 			var exist bool
-			for _, v := range rr.open {
+			for _, v := range rr.addrs {
-				if addr == v {
+				if addr == v.addr {
 					exist = true
 					grpclog.Println("grpc: The name resolver wanted to add an existing address: ", addr)
 					break
@@ -176,12 +180,12 @@ func (rr *roundRobin) watchAddrUpdates() error {
 			if exist {
 				continue
 			}
-			rr.open = append(rr.open, addr)
+			rr.addrs = append(rr.addrs, &addrInfo{addr: addr})
 		case naming.Delete:
-			for i, v := range rr.open {
+			for i, v := range rr.addrs {
-				if v == addr {
+				if addr == v.addr {
-					copy(rr.open[i:], rr.open[i+1:])
+					copy(rr.addrs[i:], rr.addrs[i+1:])
-					rr.open = rr.open[:len(rr.open)-1]
+					rr.addrs = rr.addrs[:len(rr.addrs)-1]
 					break
 				}
 			}
@@ -189,9 +193,11 @@ func (rr *roundRobin) watchAddrUpdates() error {
 			grpclog.Println("Unknown update.Op ", update.Op)
 		}
 	}
-	// Make a copy of rr.open and write it onto rr.addrCh so that gRPC internals gets notified.
+	// Make a copy of rr.addrs and write it onto rr.addrCh so that gRPC internals gets notified.
-	open := make([]Address, len(rr.open), len(rr.open))
+	open := make([]Address, len(rr.addrs))
-	copy(open, rr.open)
+	for i, v := range rr.addrs {
+		open[i] = v.addr
+	}
 	if rr.done {
 		return ErrClientConnClosing
 	}
@@ -202,7 +208,9 @@ func (rr *roundRobin) watchAddrUpdates() error {
 func (rr *roundRobin) Start(target string) error {
 	if rr.r == nil {
 		// If there is no name resolver installed, it is not needed to
-		// do name resolution. In this case, rr.addrCh stays nil.
+		// do name resolution. In this case, target is added into rr.addrs
+		// as the only address available and rr.addrCh stays nil.
+		rr.addrs = append(rr.addrs, &addrInfo{addr: Address{Addr: target}})
 		return nil
 	}
 	w, err := rr.r.Resolve(target)
@@ -221,38 +229,41 @@ func (rr *roundRobin) Start(target string) error {
 	return nil
 }
 
-// Up appends addr to the end of rr.connected and sends notification if there
+// Up sets the connected state of addr and sends notification if there are pending
-// are pending Get() calls.
+// Get() calls.
 func (rr *roundRobin) Up(addr Address) func(error) {
 	rr.mu.Lock()
 	defer rr.mu.Unlock()
-	for _, a := range rr.connected {
+	var cnt int
-		if a == addr {
+	for _, a := range rr.addrs {
-			return nil
+		if a.addr == addr {
+			if a.connected {
+				return nil
+			}
+			a.connected = true
+		}
+		if a.connected {
+			cnt++
 		}
 	}
-	rr.connected = append(rr.connected, addr)
+	// addr is only one which is connected. Notify the Get() callers who are blocking.
-	if len(rr.connected) == 1 {
+	if cnt == 1 && rr.waitCh != nil {
-		// addr is only one available. Notify the Get() callers who are blocking.
+		close(rr.waitCh)
-		if rr.waitCh != nil {
+		rr.waitCh = nil
-			close(rr.waitCh)
-			rr.waitCh = nil
-		}
 	}
 	return func(err error) {
 		rr.down(addr, err)
 	}
 }
 
-// down removes addr from rr.connected and moves the remaining addrs forward.
+// down unsets the connected state of addr.
 func (rr *roundRobin) down(addr Address, err error) {
 	rr.mu.Lock()
 	defer rr.mu.Unlock()
-	for i, a := range rr.connected {
+	for _, a := range rr.addrs {
-		if a == addr {
+		if addr == a.addr {
-			copy(rr.connected[i:], rr.connected[i+1:])
+			a.connected = false
-			rr.connected = rr.connected[:len(rr.connected)-1]
+			break
-			return
 		}
 	}
 }
@@ -266,17 +277,40 @@ func (rr *roundRobin) Get(ctx context.Context, opts BalancerGetOptions) (addr Ad
 		err = ErrClientConnClosing
 		return
 	}
-	if rr.next >= len(rr.connected) {
+
-		rr.next = 0
+	if len(rr.addrs) > 0 {
+		if rr.next >= len(rr.addrs) {
+			rr.next = 0
+		}
+		next := rr.next
+		for {
+			a := rr.addrs[next]
+			next = (next + 1) % len(rr.addrs)
+			if a.connected {
+				addr = a.addr
+				rr.next = next
+				rr.mu.Unlock()
+				return
+			}
+			if next == rr.next {
+				// Has iterated all the possible address but none is connected.
+				break
+			}
+		}
 	}
-	if len(rr.connected) > 0 {
+	if !opts.BlockingWait {
-		addr = rr.connected[rr.next]
+		if len(rr.addrs) == 0 {
+			rr.mu.Unlock()
+			err = fmt.Errorf("there is no address available")
+			return
+		}
+		// Returns the next addr on rr.addrs for failfast RPCs.
+		addr = rr.addrs[rr.next].addr
 		rr.next++
 		rr.mu.Unlock()
 		return
 	}
-	// There is no address available. Wait on rr.waitCh.
+	// Wait on rr.waitCh for non-failfast RPCs.
-	// TODO(zhaoq): Handle the case when opts.BlockingWait is false.
 	if rr.waitCh == nil {
 		ch = make(chan struct{})
 		rr.waitCh = ch
@@ -287,7 +321,7 @@ func (rr *roundRobin) Get(ctx context.Context, opts BalancerGetOptions) (addr Ad
 	for {
 		select {
 		case <-ctx.Done():
-			err = transport.ContextErr(ctx.Err())
+			err = ctx.Err()
 			return
 		case <-ch:
 			rr.mu.Lock()
@@ -296,24 +330,35 @@ func (rr *roundRobin) Get(ctx context.Context, opts BalancerGetOptions) (addr Ad
 				err = ErrClientConnClosing
 				return
 			}
-			if len(rr.connected) == 0 {
+
-				// The newly added addr got removed by Down() again.
+			if len(rr.addrs) > 0 {
-				if rr.waitCh == nil {
+				if rr.next >= len(rr.addrs) {
-					ch = make(chan struct{})
+					rr.next = 0
-					rr.waitCh = ch
+				}
-				} else {
+				next := rr.next
-					ch = rr.waitCh
+				for {
+					a := rr.addrs[next]
+					next = (next + 1) % len(rr.addrs)
+					if a.connected {
+						addr = a.addr
+						rr.next = next
+						rr.mu.Unlock()
+						return
+					}
+					if next == rr.next {
+						// Has iterated all the possible address but none is connected.
+						break
+					}
 				}
-				rr.mu.Unlock()
-				continue
 			}
-			if rr.next >= len(rr.connected) {
+			// The newly added addr got removed by Down() again.
-				rr.next = 0
+			if rr.waitCh == nil {
+				ch = make(chan struct{})
+				rr.waitCh = ch
+			} else {
+				ch = rr.waitCh
 			}
-			addr = rr.connected[rr.next]
-			rr.next++
 			rr.mu.Unlock()
-			return
 		}
 	}
 }

cmd/vendor/google.golang.org/grpc/call.go

@@ -36,6 +36,7 @@ package grpc
 import (
 	"bytes"
 	"io"
+	"math"
 	"time"
 
 	"golang.org/x/net/context"
@@ -51,13 +52,20 @@ import (
 func recvResponse(dopts dialOptions, t transport.ClientTransport, c *callInfo, stream *transport.Stream, reply interface{}) error {
 	// Try to acquire header metadata from the server if there is any.
 	var err error
+	defer func() {
+		if err != nil {
+			if _, ok := err.(transport.ConnectionError); !ok {
+				t.CloseStream(stream, err)
+			}
+		}
+	}()
 	c.headerMD, err = stream.Header()
 	if err != nil {
 		return err
 	}
 	p := &parser{r: stream}
 	for {
-		if err = recv(p, dopts.codec, stream, dopts.dc, reply); err != nil {
+		if err = recv(p, dopts.codec, stream, dopts.dc, reply, math.MaxInt32); err != nil {
 			if err == io.EOF {
 				break
 			}
@@ -76,6 +84,7 @@ func sendRequest(ctx context.Context, codec Codec, compressor Compressor, callHd
 	}
 	defer func() {
 		if err != nil {
+			// If err is connection error, t will be closed, no need to close stream here.
 			if _, ok := err.(transport.ConnectionError); !ok {
 				t.CloseStream(stream, err)
 			}
@@ -90,7 +99,10 @@ func sendRequest(ctx context.Context, codec Codec, compressor Compressor, callHd
 		return nil, transport.StreamErrorf(codes.Internal, "grpc: %v", err)
 	}
 	err = t.Write(stream, outBuf, opts)
-	if err != nil {
+	// t.NewStream(...) could lead to an early rejection of the RPC (e.g., the service/method
+	// does not exist.) so that t.Write could get io.EOF from wait(...). Leave the following
+	// recvResponse to get the final status.
+	if err != nil && err != io.EOF {
 		return nil, err
 	}
 	// Sent successfully.
@@ -101,7 +113,7 @@ func sendRequest(ctx context.Context, codec Codec, compressor Compressor, callHd
 // Invoke is called by generated code. Also users can call Invoke directly when it
 // is really needed in their use cases.
 func Invoke(ctx context.Context, method string, args, reply interface{}, cc *ClientConn, opts ...CallOption) (err error) {
-	var c callInfo
+	c := defaultCallInfo
 	for _, o := range opts {
 		if err := o.before(&c); err != nil {
 			return toRPCErr(err)
@@ -155,19 +167,17 @@ func Invoke(ctx context.Context, method string, args, reply interface{}, cc *Cli
 		t, put, err = cc.getTransport(ctx, gopts)
 		if err != nil {
 			// TODO(zhaoq): Probably revisit the error handling.
-			if err == ErrClientConnClosing {
+			if _, ok := err.(*rpcError); ok {
-				return Errorf(codes.FailedPrecondition, "%v", err)
+				return err
 			}
-			if _, ok := err.(transport.StreamError); ok {
+			if err == errConnClosing || err == errConnUnavailable {
-				return toRPCErr(err)
-			}
-			if _, ok := err.(transport.ConnectionError); ok {
 				if c.failFast {
-					return toRPCErr(err)
+					return Errorf(codes.Unavailable, "%v", err)
 				}
+				continue
 			}
-			// All the remaining cases are treated as retryable.
+			// All the other errors are treated as Internal errors.
-			continue
+			return Errorf(codes.Internal, "%v", err)
 		}
 		if c.traceInfo.tr != nil {
 			c.traceInfo.tr.LazyLog(&payload{sent: true, msg: args}, true)
@@ -178,7 +188,10 @@ func Invoke(ctx context.Context, method string, args, reply interface{}, cc *Cli
 				put()
 				put = nil
 			}
-			if _, ok := err.(transport.ConnectionError); ok {
+			// Retry a non-failfast RPC when
+			// i) there is a connection error; or
+			// ii) the server started to drain before this RPC was initiated.
+			if _, ok := err.(transport.ConnectionError); ok || err == transport.ErrStreamDrain {
 				if c.failFast {
 					return toRPCErr(err)
 				}
@@ -186,20 +199,18 @@ func Invoke(ctx context.Context, method string, args, reply interface{}, cc *Cli
 			}
 			return toRPCErr(err)
 		}
-		// Receive the response
 		err = recvResponse(cc.dopts, t, &c, stream, reply)
 		if err != nil {
 			if put != nil {
 				put()
 				put = nil
 			}
-			if _, ok := err.(transport.ConnectionError); ok {
+			if _, ok := err.(transport.ConnectionError); ok || err == transport.ErrStreamDrain {
 				if c.failFast {
 					return toRPCErr(err)
 				}
 				continue
 			}
-			t.CloseStream(stream, err)
 			return toRPCErr(err)
 		}
 		if c.traceInfo.tr != nil {

cmd/vendor/google.golang.org/grpc/clientconn.go

@@ -43,7 +43,6 @@ import (
 
 	"golang.org/x/net/context"
 	"golang.org/x/net/trace"
-	"google.golang.org/grpc/codes"
 	"google.golang.org/grpc/credentials"
 	"google.golang.org/grpc/grpclog"
 	"google.golang.org/grpc/transport"
@@ -68,13 +67,15 @@ var (
 	// errCredentialsConflict indicates that grpc.WithTransportCredentials()
 	// and grpc.WithInsecure() are both called for a connection.
 	errCredentialsConflict = errors.New("grpc: transport credentials are set for an insecure connection (grpc.WithTransportCredentials() and grpc.WithInsecure() are both called)")
-	// errNetworkIP indicates that the connection is down due to some network I/O error.
+	// errNetworkIO indicates that the connection is down due to some network I/O error.
 	errNetworkIO = errors.New("grpc: failed with network I/O error")
 	// errConnDrain indicates that the connection starts to be drained and does not accept any new RPCs.
 	errConnDrain = errors.New("grpc: the connection is drained")
 	// errConnClosing indicates that the connection is closing.
 	errConnClosing = errors.New("grpc: the connection is closing")
-	errNoAddr      = errors.New("grpc: there is no address available to dial")
+	// errConnUnavailable indicates that the connection is unavailable.
+	errConnUnavailable = errors.New("grpc: the connection is unavailable")
+	errNoAddr          = errors.New("grpc: there is no address available to dial")
 	// minimum time to give a connection to complete
 	minConnectTimeout = 20 * time.Second
 )
@@ -196,9 +197,14 @@ func WithTimeout(d time.Duration) DialOption {
 }
 
 // WithDialer returns a DialOption that specifies a function to use for dialing network addresses.
-func WithDialer(f func(addr string, timeout time.Duration) (net.Conn, error)) DialOption {
+func WithDialer(f func(string, time.Duration) (net.Conn, error)) DialOption {
 	return func(o *dialOptions) {
-		o.copts.Dialer = f
+		o.copts.Dialer = func(ctx context.Context, addr string) (net.Conn, error) {
+			if deadline, ok := ctx.Deadline(); ok {
+				return f(addr, deadline.Sub(time.Now()))
+			}
+			return f(addr, 0)
+		}
 	}
 }
 
@@ -209,49 +215,57 @@ func WithUserAgent(s string) DialOption {
 	}
 }
 
-// Dial creates a client connection the given target.
+// Dial creates a client connection to the given target.
 func Dial(target string, opts ...DialOption) (*ClientConn, error) {
+	return DialContext(context.Background(), target, opts...)
+}
+
+// DialContext creates a client connection to the given target
+// using the supplied context.
+func DialContext(ctx context.Context, target string, opts ...DialOption) (*ClientConn, error) {
 	cc := &ClientConn{
 		target: target,
 		conns:  make(map[Address]*addrConn),
 	}
+	cc.ctx, cc.cancel = context.WithCancel(ctx)
 	for _, opt := range opts {
 		opt(&cc.dopts)
 	}
+
+	// Set defaults.
 	if cc.dopts.codec == nil {
-		// Set the default codec.
 		cc.dopts.codec = protoCodec{}
 	}
-
 	if cc.dopts.bs == nil {
 		cc.dopts.bs = DefaultBackoffConfig
 	}
 
-	cc.balancer = cc.dopts.balancer
-	if cc.balancer == nil {
-		cc.balancer = RoundRobin(nil)
-	}
-	if err := cc.balancer.Start(target); err != nil {
-		return nil, err
-	}
 	var (
 		ok    bool
 		addrs []Address
 	)
-	ch := cc.balancer.Notify()
+	if cc.dopts.balancer == nil {
-	if ch == nil {
+		// Connect to target directly if balancer is nil.
-		// There is no name resolver installed.
 		addrs = append(addrs, Address{Addr: target})
 	} else {
-		addrs, ok = <-ch
+		if err := cc.dopts.balancer.Start(target); err != nil {
-		if !ok || len(addrs) == 0 {
+			return nil, err
-			return nil, errNoAddr
+		}
+		ch := cc.dopts.balancer.Notify()
+		if ch == nil {
+			// There is no name resolver installed.
+			addrs = append(addrs, Address{Addr: target})
+		} else {
+			addrs, ok = <-ch
+			if !ok || len(addrs) == 0 {
+				return nil, errNoAddr
+			}
 		}
 	}
 	waitC := make(chan error, 1)
 	go func() {
 		for _, a := range addrs {
-			if err := cc.newAddrConn(a, false); err != nil {
+			if err := cc.resetAddrConn(a, false, nil); err != nil {
 				waitC <- err
 				return
 			}
@@ -268,10 +282,15 @@ func Dial(target string, opts ...DialOption) (*ClientConn, error) {
 			cc.Close()
 			return nil, err
 		}
+	case <-cc.ctx.Done():
+		cc.Close()
+		return nil, cc.ctx.Err()
 	case <-timeoutCh:
 		cc.Close()
 		return nil, ErrClientConnTimeout
 	}
+	// If balancer is nil or balancer.Notify() is nil, ok will be false here.
+	// The lbWatcher goroutine will not be created.
 	if ok {
 		go cc.lbWatcher()
 	}
@@ -318,8 +337,10 @@ func (s ConnectivityState) String() string {
 
 // ClientConn represents a client connection to an RPC server.
 type ClientConn struct {
+	ctx    context.Context
+	cancel context.CancelFunc
+
 	target    string
-	balancer  Balancer
 	authority string
 	dopts     dialOptions
 
@@ -328,7 +349,7 @@ type ClientConn struct {
 }
 
 func (cc *ClientConn) lbWatcher() {
-	for addrs := range cc.balancer.Notify() {
+	for addrs := range cc.dopts.balancer.Notify() {
 		var (
 			add []Address   // Addresses need to setup connections.
 			del []*addrConn // Connections need to tear down.
@@ -349,11 +370,12 @@ func (cc *ClientConn) lbWatcher() {
 			}
 			if !keep {
 				del = append(del, c)
+				delete(cc.conns, c.addr)
 			}
 		}
 		cc.mu.Unlock()
 		for _, a := range add {
-			cc.newAddrConn(a, true)
+			cc.resetAddrConn(a, true, nil)
 		}
 		for _, c := range del {
 			c.tearDown(errConnDrain)
@@ -361,13 +383,17 @@ func (cc *ClientConn) lbWatcher() {
 	}
 }
 
-func (cc *ClientConn) newAddrConn(addr Address, skipWait bool) error {
+// resetAddrConn creates an addrConn for addr and adds it to cc.conns.
+// If there is an old addrConn for addr, it will be torn down, using tearDownErr as the reason.
+// If tearDownErr is nil, errConnDrain will be used instead.
+func (cc *ClientConn) resetAddrConn(addr Address, skipWait bool, tearDownErr error) error {
 	ac := &addrConn{
-		cc:           cc,
+		cc:    cc,
-		addr:         addr,
+		addr:  addr,
-		dopts:        cc.dopts,
+		dopts: cc.dopts,
-		shutdownChan: make(chan struct{}),
 	}
+	ac.ctx, ac.cancel = context.WithCancel(cc.ctx)
+	ac.stateCV = sync.NewCond(&ac.mu)
 	if EnableTracing {
 		ac.events = trace.NewEventLog("grpc.ClientConn", ac.addr.Addr)
 	}
@@ -385,26 +411,44 @@ func (cc *ClientConn) newAddrConn(addr Address, skipWait bool) error {
 			}
 		}
 	}
-	// Insert ac into ac.cc.conns. This needs to be done before any getTransport(...) is called.
+	// Track ac in cc. This needs to be done before any getTransport(...) is called.
-	ac.cc.mu.Lock()
+	cc.mu.Lock()
-	if ac.cc.conns == nil {
+	if cc.conns == nil {
-		ac.cc.mu.Unlock()
+		cc.mu.Unlock()
 		return ErrClientConnClosing
 	}
-	stale := ac.cc.conns[ac.addr]
+	stale := cc.conns[ac.addr]
-	ac.cc.conns[ac.addr] = ac
+	cc.conns[ac.addr] = ac
-	ac.cc.mu.Unlock()
+	cc.mu.Unlock()
 	if stale != nil {
 		// There is an addrConn alive on ac.addr already. This could be due to
-		// i) stale's Close is undergoing;
+		// 1) a buggy Balancer notifies duplicated Addresses;
-		// ii) a buggy Balancer notifies duplicated Addresses.
+		// 2) goaway was received, a new ac will replace the old ac.
-		stale.tearDown(errConnDrain)
+		//    The old ac should be deleted from cc.conns, but the
+		//    underlying transport should drain rather than close.
+		if tearDownErr == nil {
+			// tearDownErr is nil if resetAddrConn is called by
+			// 1) Dial
+			// 2) lbWatcher
+			// In both cases, the stale ac should drain, not close.
+			stale.tearDown(errConnDrain)
+		} else {
+			stale.tearDown(tearDownErr)
+		}
 	}
-	ac.stateCV = sync.NewCond(&ac.mu)
 	// skipWait may overwrite the decision in ac.dopts.block.
 	if ac.dopts.block && !skipWait {
 		if err := ac.resetTransport(false); err != nil {
-			ac.tearDown(err)
+			if err != errConnClosing {
+				// Tear down ac and delete it from cc.conns.
+				cc.mu.Lock()
+				delete(cc.conns, ac.addr)
+				cc.mu.Unlock()
+				ac.tearDown(err)
+			}
+			if e, ok := err.(transport.ConnectionError); ok && !e.Temporary() {
+				return e.Origin()
+			}
 			return err
 		}
 		// Start to monitor the error status of transport.
@@ -414,7 +458,10 @@ func (cc *ClientConn) newAddrConn(addr Address, skipWait bool) error {
 		go func() {
 			if err := ac.resetTransport(false); err != nil {
 				grpclog.Printf("Failed to dial %s: %v; please retry.", ac.addr.Addr, err)
-				ac.tearDown(err)
+				if err != errConnClosing {
+					// Keep this ac in cc.conns, to get the reason it's torn down.
+					ac.tearDown(err)
+				}
 				return
 			}
 			ac.transportMonitor()
@@ -424,25 +471,48 @@ func (cc *ClientConn) newAddrConn(addr Address, skipWait bool) error {
 }
 
 func (cc *ClientConn) getTransport(ctx context.Context, opts BalancerGetOptions) (transport.ClientTransport, func(), error) {
-	// TODO(zhaoq): Implement fail-fast logic.
+	var (
-	addr, put, err := cc.balancer.Get(ctx, opts)
+		ac  *addrConn
-	if err != nil {
+		ok  bool
-		return nil, nil, err
+		put func()
-	}
+	)
-	cc.mu.RLock()
+	if cc.dopts.balancer == nil {
-	if cc.conns == nil {
+		// If balancer is nil, there should be only one addrConn available.
+		cc.mu.RLock()
+		if cc.conns == nil {
+			cc.mu.RUnlock()
+			return nil, nil, toRPCErr(ErrClientConnClosing)
+		}
+		for _, ac = range cc.conns {
+			// Break after the first iteration to get the first addrConn.
+			ok = true
+			break
+		}
+		cc.mu.RUnlock()
+	} else {
+		var (
+			addr Address
+			err  error
+		)
+		addr, put, err = cc.dopts.balancer.Get(ctx, opts)
+		if err != nil {
+			return nil, nil, toRPCErr(err)
+		}
+		cc.mu.RLock()
+		if cc.conns == nil {
+			cc.mu.RUnlock()
+			return nil, nil, toRPCErr(ErrClientConnClosing)
+		}
+		ac, ok = cc.conns[addr]
 		cc.mu.RUnlock()
-		return nil, nil, ErrClientConnClosing
 	}
-	ac, ok := cc.conns[addr]
-	cc.mu.RUnlock()
 	if !ok {
 		if put != nil {
 			put()
 		}
-		return nil, nil, transport.StreamErrorf(codes.Internal, "grpc: failed to find the transport to send the rpc")
+		return nil, nil, errConnClosing
 	}
-	t, err := ac.wait(ctx)
+	t, err := ac.wait(ctx, cc.dopts.balancer != nil, !opts.BlockingWait)
 	if err != nil {
 		if put != nil {
 			put()
@@ -454,6 +524,8 @@ func (cc *ClientConn) getTransport(ctx context.Context, opts BalancerGetOptions)
 
 // Close tears down the ClientConn and all underlying connections.
 func (cc *ClientConn) Close() error {
+	cc.cancel()
+
 	cc.mu.Lock()
 	if cc.conns == nil {
 		cc.mu.Unlock()
@@ -462,7 +534,9 @@ func (cc *ClientConn) Close() error {
 	conns := cc.conns
 	cc.conns = nil
 	cc.mu.Unlock()
-	cc.balancer.Close()
+	if cc.dopts.balancer != nil {
+		cc.dopts.balancer.Close()
+	}
 	for _, ac := range conns {
 		ac.tearDown(ErrClientConnClosing)
 	}
@@ -471,11 +545,13 @@ func (cc *ClientConn) Close() error {
 
 // addrConn is a network connection to a given address.
 type addrConn struct {
-	cc           *ClientConn
+	ctx    context.Context
-	addr         Address
+	cancel context.CancelFunc
-	dopts        dialOptions
+
-	shutdownChan chan struct{}
+	cc     *ClientConn
-	events       trace.EventLog
+	addr   Address
+	dopts  dialOptions
+	events trace.EventLog
 
 	mu      sync.Mutex
 	state   ConnectivityState
@@ -485,6 +561,9 @@ type addrConn struct {
 	// due to timeout.
 	ready     chan struct{}
 	transport transport.ClientTransport
+
+	// The reason this addrConn is torn down.
+	tearDownErr error
 }
 
 // printf records an event in ac's event log, unless ac has been closed.
@@ -540,8 +619,7 @@ func (ac *addrConn) waitForStateChange(ctx context.Context, sourceState Connecti
 }
 
 func (ac *addrConn) resetTransport(closeTransport bool) error {
-	var retries int
+	for retries := 0; ; retries++ {
-	for {
 		ac.mu.Lock()
 		ac.printf("connecting")
 		if ac.state == Shutdown {
@@ -561,13 +639,20 @@ func (ac *addrConn) resetTransport(closeTransport bool) error {
 			t.Close()
 		}
 		sleepTime := ac.dopts.bs.backoff(retries)
-		ac.dopts.copts.Timeout = sleepTime
+		timeout := minConnectTimeout
-		if sleepTime < minConnectTimeout {
+		if timeout < sleepTime {
-			ac.dopts.copts.Timeout = minConnectTimeout
+			timeout = sleepTime
 		}
+		ctx, cancel := context.WithTimeout(ac.ctx, timeout)
 		connectTime := time.Now()
-		newTransport, err := transport.NewClientTransport(ac.addr.Addr, &ac.dopts.copts)
+		newTransport, err := transport.NewClientTransport(ctx, ac.addr.Addr, ac.dopts.copts)
 		if err != nil {
+			cancel()
+
+			if e, ok := err.(transport.ConnectionError); ok && !e.Temporary() {
+				return err
+			}
+			grpclog.Printf("grpc: addrConn.resetTransport failed to create client transport: %v; Reconnecting to %q", err, ac.addr)
 			ac.mu.Lock()
 			if ac.state == Shutdown {
 				// ac.tearDown(...) has been invoked.
@@ -582,17 +667,12 @@ func (ac *addrConn) resetTransport(closeTransport bool) error {
 				ac.ready = nil
 			}
 			ac.mu.Unlock()
-			sleepTime -= time.Since(connectTime)
-			if sleepTime < 0 {
-				sleepTime = 0
-			}
 			closeTransport = false
 			select {
-			case <-time.After(sleepTime):
+			case <-time.After(sleepTime - time.Since(connectTime)):
-			case <-ac.shutdownChan:
+			case <-ac.ctx.Done():
+				return ac.ctx.Err()
 			}
-			retries++
-			grpclog.Printf("grpc: addrConn.resetTransport failed to create client transport: %v; Reconnecting to %q", err, ac.addr)
 			continue
 		}
 		ac.mu.Lock()
@@ -610,7 +690,9 @@ func (ac *addrConn) resetTransport(closeTransport bool) error {
 			close(ac.ready)
 			ac.ready = nil
 		}
-		ac.down = ac.cc.balancer.Up(ac.addr)
+		if ac.cc.dopts.balancer != nil {
+			ac.down = ac.cc.dopts.balancer.Up(ac.addr)
+		}
 		ac.mu.Unlock()
 		return nil
 	}
@@ -624,14 +706,42 @@ func (ac *addrConn) transportMonitor() {
 		t := ac.transport
 		ac.mu.Unlock()
 		select {
-		// shutdownChan is needed to detect the teardown when
+		// This is needed to detect the teardown when
 		// the addrConn is idle (i.e., no RPC in flight).
-		case <-ac.shutdownChan:
+		case <-ac.ctx.Done():
+			select {
+			case <-t.Error():
+				t.Close()
+			default:
+			}
+			return
+		case <-t.GoAway():
+			// If GoAway happens without any network I/O error, ac is closed without shutting down the
+			// underlying transport (the transport will be closed when all the pending RPCs finished or
+			// failed.).
+			// If GoAway and some network I/O error happen concurrently, ac and its underlying transport
+			// are closed.
+			// In both cases, a new ac is created.
+			select {
+			case <-t.Error():
+				ac.cc.resetAddrConn(ac.addr, true, errNetworkIO)
+			default:
+				ac.cc.resetAddrConn(ac.addr, true, errConnDrain)
+			}
 			return
 		case <-t.Error():
+			select {
+			case <-ac.ctx.Done():
+				t.Close()
+				return
+			case <-t.GoAway():
+				ac.cc.resetAddrConn(ac.addr, true, errNetworkIO)
+				return
+			default:
+			}
 			ac.mu.Lock()
 			if ac.state == Shutdown {
-				// ac.tearDown(...) has been invoked.
+				// ac has been shutdown.
 				ac.mu.Unlock()
 				return
 			}
@@ -643,38 +753,53 @@ func (ac *addrConn) transportMonitor() {
 				ac.printf("transport exiting: %v", err)
 				ac.mu.Unlock()
 				grpclog.Printf("grpc: addrConn.transportMonitor exits due to: %v", err)
+				if err != errConnClosing {
+					// Keep this ac in cc.conns, to get the reason it's torn down.
+					ac.tearDown(err)
+				}
 				return
 			}
 		}
 	}
 }
 
-// wait blocks until i) the new transport is up or ii) ctx is done or iii) ac is closed.
+// wait blocks until i) the new transport is up or ii) ctx is done or iii) ac is closed or
-func (ac *addrConn) wait(ctx context.Context) (transport.ClientTransport, error) {
+// iv) transport is in TransientFailure and there's no balancer/failfast is true.
+func (ac *addrConn) wait(ctx context.Context, hasBalancer, failfast bool) (transport.ClientTransport, error) {
 	for {
 		ac.mu.Lock()
 		switch {
 		case ac.state == Shutdown:
+			if failfast || !hasBalancer {
+				// RPC is failfast or balancer is nil. This RPC should fail with ac.tearDownErr.
+				err := ac.tearDownErr
+				ac.mu.Unlock()
+				return nil, err
+			}
 			ac.mu.Unlock()
 			return nil, errConnClosing
 		case ac.state == Ready:
 			ct := ac.transport
 			ac.mu.Unlock()
 			return ct, nil
-		default:
+		case ac.state == TransientFailure:
-			ready := ac.ready
+			if failfast || hasBalancer {
-			if ready == nil {
+				ac.mu.Unlock()
-				ready = make(chan struct{})
+				return nil, errConnUnavailable
-				ac.ready = ready
-			}
-			ac.mu.Unlock()
-			select {
-			case <-ctx.Done():
-				return nil, transport.ContextErr(ctx.Err())
-			// Wait until the new transport is ready or failed.
-			case <-ready:
 			}
 		}
+		ready := ac.ready
+		if ready == nil {
+			ready = make(chan struct{})
+			ac.ready = ready
+		}
+		ac.mu.Unlock()
+		select {
+		case <-ctx.Done():
+			return nil, toRPCErr(ctx.Err())
+		// Wait until the new transport is ready or failed.
+		case <-ready:
+		}
 	}
 }
 
@@ -682,24 +807,28 @@ func (ac *addrConn) wait(ctx context.Context) (transport.ClientTransport, error)
 // TODO(zhaoq): Make this synchronous to avoid unbounded memory consumption in
 // some edge cases (e.g., the caller opens and closes many addrConn's in a
 // tight loop.
+// tearDown doesn't remove ac from ac.cc.conns.
 func (ac *addrConn) tearDown(err error) {
+	ac.cancel()
+
 	ac.mu.Lock()
-	defer func() {
+	defer ac.mu.Unlock()
-		ac.mu.Unlock()
-		ac.cc.mu.Lock()
-		if ac.cc.conns != nil {
-			delete(ac.cc.conns, ac.addr)
-		}
-		ac.cc.mu.Unlock()
-	}()
-	if ac.state == Shutdown {
-		return
-	}
-	ac.state = Shutdown
 	if ac.down != nil {
 		ac.down(downErrorf(false, false, "%v", err))
 		ac.down = nil
 	}
+	if err == errConnDrain && ac.transport != nil {
+		// GracefulClose(...) may be executed multiple times when
+		// i) receiving multiple GoAway frames from the server; or
+		// ii) there are concurrent name resolver/Balancer triggered
+		// address removal and GoAway.
+		ac.transport.GracefulClose()
+	}
+	if ac.state == Shutdown {
+		return
+	}
+	ac.state = Shutdown
+	ac.tearDownErr = err
 	ac.stateCV.Broadcast()
 	if ac.events != nil {
 		ac.events.Finish()
@@ -709,15 +838,8 @@ func (ac *addrConn) tearDown(err error) {
 		close(ac.ready)
 		ac.ready = nil
 	}
-	if ac.transport != nil {
+	if ac.transport != nil && err != errConnDrain {
-		if err == errConnDrain {
+		ac.transport.Close()
-			ac.transport.GracefulClose()
-		} else {
-			ac.transport.Close()
-		}
-	}
-	if ac.shutdownChan != nil {
-		close(ac.shutdownChan)
 	}
 	return
 }

cmd/vendor/google.golang.org/grpc/credentials/credentials.go

@@ -44,7 +44,6 @@ import (
 	"io/ioutil"
 	"net"
 	"strings"
-	"time"
 
 	"golang.org/x/net/context"
 )
@@ -66,7 +65,7 @@ type PerRPCCredentials interface {
 	// TODO(zhaoq): Define the set of the qualified keys instead of leaving
 	// it as an arbitrary string.
 	GetRequestMetadata(ctx context.Context, uri ...string) (map[string]string, error)
-	// RequireTransportSecurity indicates whether the credentails requires
+	// RequireTransportSecurity indicates whether the credentials requires
 	// transport security.
 	RequireTransportSecurity() bool
 }
@@ -93,11 +92,12 @@ type TransportCredentials interface {
 	// ClientHandshake does the authentication handshake specified by the corresponding
 	// authentication protocol on rawConn for clients. It returns the authenticated
 	// connection and the corresponding auth information about the connection.
-	ClientHandshake(addr string, rawConn net.Conn, timeout time.Duration) (net.Conn, AuthInfo, error)
+	// Implementations must use the provided context to implement timely cancellation.
+	ClientHandshake(context.Context, string, net.Conn) (net.Conn, AuthInfo, error)
 	// ServerHandshake does the authentication handshake for servers. It returns
 	// the authenticated connection and the corresponding auth information about
 	// the connection.
-	ServerHandshake(rawConn net.Conn) (net.Conn, AuthInfo, error)
+	ServerHandshake(net.Conn) (net.Conn, AuthInfo, error)
 	// Info provides the ProtocolInfo of this TransportCredentials.
 	Info() ProtocolInfo
 }
@@ -116,7 +116,7 @@ func (t TLSInfo) AuthType() string {
 // tlsCreds is the credentials required for authenticating a connection using TLS.
 type tlsCreds struct {
 	// TLS configuration
-	config tls.Config
+	config *tls.Config
 }
 
 func (c tlsCreds) Info() ProtocolInfo {
@@ -136,40 +136,28 @@ func (c *tlsCreds) RequireTransportSecurity() bool {
 	return true
 }
 
-type timeoutError struct{}
+func (c *tlsCreds) ClientHandshake(ctx context.Context, addr string, rawConn net.Conn) (_ net.Conn, _ AuthInfo, err error) {
-
+	// use local cfg to avoid clobbering ServerName if using multiple endpoints
-func (timeoutError) Error() string   { return "credentials: Dial timed out" }
+	cfg := cloneTLSConfig(c.config)
-func (timeoutError) Timeout() bool   { return true }
+	if cfg.ServerName == "" {
-func (timeoutError) Temporary() bool { return true }
-
-func (c *tlsCreds) ClientHandshake(addr string, rawConn net.Conn, timeout time.Duration) (_ net.Conn, _ AuthInfo, err error) {
-	// borrow some code from tls.DialWithDialer
-	var errChannel chan error
-	if timeout != 0 {
-		errChannel = make(chan error, 2)
-		time.AfterFunc(timeout, func() {
-			errChannel <- timeoutError{}
-		})
-	}
-	if c.config.ServerName == "" {
 		colonPos := strings.LastIndex(addr, ":")
 		if colonPos == -1 {
 			colonPos = len(addr)
 		}
-		c.config.ServerName = addr[:colonPos]
+		cfg.ServerName = addr[:colonPos]
 	}
-	conn := tls.Client(rawConn, &c.config)
+	conn := tls.Client(rawConn, cfg)
-	if timeout == 0 {
+	errChannel := make(chan error, 1)
-		err = conn.Handshake()
+	go func() {
-	} else {
+		errChannel <- conn.Handshake()
-		go func() {
+	}()
-			errChannel <- conn.Handshake()
+	select {
-		}()
+	case err := <-errChannel:
-		err = <-errChannel
+		if err != nil {
-	}
+			return nil, nil, err
-	if err != nil {
+		}
-		rawConn.Close()
+	case <-ctx.Done():
-		return nil, nil, err
+		return nil, nil, ctx.Err()
 	}
 	// TODO(zhaoq): Omit the auth info for client now. It is more for
 	// information than anything else.
@@ -177,9 +165,8 @@ func (c *tlsCreds) ClientHandshake(addr string, rawConn net.Conn, timeout time.D
 }
 
 func (c *tlsCreds) ServerHandshake(rawConn net.Conn) (net.Conn, AuthInfo, error) {
-	conn := tls.Server(rawConn, &c.config)
+	conn := tls.Server(rawConn, c.config)
 	if err := conn.Handshake(); err != nil {
-		rawConn.Close()
 		return nil, nil, err
 	}
 	return conn, TLSInfo{conn.ConnectionState()}, nil
@@ -187,7 +174,7 @@ func (c *tlsCreds) ServerHandshake(rawConn net.Conn) (net.Conn, AuthInfo, error)
 
 // NewTLS uses c to construct a TransportCredentials based on TLS.
 func NewTLS(c *tls.Config) TransportCredentials {
-	tc := &tlsCreds{*c}
+	tc := &tlsCreds{cloneTLSConfig(c)}
 	tc.config.NextProtos = alpnProtoStr
 	return tc
 }

cmd/vendor/google.golang.org/grpc/credentials/credentials_util_go17.go

@@ -0,0 +1,76 @@
+// +build go1.7
+
+/*
+ *
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+package credentials
+
+import (
+	"crypto/tls"
+)
+
+// cloneTLSConfig returns a shallow clone of the exported
+// fields of cfg, ignoring the unexported sync.Once, which
+// contains a mutex and must not be copied.
+//
+// If cfg is nil, a new zero tls.Config is returned.
+//
+// TODO replace this function with official clone function.
+func cloneTLSConfig(cfg *tls.Config) *tls.Config {
+	if cfg == nil {
+		return &tls.Config{}
+	}
+	return &tls.Config{
+		Rand:                        cfg.Rand,
+		Time:                        cfg.Time,
+		Certificates:                cfg.Certificates,
+		NameToCertificate:           cfg.NameToCertificate,
+		GetCertificate:              cfg.GetCertificate,
+		RootCAs:                     cfg.RootCAs,
+		NextProtos:                  cfg.NextProtos,
+		ServerName:                  cfg.ServerName,
+		ClientAuth:                  cfg.ClientAuth,
+		ClientCAs:                   cfg.ClientCAs,
+		InsecureSkipVerify:          cfg.InsecureSkipVerify,
+		CipherSuites:                cfg.CipherSuites,
+		PreferServerCipherSuites:    cfg.PreferServerCipherSuites,
+		SessionTicketsDisabled:      cfg.SessionTicketsDisabled,
+		SessionTicketKey:            cfg.SessionTicketKey,
+		ClientSessionCache:          cfg.ClientSessionCache,
+		MinVersion:                  cfg.MinVersion,
+		MaxVersion:                  cfg.MaxVersion,
+		CurvePreferences:            cfg.CurvePreferences,
+		DynamicRecordSizingDisabled: cfg.DynamicRecordSizingDisabled,
+		Renegotiation:               cfg.Renegotiation,
+	}
+}

cmd/vendor/google.golang.org/grpc/credentials/credentials_util_pre_go17.go

@@ -0,0 +1,74 @@
+// +build !go1.7
+
+/*
+ *
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+package credentials
+
+import (
+	"crypto/tls"
+)
+
+// cloneTLSConfig returns a shallow clone of the exported
+// fields of cfg, ignoring the unexported sync.Once, which
+// contains a mutex and must not be copied.
+//
+// If cfg is nil, a new zero tls.Config is returned.
+//
+// TODO replace this function with official clone function.
+func cloneTLSConfig(cfg *tls.Config) *tls.Config {
+	if cfg == nil {
+		return &tls.Config{}
+	}
+	return &tls.Config{
+		Rand:                     cfg.Rand,
+		Time:                     cfg.Time,
+		Certificates:             cfg.Certificates,
+		NameToCertificate:        cfg.NameToCertificate,
+		GetCertificate:           cfg.GetCertificate,
+		RootCAs:                  cfg.RootCAs,
+		NextProtos:               cfg.NextProtos,
+		ServerName:               cfg.ServerName,
+		ClientAuth:               cfg.ClientAuth,
+		ClientCAs:                cfg.ClientCAs,
+		InsecureSkipVerify:       cfg.InsecureSkipVerify,
+		CipherSuites:             cfg.CipherSuites,
+		PreferServerCipherSuites: cfg.PreferServerCipherSuites,
+		SessionTicketsDisabled:   cfg.SessionTicketsDisabled,
+		SessionTicketKey:         cfg.SessionTicketKey,
+		ClientSessionCache:       cfg.ClientSessionCache,
+		MinVersion:               cfg.MinVersion,
+		MaxVersion:               cfg.MaxVersion,
+		CurvePreferences:         cfg.CurvePreferences,
+	}
+}

cmd/vendor/google.golang.org/grpc/metadata/metadata.go

@@ -60,15 +60,21 @@ func encodeKeyValue(k, v string) (string, string) {
 
 // DecodeKeyValue returns the original key and value corresponding to the
 // encoded data in k, v.
+// If k is a binary header and v contains comma, v is split on comma before decoded,
+// and the decoded v will be joined with comma before returned.
 func DecodeKeyValue(k, v string) (string, string, error) {
 	if !strings.HasSuffix(k, binHdrSuffix) {
 		return k, v, nil
 	}
-	val, err := base64.StdEncoding.DecodeString(v)
+	vvs := strings.Split(v, ",")
-	if err != nil {
+	for i, vv := range vvs {
-		return "", "", err
+		val, err := base64.StdEncoding.DecodeString(vv)
+		if err != nil {
+			return "", "", err
+		}
+		vvs[i] = string(val)
 	}
-	return k, string(val), nil
+	return k, strings.Join(vvs, ","), nil
 }
 
 // MD is a mapping from metadata keys to values. Users should use the following

cmd/vendor/google.golang.org/grpc/rpc_util.go

@@ -141,6 +141,8 @@ type callInfo struct {
 	traceInfo traceInfo // in trace.go
 }
 
+var defaultCallInfo = callInfo{failFast: true}
+
 // CallOption configures a Call before it starts or extracts information from
 // a Call after it completes.
 type CallOption interface {
@@ -179,6 +181,19 @@ func Trailer(md *metadata.MD) CallOption {
 	})
 }
 
+// FailFast configures the action to take when an RPC is attempted on broken
+// connections or unreachable servers. If failfast is true, the RPC will fail
+// immediately. Otherwise, the RPC client will block the call until a
+// connection is available (or the call is canceled or times out) and will retry
+// the call if it fails due to a transient error. Please refer to
+// https://github.com/grpc/grpc/blob/master/doc/fail_fast.md
+func FailFast(failFast bool) CallOption {
+	return beforeCall(func(c *callInfo) error {
+		c.failFast = failFast
+		return nil
+	})
+}
+
 // The format of the payload: compressed or not?
 type payloadFormat uint8
 
@@ -212,7 +227,7 @@ type parser struct {
 // No other error values or types must be returned, which also means
 // that the underlying io.Reader must not return an incompatible
 // error.
-func (p *parser) recvMsg() (pf payloadFormat, msg []byte, err error) {
+func (p *parser) recvMsg(maxMsgSize int) (pf payloadFormat, msg []byte, err error) {
 	if _, err := io.ReadFull(p.r, p.header[:]); err != nil {
 		return 0, nil, err
 	}
@@ -223,6 +238,9 @@ func (p *parser) recvMsg() (pf payloadFormat, msg []byte, err error) {
 	if length == 0 {
 		return pf, nil, nil
 	}
+	if length > uint32(maxMsgSize) {
+		return 0, nil, Errorf(codes.Internal, "grpc: received message length %d exceeding the max size %d", length, maxMsgSize)
+	}
 	// TODO(bradfitz,zhaoq): garbage. reuse buffer after proto decoding instead
 	// of making it for each message:
 	msg = make([]byte, int(length))
@@ -293,8 +311,8 @@ func checkRecvPayload(pf payloadFormat, recvCompress string, dc Decompressor) er
 	return nil
 }
 
-func recv(p *parser, c Codec, s *transport.Stream, dc Decompressor, m interface{}) error {
+func recv(p *parser, c Codec, s *transport.Stream, dc Decompressor, m interface{}, maxMsgSize int) error {
-	pf, d, err := p.recvMsg()
+	pf, d, err := p.recvMsg(maxMsgSize)
 	if err != nil {
 		return err
 	}
@@ -304,11 +322,16 @@ func recv(p *parser, c Codec, s *transport.Stream, dc Decompressor, m interface{
 	if pf == compressionMade {
 		d, err = dc.Do(bytes.NewReader(d))
 		if err != nil {
-			return transport.StreamErrorf(codes.Internal, "grpc: failed to decompress the received message %v", err)
+			return Errorf(codes.Internal, "grpc: failed to decompress the received message %v", err)
 		}
 	}
+	if len(d) > maxMsgSize {
+		// TODO: Revisit the error code. Currently keep it consistent with java
+		// implementation.
+		return Errorf(codes.Internal, "grpc: received a message of %d bytes exceeding %d limit", len(d), maxMsgSize)
+	}
 	if err := c.Unmarshal(d, m); err != nil {
-		return transport.StreamErrorf(codes.Internal, "grpc: failed to unmarshal the received message %v", err)
+		return Errorf(codes.Internal, "grpc: failed to unmarshal the received message %v", err)
 	}
 	return nil
 }
@@ -319,7 +342,7 @@ type rpcError struct {
 	desc string
 }
 
-func (e rpcError) Error() string {
+func (e *rpcError) Error() string {
 	return fmt.Sprintf("rpc error: code = %d desc = %s", e.code, e.desc)
 }
 
@@ -329,7 +352,7 @@ func Code(err error) codes.Code {
 	if err == nil {
 		return codes.OK
 	}
-	if e, ok := err.(rpcError); ok {
+	if e, ok := err.(*rpcError); ok {
 		return e.code
 	}
 	return codes.Unknown
@@ -341,7 +364,7 @@ func ErrorDesc(err error) string {
 	if err == nil {
 		return ""
 	}
-	if e, ok := err.(rpcError); ok {
+	if e, ok := err.(*rpcError); ok {
 		return e.desc
 	}
 	return err.Error()
@@ -353,7 +376,7 @@ func Errorf(c codes.Code, format string, a ...interface{}) error {
 	if c == codes.OK {
 		return nil
 	}
-	return rpcError{
+	return &rpcError{
 		code: c,
 		desc: fmt.Sprintf(format, a...),
 	}
@@ -362,18 +385,37 @@ func Errorf(c codes.Code, format string, a ...interface{}) error {
 // toRPCErr converts an error into a rpcError.
 func toRPCErr(err error) error {
 	switch e := err.(type) {
-	case rpcError:
+	case *rpcError:
 		return err
 	case transport.StreamError:
-		return rpcError{
+		return &rpcError{
 			code: e.Code,
 			desc: e.Desc,
 		}
 	case transport.ConnectionError:
-		return rpcError{
+		return &rpcError{
 			code: codes.Internal,
 			desc: e.Desc,
 		}
+	default:
+		switch err {
+		case context.DeadlineExceeded:
+			return &rpcError{
+				code: codes.DeadlineExceeded,
+				desc: err.Error(),
+			}
+		case context.Canceled:
+			return &rpcError{
+				code: codes.Canceled,
+				desc: err.Error(),
+			}
+		case ErrClientConnClosing:
+			return &rpcError{
+				code: codes.FailedPrecondition,
+				desc: err.Error(),
+			}
+		}
+
 	}
 	return Errorf(codes.Unknown, "%v", err)
 }

cmd/vendor/google.golang.org/grpc/server.go

@@ -82,15 +82,20 @@ type service struct {
 	server interface{} // the server for service methods
 	md     map[string]*MethodDesc
 	sd     map[string]*StreamDesc
+	mdata  interface{}
 }
 
 // Server is a gRPC server to serve RPC requests.
 type Server struct {
 	opts options
 
-	mu     sync.Mutex // guards following
+	mu    sync.Mutex // guards following
-	lis    map[net.Listener]bool
+	lis   map[net.Listener]bool
-	conns  map[io.Closer]bool
+	conns map[io.Closer]bool
+	drain bool
+	// A CondVar to let GracefulStop() blocks until all the pending RPCs are finished
+	// and all the transport goes away.
+	cv     *sync.Cond
 	m      map[string]*service // service name -> service info
 	events trace.EventLog
 }
@@ -100,12 +105,15 @@ type options struct {
 	codec                Codec
 	cp                   Compressor
 	dc                   Decompressor
+	maxMsgSize           int
 	unaryInt             UnaryServerInterceptor
 	streamInt            StreamServerInterceptor
 	maxConcurrentStreams uint32
 	useHandlerImpl       bool // use http.Handler-based server
 }
 
+var defaultMaxMsgSize = 1024 * 1024 * 4 // use 4MB as the default message size limit
+
 // A ServerOption sets options.
 type ServerOption func(*options)
 
@@ -116,20 +124,28 @@ func CustomCodec(codec Codec) ServerOption {
 	}
 }
 
-// RPCCompressor returns a ServerOption that sets a compressor for outbound message.
+// RPCCompressor returns a ServerOption that sets a compressor for outbound messages.
 func RPCCompressor(cp Compressor) ServerOption {
 	return func(o *options) {
 		o.cp = cp
 	}
 }
 
-// RPCDecompressor returns a ServerOption that sets a decompressor for inbound message.
+// RPCDecompressor returns a ServerOption that sets a decompressor for inbound messages.
 func RPCDecompressor(dc Decompressor) ServerOption {
 	return func(o *options) {
 		o.dc = dc
 	}
 }
 
+// MaxMsgSize returns a ServerOption to set the max message size in bytes for inbound mesages.
+// If this is not set, gRPC uses the default 4MB.
+func MaxMsgSize(m int) ServerOption {
+	return func(o *options) {
+		o.maxMsgSize = m
+	}
+}
+
 // MaxConcurrentStreams returns a ServerOption that will apply a limit on the number
 // of concurrent streams to each ServerTransport.
 func MaxConcurrentStreams(n uint32) ServerOption {
@@ -172,6 +188,7 @@ func StreamInterceptor(i StreamServerInterceptor) ServerOption {
 // started to accept requests yet.
 func NewServer(opt ...ServerOption) *Server {
 	var opts options
+	opts.maxMsgSize = defaultMaxMsgSize
 	for _, o := range opt {
 		o(&opts)
 	}
@@ -185,6 +202,7 @@ func NewServer(opt ...ServerOption) *Server {
 		conns: make(map[io.Closer]bool),
 		m:     make(map[string]*service),
 	}
+	s.cv = sync.NewCond(&s.mu)
 	if EnableTracing {
 		_, file, line, _ := runtime.Caller(1)
 		s.events = trace.NewEventLog("grpc.Server", fmt.Sprintf("%s:%d", file, line))
@@ -231,6 +249,7 @@ func (s *Server) register(sd *ServiceDesc, ss interface{}) {
 		server: ss,
 		md:     make(map[string]*MethodDesc),
 		sd:     make(map[string]*StreamDesc),
+		mdata:  sd.Metadata,
 	}
 	for i := range sd.Methods {
 		d := &sd.Methods[i]
@@ -243,6 +262,52 @@ func (s *Server) register(sd *ServiceDesc, ss interface{}) {
 	s.m[sd.ServiceName] = srv
 }
 
+// MethodInfo contains the information of an RPC including its method name and type.
+type MethodInfo struct {
+	// Name is the method name only, without the service name or package name.
+	Name string
+	// IsClientStream indicates whether the RPC is a client streaming RPC.
+	IsClientStream bool
+	// IsServerStream indicates whether the RPC is a server streaming RPC.
+	IsServerStream bool
+}
+
+// ServiceInfo contains unary RPC method info, streaming RPC methid info and metadata for a service.
+type ServiceInfo struct {
+	Methods []MethodInfo
+	// Metadata is the metadata specified in ServiceDesc when registering service.
+	Metadata interface{}
+}
+
+// GetServiceInfo returns a map from service names to ServiceInfo.
+// Service names include the package names, in the form of <package>.<service>.
+func (s *Server) GetServiceInfo() map[string]ServiceInfo {
+	ret := make(map[string]ServiceInfo)
+	for n, srv := range s.m {
+		methods := make([]MethodInfo, 0, len(srv.md)+len(srv.sd))
+		for m := range srv.md {
+			methods = append(methods, MethodInfo{
+				Name:           m,
+				IsClientStream: false,
+				IsServerStream: false,
+			})
+		}
+		for m, d := range srv.sd {
+			methods = append(methods, MethodInfo{
+				Name:           m,
+				IsClientStream: d.ClientStreams,
+				IsServerStream: d.ServerStreams,
+			})
+		}
+
+		ret[n] = ServiceInfo{
+			Methods:  methods,
+			Metadata: srv.mdata,
+		}
+	}
+	return ret
+}
+
 var (
 	// ErrServerStopped indicates that the operation is now illegal because of
 	// the server being stopped.
@@ -272,9 +337,11 @@ func (s *Server) Serve(lis net.Listener) error {
 	s.lis[lis] = true
 	s.mu.Unlock()
 	defer func() {
-		lis.Close()
 		s.mu.Lock()
-		delete(s.lis, lis)
+		if s.lis != nil && s.lis[lis] {
+			lis.Close()
+			delete(s.lis, lis)
+		}
 		s.mu.Unlock()
 	}()
 	for {
@@ -418,7 +485,7 @@ func (s *Server) traceInfo(st transport.ServerTransport, stream *transport.Strea
 func (s *Server) addConn(c io.Closer) bool {
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	if s.conns == nil {
+	if s.conns == nil || s.drain {
 		return false
 	}
 	s.conns[c] = true
@@ -430,6 +497,7 @@ func (s *Server) removeConn(c io.Closer) {
 	defer s.mu.Unlock()
 	if s.conns != nil {
 		delete(s.conns, c)
+		s.cv.Signal()
 	}
 }
 
@@ -470,7 +538,7 @@ func (s *Server) processUnaryRPC(t transport.ServerTransport, stream *transport.
 	}
 	p := &parser{r: stream}
 	for {
-		pf, req, err := p.recvMsg()
+		pf, req, err := p.recvMsg(s.opts.maxMsgSize)
 		if err == io.EOF {
 			// The entire stream is done (for unary RPC only).
 			return err
@@ -480,6 +548,10 @@ func (s *Server) processUnaryRPC(t transport.ServerTransport, stream *transport.
 		}
 		if err != nil {
 			switch err := err.(type) {
+			case *rpcError:
+				if err := t.WriteStatus(stream, err.code, err.desc); err != nil {
+					grpclog.Printf("grpc: Server.processUnaryRPC failed to write status %v", err)
+				}
 			case transport.ConnectionError:
 				// Nothing to do here.
 			case transport.StreamError:
@@ -519,6 +591,12 @@ func (s *Server) processUnaryRPC(t transport.ServerTransport, stream *transport.
 					return err
 				}
 			}
+			if len(req) > s.opts.maxMsgSize {
+				// TODO: Revisit the error code. Currently keep it consistent with
+				// java implementation.
+				statusCode = codes.Internal
+				statusDesc = fmt.Sprintf("grpc: server received a message of %d bytes exceeding %d limit", len(req), s.opts.maxMsgSize)
+			}
 			if err := s.opts.codec.Unmarshal(req, v); err != nil {
 				return err
 			}
@@ -529,7 +607,7 @@ func (s *Server) processUnaryRPC(t transport.ServerTransport, stream *transport.
 		}
 		reply, appErr := md.Handler(srv.server, stream.Context(), df, s.opts.unaryInt)
 		if appErr != nil {
-			if err, ok := appErr.(rpcError); ok {
+			if err, ok := appErr.(*rpcError); ok {
 				statusCode = err.code
 				statusDesc = err.desc
 			} else {
@@ -578,13 +656,14 @@ func (s *Server) processStreamingRPC(t transport.ServerTransport, stream *transp
 		stream.SetSendCompress(s.opts.cp.Type())
 	}
 	ss := &serverStream{
-		t:      t,
+		t:          t,
-		s:      stream,
+		s:          stream,
-		p:      &parser{r: stream},
+		p:          &parser{r: stream},
-		codec:  s.opts.codec,
+		codec:      s.opts.codec,
-		cp:     s.opts.cp,
+		cp:         s.opts.cp,
-		dc:     s.opts.dc,
+		dc:         s.opts.dc,
-		trInfo: trInfo,
+		maxMsgSize: s.opts.maxMsgSize,
+		trInfo:     trInfo,
 	}
 	if ss.cp != nil {
 		ss.cbuf = new(bytes.Buffer)
@@ -614,7 +693,7 @@ func (s *Server) processStreamingRPC(t transport.ServerTransport, stream *transp
 		appErr = s.opts.streamInt(srv.server, ss, info, sd.Handler)
 	}
 	if appErr != nil {
-		if err, ok := appErr.(rpcError); ok {
+		if err, ok := appErr.(*rpcError); ok {
 			ss.statusCode = err.code
 			ss.statusDesc = err.desc
 		} else if err, ok := appErr.(transport.StreamError); ok {
@@ -716,14 +795,16 @@ func (s *Server) Stop() {
 	s.mu.Lock()
 	listeners := s.lis
 	s.lis = nil
-	cs := s.conns
+	st := s.conns
 	s.conns = nil
+	// interrupt GracefulStop if Stop and GracefulStop are called concurrently.
+	s.cv.Signal()
 	s.mu.Unlock()
 
 	for lis := range listeners {
 		lis.Close()
 	}
-	for c := range cs {
+	for c := range st {
 		c.Close()
 	}
 
@@ -735,6 +816,32 @@ func (s *Server) Stop() {
 	s.mu.Unlock()
 }
 
+// GracefulStop stops the gRPC server gracefully. It stops the server to accept new
+// connections and RPCs and blocks until all the pending RPCs are finished.
+func (s *Server) GracefulStop() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if s.drain == true || s.conns == nil {
+		return
+	}
+	s.drain = true
+	for lis := range s.lis {
+		lis.Close()
+	}
+	s.lis = nil
+	for c := range s.conns {
+		c.(transport.ServerTransport).Drain()
+	}
+	for len(s.conns) != 0 {
+		s.cv.Wait()
+	}
+	s.conns = nil
+	if s.events != nil {
+		s.events.Finish()
+		s.events = nil
+	}
+}
+
 func init() {
 	internal.TestingCloseConns = func(arg interface{}) {
 		arg.(*Server).testingCloseConns()

cmd/vendor/google.golang.org/grpc/stream.go

@@ -37,6 +37,7 @@ import (
 	"bytes"
 	"errors"
 	"io"
+	"math"
 	"sync"
 	"time"
 
@@ -84,12 +85,9 @@ type ClientStream interface {
 	// Header returns the header metadata received from the server if there
 	// is any. It blocks if the metadata is not ready to read.
 	Header() (metadata.MD, error)
-	// Trailer returns the trailer metadata from the server. It must be called
+	// Trailer returns the trailer metadata from the server, if there is any.
-	// after stream.Recv() returns non-nil error (including io.EOF) for
+	// It must only be called after stream.CloseAndRecv has returned, or
-	// bi-directional streaming and server streaming or stream.CloseAndRecv()
+	// stream.Recv has returned a non-nil error (including io.EOF).
-	// returns for client streaming in order to receive trailer metadata if
-	// present. Otherwise, it could returns an empty MD even though trailer
-	// is present.
 	Trailer() metadata.MD
 	// CloseSend closes the send direction of the stream. It closes the stream
 	// when non-nil error is met.
@@ -99,19 +97,17 @@ type ClientStream interface {
 
 // NewClientStream creates a new Stream for the client side. This is called
 // by generated code.
-func NewClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, opts ...CallOption) (ClientStream, error) {
+func NewClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, method string, opts ...CallOption) (_ ClientStream, err error) {
 	var (
 		t   transport.ClientTransport
-		err error
+		s   *transport.Stream
 		put func()
 	)
-	// TODO(zhaoq): CallOption is omitted. Add support when it is needed.
+	c := defaultCallInfo
-	gopts := BalancerGetOptions{
+	for _, o := range opts {
-		BlockingWait: false,
+		if err := o.before(&c); err != nil {
-	}
+			return nil, toRPCErr(err)
-	t, put, err = cc.getTransport(ctx, gopts)
+		}
-	if err != nil {
-		return nil, toRPCErr(err)
 	}
 	callHdr := &transport.CallHdr{
 		Host:   cc.authority,
@@ -121,41 +117,98 @@ func NewClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, meth
 	if cc.dopts.cp != nil {
 		callHdr.SendCompress = cc.dopts.cp.Type()
 	}
+	var trInfo traceInfo
+	if EnableTracing {
+		trInfo.tr = trace.New("grpc.Sent."+methodFamily(method), method)
+		trInfo.firstLine.client = true
+		if deadline, ok := ctx.Deadline(); ok {
+			trInfo.firstLine.deadline = deadline.Sub(time.Now())
+		}
+		trInfo.tr.LazyLog(&trInfo.firstLine, false)
+		ctx = trace.NewContext(ctx, trInfo.tr)
+		defer func() {
+			if err != nil {
+				// Need to call tr.finish() if error is returned.
+				// Because tr will not be returned to caller.
+				trInfo.tr.LazyPrintf("RPC: [%v]", err)
+				trInfo.tr.SetError()
+				trInfo.tr.Finish()
+			}
+		}()
+	}
+	gopts := BalancerGetOptions{
+		BlockingWait: !c.failFast,
+	}
+	for {
+		t, put, err = cc.getTransport(ctx, gopts)
+		if err != nil {
+			// TODO(zhaoq): Probably revisit the error handling.
+			if _, ok := err.(*rpcError); ok {
+				return nil, err
+			}
+			if err == errConnClosing || err == errConnUnavailable {
+				if c.failFast {
+					return nil, Errorf(codes.Unavailable, "%v", err)
+				}
+				continue
+			}
+			// All the other errors are treated as Internal errors.
+			return nil, Errorf(codes.Internal, "%v", err)
+		}
+
+		s, err = t.NewStream(ctx, callHdr)
+		if err != nil {
+			if put != nil {
+				put()
+				put = nil
+			}
+			if _, ok := err.(transport.ConnectionError); ok || err == transport.ErrStreamDrain {
+				if c.failFast {
+					return nil, toRPCErr(err)
+				}
+				continue
+			}
+			return nil, toRPCErr(err)
+		}
+		break
+	}
 	cs := &clientStream{
-		desc:    desc,
+		opts:  opts,
-		put:     put,
+		c:     c,
-		codec:   cc.dopts.codec,
+		desc:  desc,
-		cp:      cc.dopts.cp,
+		codec: cc.dopts.codec,
-		dc:      cc.dopts.dc,
+		cp:    cc.dopts.cp,
+		dc:    cc.dopts.dc,
+
+		put: put,
+		t:   t,
+		s:   s,
+		p:   &parser{r: s},
+
 		tracing: EnableTracing,
+		trInfo:  trInfo,
 	}
 	if cc.dopts.cp != nil {
-		callHdr.SendCompress = cc.dopts.cp.Type()
 		cs.cbuf = new(bytes.Buffer)
 	}
-	if cs.tracing {
+	// Listen on ctx.Done() to detect cancellation and s.Done() to detect normal termination
-		cs.trInfo.tr = trace.New("grpc.Sent."+methodFamily(method), method)
+	// when there is no pending I/O operations on this stream.
-		cs.trInfo.firstLine.client = true
-		if deadline, ok := ctx.Deadline(); ok {
-			cs.trInfo.firstLine.deadline = deadline.Sub(time.Now())
-		}
-		cs.trInfo.tr.LazyLog(&cs.trInfo.firstLine, false)
-		ctx = trace.NewContext(ctx, cs.trInfo.tr)
-	}
-	s, err := t.NewStream(ctx, callHdr)
-	if err != nil {
-		cs.finish(err)
-		return nil, toRPCErr(err)
-	}
-	cs.t = t
-	cs.s = s
-	cs.p = &parser{r: s}
-	// Listen on ctx.Done() to detect cancellation when there is no pending
-	// I/O operations on this stream.
 	go func() {
 		select {
 		case <-t.Error():
 			// Incur transport error, simply exit.
+		case <-s.Done():
+			// TODO: The trace of the RPC is terminated here when there is no pending
+			// I/O, which is probably not the optimal solution.
+			if s.StatusCode() == codes.OK {
+				cs.finish(nil)
+			} else {
+				cs.finish(Errorf(s.StatusCode(), "%s", s.StatusDesc()))
+			}
+			cs.closeTransportStream(nil)
+		case <-s.GoAway():
+			cs.finish(errConnDrain)
+			cs.closeTransportStream(errConnDrain)
 		case <-s.Context().Done():
 			err := s.Context().Err()
 			cs.finish(err)
@@ -167,6 +220,8 @@ func NewClientStream(ctx context.Context, desc *StreamDesc, cc *ClientConn, meth
 
 // clientStream implements a client side Stream.
 type clientStream struct {
+	opts  []CallOption
+	c     callInfo
 	t     transport.ClientTransport
 	s     *transport.Stream
 	p     *parser
@@ -216,7 +271,17 @@ func (cs *clientStream) SendMsg(m interface{}) (err error) {
 		if err != nil {
 			cs.finish(err)
 		}
-		if err == nil || err == io.EOF {
+		if err == nil {
+			return
+		}
+		if err == io.EOF {
+			// Specialize the process for server streaming. SendMesg is only called
+			// once when creating the stream object. io.EOF needs to be skipped when
+			// the rpc is early finished (before the stream object is created.).
+			// TODO: It is probably better to move this into the generated code.
+			if !cs.desc.ClientStreams && cs.desc.ServerStreams {
+				err = nil
+			}
 			return
 		}
 		if _, ok := err.(transport.ConnectionError); !ok {
@@ -237,7 +302,7 @@ func (cs *clientStream) SendMsg(m interface{}) (err error) {
 }
 
 func (cs *clientStream) RecvMsg(m interface{}) (err error) {
-	err = recv(cs.p, cs.codec, cs.s, cs.dc, m)
+	err = recv(cs.p, cs.codec, cs.s, cs.dc, m, math.MaxInt32)
 	defer func() {
 		// err != nil indicates the termination of the stream.
 		if err != nil {
@@ -256,7 +321,7 @@ func (cs *clientStream) RecvMsg(m interface{}) (err error) {
 			return
 		}
 		// Special handling for client streaming rpc.
-		err = recv(cs.p, cs.codec, cs.s, cs.dc, m)
+		err = recv(cs.p, cs.codec, cs.s, cs.dc, m, math.MaxInt32)
 		cs.closeTransportStream(err)
 		if err == nil {
 			return toRPCErr(errors.New("grpc: client streaming protocol violation: get <nil>, want <EOF>"))
@@ -291,7 +356,7 @@ func (cs *clientStream) CloseSend() (err error) {
 		}
 	}()
 	if err == nil || err == io.EOF {
-		return
+		return nil
 	}
 	if _, ok := err.(transport.ConnectionError); !ok {
 		cs.closeTransportStream(err)
@@ -312,15 +377,18 @@ func (cs *clientStream) closeTransportStream(err error) {
 }
 
 func (cs *clientStream) finish(err error) {
-	if !cs.tracing {
-		return
-	}
 	cs.mu.Lock()
 	defer cs.mu.Unlock()
+	for _, o := range cs.opts {
+		o.after(&cs.c)
+	}
 	if cs.put != nil {
 		cs.put()
 		cs.put = nil
 	}
+	if !cs.tracing {
+		return
+	}
 	if cs.trInfo.tr != nil {
 		if err == nil || err == io.EOF {
 			cs.trInfo.tr.LazyPrintf("RPC: [OK]")
@@ -354,6 +422,7 @@ type serverStream struct {
 	cp         Compressor
 	dc         Decompressor
 	cbuf       *bytes.Buffer
+	maxMsgSize int
 	statusCode codes.Code
 	statusDesc string
 	trInfo     *traceInfo
@@ -420,5 +489,5 @@ func (ss *serverStream) RecvMsg(m interface{}) (err error) {
 			ss.mu.Unlock()
 		}
 	}()
-	return recv(ss.p, ss.codec, ss.s, ss.dc, m)
+	return recv(ss.p, ss.codec, ss.s, ss.dc, m, ss.maxMsgSize)
 }

cmd/vendor/google.golang.org/grpc/transport/control.go

@@ -72,6 +72,11 @@ type resetStream struct {
 
 func (*resetStream) item() {}
 
+type goAway struct {
+}
+
+func (*goAway) item() {}
+
 type flushIO struct {
 }
 

cmd/vendor/google.golang.org/grpc/transport/go16.go

@@ -0,0 +1,46 @@
+// +build go1.6,!go1.7
+
+/*
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+package transport
+
+import (
+	"net"
+
+	"golang.org/x/net/context"
+)
+
+// dialContext connects to the address on the named network.
+func dialContext(ctx context.Context, network, address string) (net.Conn, error) {
+	return (&net.Dialer{Cancel: ctx.Done()}).Dial(network, address)
+}

cmd/vendor/google.golang.org/grpc/transport/go17.go

@@ -0,0 +1,46 @@
+// +build go1.7
+
+/*
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+package transport
+
+import (
+	"net"
+
+	"golang.org/x/net/context"
+)
+
+// dialContext connects to the address on the named network.
+func dialContext(ctx context.Context, network, address string) (net.Conn, error) {
+	return (&net.Dialer{}).DialContext(ctx, network, address)
+}

cmd/vendor/google.golang.org/grpc/transport/handler_server.go

@@ -83,7 +83,7 @@ func NewServerHandlerTransport(w http.ResponseWriter, r *http.Request) (ServerTr
 	}
 
 	if v := r.Header.Get("grpc-timeout"); v != "" {
-		to, err := timeoutDecode(v)
+		to, err := decodeTimeout(v)
 		if err != nil {
 			return nil, StreamErrorf(codes.Internal, "malformed time-out: %v", err)
 		}
@@ -194,7 +194,7 @@ func (ht *serverHandlerTransport) WriteStatus(s *Stream, statusCode codes.Code,
 		h := ht.rw.Header()
 		h.Set("Grpc-Status", fmt.Sprintf("%d", statusCode))
 		if statusDesc != "" {
-			h.Set("Grpc-Message", statusDesc)
+			h.Set("Grpc-Message", encodeGrpcMessage(statusDesc))
 		}
 		if md := s.Trailer(); len(md) > 0 {
 			for k, vv := range md {
@@ -312,7 +312,7 @@ func (ht *serverHandlerTransport) HandleStreams(startStream func(*Stream)) {
 		Addr: ht.RemoteAddr(),
 	}
 	if req.TLS != nil {
-		pr.AuthInfo = credentials.TLSInfo{*req.TLS}
+		pr.AuthInfo = credentials.TLSInfo{State: *req.TLS}
 	}
 	ctx = metadata.NewContext(ctx, ht.headerMD)
 	ctx = peer.NewContext(ctx, pr)
@@ -370,6 +370,10 @@ func (ht *serverHandlerTransport) runStream() {
 	}
 }
 
+func (ht *serverHandlerTransport) Drain() {
+	panic("Drain() is not implemented")
+}
+
 // mapRecvMsgError returns the non-nil err into the appropriate
 // error value as expected by callers of *grpc.parser.recvMsg.
 // In particular, in can only be:

cmd/vendor/google.golang.org/grpc/transport/http2_client.go

@@ -35,6 +35,7 @@ package transport
 
 import (
 	"bytes"
+	"fmt"
 	"io"
 	"math"
 	"net"
@@ -71,6 +72,9 @@ type http2Client struct {
 	shutdownChan chan struct{}
 	// errorChan is closed to notify the I/O error to the caller.
 	errorChan chan struct{}
+	// goAway is closed to notify the upper layer (i.e., addrConn.transportMonitor)
+	// that the server sent GoAway on this transport.
+	goAway chan struct{}
 
 	framer *framer
 	hBuf   *bytes.Buffer  // the buffer for HPACK encoding
@@ -97,41 +101,44 @@ type http2Client struct {
 	maxStreams int
 	// the per-stream outbound flow control window size set by the peer.
 	streamSendQuota uint32
+	// goAwayID records the Last-Stream-ID in the GoAway frame from the server.
+	goAwayID uint32
+	// prevGoAway ID records the Last-Stream-ID in the previous GOAway frame.
+	prevGoAwayID uint32
+}
+
+func dial(fn func(context.Context, string) (net.Conn, error), ctx context.Context, addr string) (net.Conn, error) {
+	if fn != nil {
+		return fn(ctx, addr)
+	}
+	return dialContext(ctx, "tcp", addr)
 }
 
 // newHTTP2Client constructs a connected ClientTransport to addr based on HTTP2
 // and starts to receive messages on it. Non-nil error returns if construction
 // fails.
-func newHTTP2Client(addr string, opts *ConnectOptions) (_ ClientTransport, err error) {
+func newHTTP2Client(ctx context.Context, addr string, opts ConnectOptions) (_ ClientTransport, err error) {
-	if opts.Dialer == nil {
-		// Set the default Dialer.
-		opts.Dialer = func(addr string, timeout time.Duration) (net.Conn, error) {
-			return net.DialTimeout("tcp", addr, timeout)
-		}
-	}
 	scheme := "http"
-	startT := time.Now()
+	conn, connErr := dial(opts.Dialer, ctx, addr)
-	timeout := opts.Timeout
-	conn, connErr := opts.Dialer(addr, timeout)
 	if connErr != nil {
-		return nil, ConnectionErrorf("transport: %v", connErr)
+		return nil, ConnectionErrorf(true, connErr, "transport: %v", connErr)
 	}
-	var authInfo credentials.AuthInfo
+	// Any further errors will close the underlying connection
-	if opts.TransportCredentials != nil {
+	defer func(conn net.Conn) {
-		scheme = "https"
-		if timeout > 0 {
-			timeout -= time.Since(startT)
-		}
-		conn, authInfo, connErr = opts.TransportCredentials.ClientHandshake(addr, conn, timeout)
-	}
-	if connErr != nil {
-		return nil, ConnectionErrorf("transport: %v", connErr)
-	}
-	defer func() {
 		if err != nil {
 			conn.Close()
 		}
-	}()
+	}(conn)
+	var authInfo credentials.AuthInfo
+	if creds := opts.TransportCredentials; creds != nil {
+		scheme = "https"
+		conn, authInfo, connErr = creds.ClientHandshake(ctx, addr, conn)
+	}
+	if connErr != nil {
+		// Credentials handshake error is not a temporary error (unless the error
+		// was the connection closing).
+		return nil, ConnectionErrorf(connErr == io.EOF, connErr, "transport: %v", connErr)
+	}
 	ua := primaryUA
 	if opts.UserAgent != "" {
 		ua = opts.UserAgent + " " + ua
@@ -147,6 +154,7 @@ func newHTTP2Client(addr string, opts *ConnectOptions) (_ ClientTransport, err e
 		writableChan:    make(chan int, 1),
 		shutdownChan:    make(chan struct{}),
 		errorChan:       make(chan struct{}),
+		goAway:          make(chan struct{}),
 		framer:          newFramer(conn),
 		hBuf:            &buf,
 		hEnc:            hpack.NewEncoder(&buf),
@@ -168,26 +176,29 @@ func newHTTP2Client(addr string, opts *ConnectOptions) (_ ClientTransport, err e
 	n, err := t.conn.Write(clientPreface)
 	if err != nil {
 		t.Close()
-		return nil, ConnectionErrorf("transport: %v", err)
+		return nil, ConnectionErrorf(true, err, "transport: %v", err)
 	}
 	if n != len(clientPreface) {
 		t.Close()
-		return nil, ConnectionErrorf("transport: preface mismatch, wrote %d bytes; want %d", n, len(clientPreface))
+		return nil, ConnectionErrorf(true, err, "transport: preface mismatch, wrote %d bytes; want %d", n, len(clientPreface))
 	}
 	if initialWindowSize != defaultWindowSize {
-		err = t.framer.writeSettings(true, http2.Setting{http2.SettingInitialWindowSize, uint32(initialWindowSize)})
+		err = t.framer.writeSettings(true, http2.Setting{
+			ID:  http2.SettingInitialWindowSize,
+			Val: uint32(initialWindowSize),
+		})
 	} else {
 		err = t.framer.writeSettings(true)
 	}
 	if err != nil {
 		t.Close()
-		return nil, ConnectionErrorf("transport: %v", err)
+		return nil, ConnectionErrorf(true, err, "transport: %v", err)
 	}
 	// Adjust the connection flow control window if needed.
 	if delta := uint32(initialConnWindowSize - defaultWindowSize); delta > 0 {
 		if err := t.framer.writeWindowUpdate(true, 0, delta); err != nil {
 			t.Close()
-			return nil, ConnectionErrorf("transport: %v", err)
+			return nil, ConnectionErrorf(true, err, "transport: %v", err)
 		}
 	}
 	go t.controller()
@@ -199,6 +210,8 @@ func (t *http2Client) newStream(ctx context.Context, callHdr *CallHdr) *Stream {
 	// TODO(zhaoq): Handle uint32 overflow of Stream.id.
 	s := &Stream{
 		id:            t.nextID,
+		done:          make(chan struct{}),
+		goAway:        make(chan struct{}),
 		method:        callHdr.Method,
 		sendCompress:  callHdr.SendCompress,
 		buf:           newRecvBuffer(),
@@ -213,8 +226,9 @@ func (t *http2Client) newStream(ctx context.Context, callHdr *CallHdr) *Stream {
 	// Make a stream be able to cancel the pending operations by itself.
 	s.ctx, s.cancel = context.WithCancel(ctx)
 	s.dec = &recvBufferReader{
-		ctx:  s.ctx,
+		ctx:    s.ctx,
-		recv: s.buf,
+		goAway: s.goAway,
+		recv:   s.buf,
 	}
 	return s
 }
@@ -268,6 +282,10 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 		t.mu.Unlock()
 		return nil, ErrConnClosing
 	}
+	if t.state == draining {
+		t.mu.Unlock()
+		return nil, ErrStreamDrain
+	}
 	if t.state != reachable {
 		t.mu.Unlock()
 		return nil, ErrConnClosing
@@ -275,7 +293,7 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 	checkStreamsQuota := t.streamsQuota != nil
 	t.mu.Unlock()
 	if checkStreamsQuota {
-		sq, err := wait(ctx, t.shutdownChan, t.streamsQuota.acquire())
+		sq, err := wait(ctx, nil, nil, t.shutdownChan, t.streamsQuota.acquire())
 		if err != nil {
 			return nil, err
 		}
@@ -284,7 +302,7 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 			t.streamsQuota.add(sq - 1)
 		}
 	}
-	if _, err := wait(ctx, t.shutdownChan, t.writableChan); err != nil {
+	if _, err := wait(ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil {
 		// Return the quota back now because there is no stream returned to the caller.
 		if _, ok := err.(StreamError); ok && checkStreamsQuota {
 			t.streamsQuota.add(1)
@@ -292,6 +310,15 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 		return nil, err
 	}
 	t.mu.Lock()
+	if t.state == draining {
+		t.mu.Unlock()
+		if checkStreamsQuota {
+			t.streamsQuota.add(1)
+		}
+		// Need to make t writable again so that the rpc in flight can still proceed.
+		t.writableChan <- 0
+		return nil, ErrStreamDrain
+	}
 	if t.state != reachable {
 		t.mu.Unlock()
 		return nil, ErrConnClosing
@@ -326,7 +353,7 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 		t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-encoding", Value: callHdr.SendCompress})
 	}
 	if timeout > 0 {
-		t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-timeout", Value: timeoutEncode(timeout)})
+		t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-timeout", Value: encodeTimeout(timeout)})
 	}
 	for k, v := range authData {
 		// Capital header names are illegal in HTTP/2.
@@ -381,7 +408,7 @@ func (t *http2Client) NewStream(ctx context.Context, callHdr *CallHdr) (_ *Strea
 		}
 		if err != nil {
 			t.notifyError(err)
-			return nil, ConnectionErrorf("transport: %v", err)
+			return nil, ConnectionErrorf(true, err, "transport: %v", err)
 		}
 	}
 	t.writableChan <- 0
@@ -400,22 +427,17 @@ func (t *http2Client) CloseStream(s *Stream, err error) {
 	if t.streamsQuota != nil {
 		updateStreams = true
 	}
-	if t.state == draining && len(t.activeStreams) == 1 {
+	delete(t.activeStreams, s.id)
+	if t.state == draining && len(t.activeStreams) == 0 {
 		// The transport is draining and s is the last live stream on t.
 		t.mu.Unlock()
 		t.Close()
 		return
 	}
-	delete(t.activeStreams, s.id)
 	t.mu.Unlock()
 	if updateStreams {
 		t.streamsQuota.add(1)
 	}
-	// In case stream sending and receiving are invoked in separate
-	// goroutines (e.g., bi-directional streaming), the caller needs
-	// to call cancel on the stream to interrupt the blocking on
-	// other goroutines.
-	s.cancel()
 	s.mu.Lock()
 	if q := s.fc.resetPendingData(); q > 0 {
 		if n := t.fc.onRead(q); n > 0 {
@@ -442,13 +464,13 @@ func (t *http2Client) CloseStream(s *Stream, err error) {
 // accessed any more.
 func (t *http2Client) Close() (err error) {
 	t.mu.Lock()
-	if t.state == reachable {
-		close(t.errorChan)
-	}
 	if t.state == closing {
 		t.mu.Unlock()
 		return
 	}
+	if t.state == reachable || t.state == draining {
+		close(t.errorChan)
+	}
 	t.state = closing
 	t.mu.Unlock()
 	close(t.shutdownChan)
@@ -472,10 +494,35 @@ func (t *http2Client) Close() (err error) {
 
 func (t *http2Client) GracefulClose() error {
 	t.mu.Lock()
-	if t.state == closing {
+	switch t.state {
+	case unreachable:
+		// The server may close the connection concurrently. t is not available for
+		// any streams. Close it now.
+		t.mu.Unlock()
+		t.Close()
+		return nil
+	case closing:
 		t.mu.Unlock()
 		return nil
 	}
+	// Notify the streams which were initiated after the server sent GOAWAY.
+	select {
+	case <-t.goAway:
+		n := t.prevGoAwayID
+		if n == 0 && t.nextID > 1 {
+			n = t.nextID - 2
+		}
+		m := t.goAwayID + 2
+		if m == 2 {
+			m = 1
+		}
+		for i := m; i <= n; i += 2 {
+			if s, ok := t.activeStreams[i]; ok {
+				close(s.goAway)
+			}
+		}
+	default:
+	}
 	if t.state == draining {
 		t.mu.Unlock()
 		return nil
@@ -501,15 +548,15 @@ func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error {
 			size := http2MaxFrameLen
 			s.sendQuotaPool.add(0)
 			// Wait until the stream has some quota to send the data.
-			sq, err := wait(s.ctx, t.shutdownChan, s.sendQuotaPool.acquire())
+			sq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, s.sendQuotaPool.acquire())
 			if err != nil {
 				return err
 			}
 			t.sendQuotaPool.add(0)
 			// Wait until the transport has some quota to send the data.
-			tq, err := wait(s.ctx, t.shutdownChan, t.sendQuotaPool.acquire())
+			tq, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.sendQuotaPool.acquire())
 			if err != nil {
-				if _, ok := err.(StreamError); ok {
+				if _, ok := err.(StreamError); ok || err == io.EOF {
 					t.sendQuotaPool.cancel()
 				}
 				return err
@@ -541,8 +588,8 @@ func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error {
 		// Indicate there is a writer who is about to write a data frame.
 		t.framer.adjustNumWriters(1)
 		// Got some quota. Try to acquire writing privilege on the transport.
-		if _, err := wait(s.ctx, t.shutdownChan, t.writableChan); err != nil {
+		if _, err := wait(s.ctx, s.done, s.goAway, t.shutdownChan, t.writableChan); err != nil {
-			if _, ok := err.(StreamError); ok {
+			if _, ok := err.(StreamError); ok || err == io.EOF {
 				// Return the connection quota back.
 				t.sendQuotaPool.add(len(p))
 			}
@@ -575,7 +622,7 @@ func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error {
 		// invoked.
 		if err := t.framer.writeData(forceFlush, s.id, endStream, p); err != nil {
 			t.notifyError(err)
-			return ConnectionErrorf("transport: %v", err)
+			return ConnectionErrorf(true, err, "transport: %v", err)
 		}
 		if t.framer.adjustNumWriters(-1) == 0 {
 			t.framer.flushWrite()
@@ -590,11 +637,7 @@ func (t *http2Client) Write(s *Stream, data []byte, opts *Options) error {
 	}
 	s.mu.Lock()
 	if s.state != streamDone {
-		if s.state == streamReadDone {
+		s.state = streamWriteDone
-			s.state = streamDone
-		} else {
-			s.state = streamWriteDone
-		}
 	}
 	s.mu.Unlock()
 	return nil
@@ -627,7 +670,7 @@ func (t *http2Client) updateWindow(s *Stream, n uint32) {
 func (t *http2Client) handleData(f *http2.DataFrame) {
 	size := len(f.Data())
 	if err := t.fc.onData(uint32(size)); err != nil {
-		t.notifyError(ConnectionErrorf("%v", err))
+		t.notifyError(ConnectionErrorf(true, err, "%v", err))
 		return
 	}
 	// Select the right stream to dispatch.
@@ -652,6 +695,7 @@ func (t *http2Client) handleData(f *http2.DataFrame) {
 			s.state = streamDone
 			s.statusCode = codes.Internal
 			s.statusDesc = err.Error()
+			close(s.done)
 			s.mu.Unlock()
 			s.write(recvMsg{err: io.EOF})
 			t.controlBuf.put(&resetStream{s.id, http2.ErrCodeFlowControl})
@@ -669,13 +713,14 @@ func (t *http2Client) handleData(f *http2.DataFrame) {
 	// the read direction is closed, and set the status appropriately.
 	if f.FrameHeader.Flags.Has(http2.FlagDataEndStream) {
 		s.mu.Lock()
-		if s.state == streamWriteDone {
+		if s.state == streamDone {
-			s.state = streamDone
+			s.mu.Unlock()
-		} else {
+			return
-			s.state = streamReadDone
 		}
+		s.state = streamDone
 		s.statusCode = codes.Internal
 		s.statusDesc = "server closed the stream without sending trailers"
+		close(s.done)
 		s.mu.Unlock()
 		s.write(recvMsg{err: io.EOF})
 	}
@@ -701,6 +746,8 @@ func (t *http2Client) handleRSTStream(f *http2.RSTStreamFrame) {
 		grpclog.Println("transport: http2Client.handleRSTStream found no mapped gRPC status for the received http2 error ", f.ErrCode)
 		s.statusCode = codes.Unknown
 	}
+	s.statusDesc = fmt.Sprintf("stream terminated by RST_STREAM with error code: %d", f.ErrCode)
+	close(s.done)
 	s.mu.Unlock()
 	s.write(recvMsg{err: io.EOF})
 }
@@ -725,7 +772,32 @@ func (t *http2Client) handlePing(f *http2.PingFrame) {
 }
 
 func (t *http2Client) handleGoAway(f *http2.GoAwayFrame) {
-	// TODO(zhaoq): GoAwayFrame handler to be implemented
+	t.mu.Lock()
+	if t.state == reachable || t.state == draining {
+		if f.LastStreamID > 0 && f.LastStreamID%2 != 1 {
+			t.mu.Unlock()
+			t.notifyError(ConnectionErrorf(true, nil, "received illegal http2 GOAWAY frame: stream ID %d is even", f.LastStreamID))
+			return
+		}
+		select {
+		case <-t.goAway:
+			id := t.goAwayID
+			// t.goAway has been closed (i.e.,multiple GoAways).
+			if id < f.LastStreamID {
+				t.mu.Unlock()
+				t.notifyError(ConnectionErrorf(true, nil, "received illegal http2 GOAWAY frame: previously recv GOAWAY frame with LastStramID %d, currently recv %d", id, f.LastStreamID))
+				return
+			}
+			t.prevGoAwayID = id
+			t.goAwayID = f.LastStreamID
+			t.mu.Unlock()
+			return
+		default:
+		}
+		t.goAwayID = f.LastStreamID
+		close(t.goAway)
+	}
+	t.mu.Unlock()
 }
 
 func (t *http2Client) handleWindowUpdate(f *http2.WindowUpdateFrame) {
@@ -777,11 +849,11 @@ func (t *http2Client) operateHeaders(frame *http2.MetaHeadersFrame) {
 	if len(state.mdata) > 0 {
 		s.trailer = state.mdata
 	}
-	s.state = streamDone
 	s.statusCode = state.statusCode
 	s.statusDesc = state.statusDesc
+	close(s.done)
+	s.state = streamDone
 	s.mu.Unlock()
-
 	s.write(recvMsg{err: io.EOF})
 }
 
@@ -934,13 +1006,22 @@ func (t *http2Client) Error() <-chan struct{} {
 	return t.errorChan
 }
 
+func (t *http2Client) GoAway() <-chan struct{} {
+	return t.goAway
+}
+
 func (t *http2Client) notifyError(err error) {
 	t.mu.Lock()
-	defer t.mu.Unlock()
 	// make sure t.errorChan is closed only once.
+	if t.state == draining {
+		t.mu.Unlock()
+		t.Close()
+		return
+	}
 	if t.state == reachable {
 		t.state = unreachable
 		close(t.errorChan)
 		grpclog.Printf("transport: http2Client.notifyError got notified that the client transport was broken %v.", err)
 	}
+	t.mu.Unlock()
 }

cmd/vendor/google.golang.org/grpc/transport/http2_server.go

@@ -100,18 +100,23 @@ func newHTTP2Server(conn net.Conn, maxStreams uint32, authInfo credentials.AuthI
 	if maxStreams == 0 {
 		maxStreams = math.MaxUint32
 	} else {
-		settings = append(settings, http2.Setting{http2.SettingMaxConcurrentStreams, maxStreams})
+		settings = append(settings, http2.Setting{
+			ID:  http2.SettingMaxConcurrentStreams,
+			Val: maxStreams,
+		})
 	}
 	if initialWindowSize != defaultWindowSize {
-		settings = append(settings, http2.Setting{http2.SettingInitialWindowSize, uint32(initialWindowSize)})
+		settings = append(settings, http2.Setting{
+			ID:  http2.SettingInitialWindowSize,
+			Val: uint32(initialWindowSize)})
 	}
 	if err := framer.writeSettings(true, settings...); err != nil {
-		return nil, ConnectionErrorf("transport: %v", err)
+		return nil, ConnectionErrorf(true, err, "transport: %v", err)
 	}
 	// Adjust the connection flow control window if needed.
 	if delta := uint32(initialConnWindowSize - defaultWindowSize); delta > 0 {
 		if err := framer.writeWindowUpdate(true, 0, delta); err != nil {
-			return nil, ConnectionErrorf("transport: %v", err)
+			return nil, ConnectionErrorf(true, err, "transport: %v", err)
 		}
 	}
 	var buf bytes.Buffer
@@ -137,7 +142,7 @@ func newHTTP2Server(conn net.Conn, maxStreams uint32, authInfo credentials.AuthI
 }
 
 // operateHeader takes action on the decoded headers.
-func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(*Stream)) {
+func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(*Stream)) (close bool) {
 	buf := newRecvBuffer()
 	s := &Stream{
 		id:  frame.Header().StreamID,
@@ -200,6 +205,13 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(
 		t.controlBuf.put(&resetStream{s.id, http2.ErrCodeRefusedStream})
 		return
 	}
+	if s.id%2 != 1 || s.id <= t.maxStreamID {
+		t.mu.Unlock()
+		// illegal gRPC stream id.
+		grpclog.Println("transport: http2Server.HandleStreams received an illegal stream id: ", s.id)
+		return true
+	}
+	t.maxStreamID = s.id
 	s.sendQuotaPool = newQuotaPool(int(t.streamSendQuota))
 	t.activeStreams[s.id] = s
 	t.mu.Unlock()
@@ -207,6 +219,7 @@ func (t *http2Server) operateHeaders(frame *http2.MetaHeadersFrame, handle func(
 		t.updateWindow(s, uint32(n))
 	}
 	handle(s)
+	return
 }
 
 // HandleStreams receives incoming streams using the given handler. This is
@@ -226,6 +239,10 @@ func (t *http2Server) HandleStreams(handle func(*Stream)) {
 	}
 
 	frame, err := t.framer.readFrame()
+	if err == io.EOF || err == io.ErrUnexpectedEOF {
+		t.Close()
+		return
+	}
 	if err != nil {
 		grpclog.Printf("transport: http2Server.HandleStreams failed to read frame: %v", err)
 		t.Close()
@@ -252,20 +269,20 @@ func (t *http2Server) HandleStreams(handle func(*Stream)) {
 				t.controlBuf.put(&resetStream{se.StreamID, se.Code})
 				continue
 			}
+			if err == io.EOF || err == io.ErrUnexpectedEOF {
+				t.Close()
+				return
+			}
+			grpclog.Printf("transport: http2Server.HandleStreams failed to read frame: %v", err)
 			t.Close()
 			return
 		}
 		switch frame := frame.(type) {
 		case *http2.MetaHeadersFrame:
-			id := frame.Header().StreamID
+			if t.operateHeaders(frame, handle) {
-			if id%2 != 1 || id <= t.maxStreamID {
-				// illegal gRPC stream id.
-				grpclog.Println("transport: http2Server.HandleStreams received an illegal stream id: ", id)
 				t.Close()
 				break
 			}
-			t.maxStreamID = id
-			t.operateHeaders(frame, handle)
 		case *http2.DataFrame:
 			t.handleData(frame)
 		case *http2.RSTStreamFrame:
@@ -277,7 +294,7 @@ func (t *http2Server) HandleStreams(handle func(*Stream)) {
 		case *http2.WindowUpdateFrame:
 			t.handleWindowUpdate(frame)
 		case *http2.GoAwayFrame:
-			break
+			// TODO: Handle GoAway from the client appropriately.
 		default:
 			grpclog.Printf("transport: http2Server.HandleStreams found unhandled frame type %v.", frame)
 		}
@@ -359,11 +376,7 @@ func (t *http2Server) handleData(f *http2.DataFrame) {
 		// Received the end of stream from the client.
 		s.mu.Lock()
 		if s.state != streamDone {
-			if s.state == streamWriteDone {
+			s.state = streamReadDone
-				s.state = streamDone
-			} else {
-				s.state = streamReadDone
-			}
 		}
 		s.mu.Unlock()
 		s.write(recvMsg{err: io.EOF})
@@ -435,7 +448,7 @@ func (t *http2Server) writeHeaders(s *Stream, b *bytes.Buffer, endStream bool) e
 		}
 		if err != nil {
 			t.Close()
-			return ConnectionErrorf("transport: %v", err)
+			return ConnectionErrorf(true, err, "transport: %v", err)
 		}
 	}
 	return nil
@@ -450,7 +463,7 @@ func (t *http2Server) WriteHeader(s *Stream, md metadata.MD) error {
 	}
 	s.headerOk = true
 	s.mu.Unlock()
-	if _, err := wait(s.ctx, t.shutdownChan, t.writableChan); err != nil {
+	if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil {
 		return err
 	}
 	t.hBuf.Reset()
@@ -490,7 +503,7 @@ func (t *http2Server) WriteStatus(s *Stream, statusCode codes.Code, statusDesc s
 		headersSent = true
 	}
 	s.mu.Unlock()
-	if _, err := wait(s.ctx, t.shutdownChan, t.writableChan); err != nil {
+	if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil {
 		return err
 	}
 	t.hBuf.Reset()
@@ -503,7 +516,7 @@ func (t *http2Server) WriteStatus(s *Stream, statusCode codes.Code, statusDesc s
 			Name:  "grpc-status",
 			Value: strconv.Itoa(int(statusCode)),
 		})
-	t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-message", Value: statusDesc})
+	t.hEnc.WriteField(hpack.HeaderField{Name: "grpc-message", Value: encodeGrpcMessage(statusDesc)})
 	// Attach the trailer metadata.
 	for k, v := range s.trailer {
 		// Clients don't tolerate reading restricted headers after some non restricted ones were sent.
@@ -539,7 +552,7 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) error {
 	}
 	s.mu.Unlock()
 	if writeHeaderFrame {
-		if _, err := wait(s.ctx, t.shutdownChan, t.writableChan); err != nil {
+		if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil {
 			return err
 		}
 		t.hBuf.Reset()
@@ -555,7 +568,7 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) error {
 		}
 		if err := t.framer.writeHeaders(false, p); err != nil {
 			t.Close()
-			return ConnectionErrorf("transport: %v", err)
+			return ConnectionErrorf(true, err, "transport: %v", err)
 		}
 		t.writableChan <- 0
 	}
@@ -567,13 +580,13 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) error {
 		size := http2MaxFrameLen
 		s.sendQuotaPool.add(0)
 		// Wait until the stream has some quota to send the data.
-		sq, err := wait(s.ctx, t.shutdownChan, s.sendQuotaPool.acquire())
+		sq, err := wait(s.ctx, nil, nil, t.shutdownChan, s.sendQuotaPool.acquire())
 		if err != nil {
 			return err
 		}
 		t.sendQuotaPool.add(0)
 		// Wait until the transport has some quota to send the data.
-		tq, err := wait(s.ctx, t.shutdownChan, t.sendQuotaPool.acquire())
+		tq, err := wait(s.ctx, nil, nil, t.shutdownChan, t.sendQuotaPool.acquire())
 		if err != nil {
 			if _, ok := err.(StreamError); ok {
 				t.sendQuotaPool.cancel()
@@ -599,7 +612,7 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) error {
 		t.framer.adjustNumWriters(1)
 		// Got some quota. Try to acquire writing privilege on the
 		// transport.
-		if _, err := wait(s.ctx, t.shutdownChan, t.writableChan); err != nil {
+		if _, err := wait(s.ctx, nil, nil, t.shutdownChan, t.writableChan); err != nil {
 			if _, ok := err.(StreamError); ok {
 				// Return the connection quota back.
 				t.sendQuotaPool.add(ps)
@@ -629,7 +642,7 @@ func (t *http2Server) Write(s *Stream, data []byte, opts *Options) error {
 		}
 		if err := t.framer.writeData(forceFlush, s.id, false, p); err != nil {
 			t.Close()
-			return ConnectionErrorf("transport: %v", err)
+			return ConnectionErrorf(true, err, "transport: %v", err)
 		}
 		if t.framer.adjustNumWriters(-1) == 0 {
 			t.framer.flushWrite()
@@ -674,6 +687,17 @@ func (t *http2Server) controller() {
 					}
 				case *resetStream:
 					t.framer.writeRSTStream(true, i.streamID, i.code)
+				case *goAway:
+					t.mu.Lock()
+					if t.state == closing {
+						t.mu.Unlock()
+						// The transport is closing.
+						return
+					}
+					sid := t.maxStreamID
+					t.state = draining
+					t.mu.Unlock()
+					t.framer.writeGoAway(true, sid, http2.ErrCodeNo, nil)
 				case *flushIO:
 					t.framer.flushWrite()
 				case *ping:
@@ -719,6 +743,9 @@ func (t *http2Server) Close() (err error) {
 func (t *http2Server) closeStream(s *Stream) {
 	t.mu.Lock()
 	delete(t.activeStreams, s.id)
+	if t.state == draining && len(t.activeStreams) == 0 {
+		defer t.Close()
+	}
 	t.mu.Unlock()
 	// In case stream sending and receiving are invoked in separate
 	// goroutines (e.g., bi-directional streaming), cancel needs to be
@@ -741,3 +768,7 @@ func (t *http2Server) closeStream(s *Stream) {
 func (t *http2Server) RemoteAddr() net.Addr {
 	return t.conn.RemoteAddr()
 }
+
+func (t *http2Server) Drain() {
+	t.controlBuf.put(&goAway{})
+}

cmd/vendor/google.golang.org/grpc/transport/http_util.go

@@ -35,6 +35,7 @@ package transport
 
 import (
 	"bufio"
+	"bytes"
 	"fmt"
 	"io"
 	"net"
@@ -174,11 +175,11 @@ func (d *decodeState) processHeaderField(f hpack.HeaderField) {
 		}
 		d.statusCode = codes.Code(code)
 	case "grpc-message":
-		d.statusDesc = f.Value
+		d.statusDesc = decodeGrpcMessage(f.Value)
 	case "grpc-timeout":
 		d.timeoutSet = true
 		var err error
-		d.timeout, err = timeoutDecode(f.Value)
+		d.timeout, err = decodeTimeout(f.Value)
 		if err != nil {
 			d.setErr(StreamErrorf(codes.Internal, "transport: malformed time-out: %v", err))
 			return
@@ -251,7 +252,7 @@ func div(d, r time.Duration) int64 {
 }
 
 // TODO(zhaoq): It is the simplistic and not bandwidth efficient. Improve it.
-func timeoutEncode(t time.Duration) string {
+func encodeTimeout(t time.Duration) string {
 	if d := div(t, time.Nanosecond); d <= maxTimeoutValue {
 		return strconv.FormatInt(d, 10) + "n"
 	}
@@ -271,7 +272,7 @@ func timeoutEncode(t time.Duration) string {
 	return strconv.FormatInt(div(t, time.Hour), 10) + "H"
 }
 
-func timeoutDecode(s string) (time.Duration, error) {
+func decodeTimeout(s string) (time.Duration, error) {
 	size := len(s)
 	if size < 2 {
 		return 0, fmt.Errorf("transport: timeout string is too short: %q", s)
@@ -288,6 +289,80 @@ func timeoutDecode(s string) (time.Duration, error) {
 	return d * time.Duration(t), nil
 }
 
+const (
+	spaceByte   = ' '
+	tildaByte   = '~'
+	percentByte = '%'
+)
+
+// encodeGrpcMessage is used to encode status code in header field
+// "grpc-message".
+// It checks to see if each individual byte in msg is an
+// allowable byte, and then either percent encoding or passing it through.
+// When percent encoding, the byte is converted into hexadecimal notation
+// with a '%' prepended.
+func encodeGrpcMessage(msg string) string {
+	if msg == "" {
+		return ""
+	}
+	lenMsg := len(msg)
+	for i := 0; i < lenMsg; i++ {
+		c := msg[i]
+		if !(c >= spaceByte && c < tildaByte && c != percentByte) {
+			return encodeGrpcMessageUnchecked(msg)
+		}
+	}
+	return msg
+}
+
+func encodeGrpcMessageUnchecked(msg string) string {
+	var buf bytes.Buffer
+	lenMsg := len(msg)
+	for i := 0; i < lenMsg; i++ {
+		c := msg[i]
+		if c >= spaceByte && c < tildaByte && c != percentByte {
+			buf.WriteByte(c)
+		} else {
+			buf.WriteString(fmt.Sprintf("%%%02X", c))
+		}
+	}
+	return buf.String()
+}
+
+// decodeGrpcMessage decodes the msg encoded by encodeGrpcMessage.
+func decodeGrpcMessage(msg string) string {
+	if msg == "" {
+		return ""
+	}
+	lenMsg := len(msg)
+	for i := 0; i < lenMsg; i++ {
+		if msg[i] == percentByte && i+2 < lenMsg {
+			return decodeGrpcMessageUnchecked(msg)
+		}
+	}
+	return msg
+}
+
+func decodeGrpcMessageUnchecked(msg string) string {
+	var buf bytes.Buffer
+	lenMsg := len(msg)
+	for i := 0; i < lenMsg; i++ {
+		c := msg[i]
+		if c == percentByte && i+2 < lenMsg {
+			parsed, err := strconv.ParseInt(msg[i+1:i+3], 16, 8)
+			if err != nil {
+				buf.WriteByte(c)
+			} else {
+				buf.WriteByte(byte(parsed))
+				i += 2
+			}
+		} else {
+			buf.WriteByte(c)
+		}
+	}
+	return buf.String()
+}
+
 type framer struct {
 	numWriters int32
 	reader     io.Reader

cmd/vendor/google.golang.org/grpc/transport/pre_go16.go

@@ -0,0 +1,51 @@
+// +build !go1.6
+
+/*
+ * Copyright 2016, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ */
+
+package transport
+
+import (
+	"net"
+	"time"
+
+	"golang.org/x/net/context"
+)
+
+// dialContext connects to the address on the named network.
+func dialContext(ctx context.Context, network, address string) (net.Conn, error) {
+	var dialer net.Dialer
+	if deadline, ok := ctx.Deadline(); ok {
+		dialer.Timeout = deadline.Sub(time.Now())
+	}
+	return dialer.Dial(network, address)
+}

cmd/vendor/google.golang.org/grpc/transport/transport.go

@@ -44,7 +44,6 @@ import (
 	"io"
 	"net"
 	"sync"
-	"time"
 
 	"golang.org/x/net/context"
 	"golang.org/x/net/trace"
@@ -120,10 +119,11 @@ func (b *recvBuffer) get() <-chan item {
 // recvBufferReader implements io.Reader interface to read the data from
 // recvBuffer.
 type recvBufferReader struct {
-	ctx  context.Context
+	ctx    context.Context
-	recv *recvBuffer
+	goAway chan struct{}
-	last *bytes.Reader // Stores the remaining data in the previous calls.
+	recv   *recvBuffer
-	err  error
+	last   *bytes.Reader // Stores the remaining data in the previous calls.
+	err    error
 }
 
 // Read reads the next len(p) bytes from last. If last is drained, it tries to
@@ -141,6 +141,8 @@ func (r *recvBufferReader) Read(p []byte) (n int, err error) {
 	select {
 	case <-r.ctx.Done():
 		return 0, ContextErr(r.ctx.Err())
+	case <-r.goAway:
+		return 0, ErrStreamDrain
 	case i := <-r.recv.get():
 		r.recv.load()
 		m := i.(*recvMsg)
@@ -158,7 +160,7 @@ const (
 	streamActive    streamState = iota
 	streamWriteDone             // EndStream sent
 	streamReadDone              // EndStream received
-	streamDone                  // sendDone and recvDone or RSTStreamFrame is sent or received.
+	streamDone                  // the entire stream is finished.
 )
 
 // Stream represents an RPC in the transport layer.
@@ -169,6 +171,10 @@ type Stream struct {
 	// ctx is the associated context of the stream.
 	ctx    context.Context
 	cancel context.CancelFunc
+	// done is closed when the final status arrives.
+	done chan struct{}
+	// goAway is closed when the server sent GoAways signal before this stream was initiated.
+	goAway chan struct{}
 	// method records the associated RPC method of the stream.
 	method       string
 	recvCompress string
@@ -214,6 +220,18 @@ func (s *Stream) SetSendCompress(str string) {
 	s.sendCompress = str
 }
 
+// Done returns a chanel which is closed when it receives the final status
+// from the server.
+func (s *Stream) Done() <-chan struct{} {
+	return s.done
+}
+
+// GoAway returns a channel which is closed when the server sent GoAways signal
+// before this stream was initiated.
+func (s *Stream) GoAway() <-chan struct{} {
+	return s.goAway
+}
+
 // Header acquires the key-value pairs of header metadata once it
 // is available. It blocks until i) the metadata is ready or ii) there is no
 // header metadata or iii) the stream is cancelled/expired.
@@ -221,6 +239,8 @@ func (s *Stream) Header() (metadata.MD, error) {
 	select {
 	case <-s.ctx.Done():
 		return nil, ContextErr(s.ctx.Err())
+	case <-s.goAway:
+		return nil, ErrStreamDrain
 	case <-s.headerChan:
 		return s.header.Copy(), nil
 	}
@@ -335,19 +355,17 @@ type ConnectOptions struct {
 	// UserAgent is the application user agent.
 	UserAgent string
 	// Dialer specifies how to dial a network address.
-	Dialer func(string, time.Duration) (net.Conn, error)
+	Dialer func(context.Context, string) (net.Conn, error)
 	// PerRPCCredentials stores the PerRPCCredentials required to issue RPCs.
 	PerRPCCredentials []credentials.PerRPCCredentials
 	// TransportCredentials stores the Authenticator required to setup a client connection.
 	TransportCredentials credentials.TransportCredentials
-	// Timeout specifies the timeout for dialing a ClientTransport.
-	Timeout time.Duration
 }
 
 // NewClientTransport establishes the transport with the required ConnectOptions
 // and returns it to the caller.
-func NewClientTransport(target string, opts *ConnectOptions) (ClientTransport, error) {
+func NewClientTransport(ctx context.Context, target string, opts ConnectOptions) (ClientTransport, error) {
-	return newHTTP2Client(target, opts)
+	return newHTTP2Client(ctx, target, opts)
 }
 
 // Options provides additional hints and information for message
@@ -417,6 +435,11 @@ type ClientTransport interface {
 	// and create a new one) in error case. It should not return nil
 	// once the transport is initiated.
 	Error() <-chan struct{}
+
+	// GoAway returns a channel that is closed when ClientTranspor
+	// receives the draining signal from the server (e.g., GOAWAY frame in
+	// HTTP/2).
+	GoAway() <-chan struct{}
 }
 
 // ServerTransport is the common interface for all gRPC server-side transport
@@ -448,6 +471,9 @@ type ServerTransport interface {
 
 	// RemoteAddr returns the remote network address.
 	RemoteAddr() net.Addr
+
+	// Drain notifies the client this ServerTransport stops accepting new RPCs.
+	Drain()
 }
 
 // StreamErrorf creates an StreamError with the specified error code and description.
@@ -459,9 +485,11 @@ func StreamErrorf(c codes.Code, format string, a ...interface{}) StreamError {
 }
 
 // ConnectionErrorf creates an ConnectionError with the specified error description.
-func ConnectionErrorf(format string, a ...interface{}) ConnectionError {
+func ConnectionErrorf(temp bool, e error, format string, a ...interface{}) ConnectionError {
 	return ConnectionError{
 		Desc: fmt.Sprintf(format, a...),
+		temp: temp,
+		err:  e,
 	}
 }
 
@@ -469,14 +497,36 @@ func ConnectionErrorf(format string, a ...interface{}) ConnectionError {
 // entire connection and the retry of all the active streams.
 type ConnectionError struct {
 	Desc string
+	temp bool
+	err  error
 }
 
 func (e ConnectionError) Error() string {
 	return fmt.Sprintf("connection error: desc = %q", e.Desc)
 }
 
-// ErrConnClosing indicates that the transport is closing.
+// Temporary indicates if this connection error is temporary or fatal.
-var ErrConnClosing = ConnectionError{Desc: "transport is closing"}
+func (e ConnectionError) Temporary() bool {
+	return e.temp
+}
+
+// Origin returns the original error of this connection error.
+func (e ConnectionError) Origin() error {
+	// Never return nil error here.
+	// If the original error is nil, return itself.
+	if e.err == nil {
+		return e
+	}
+	return e.err
+}
+
+var (
+	// ErrConnClosing indicates that the transport is closing.
+	ErrConnClosing = ConnectionError{Desc: "transport is closing", temp: true}
+	// ErrStreamDrain indicates that the stream is rejected by the server because
+	// the server stops accepting new RPCs.
+	ErrStreamDrain = StreamErrorf(codes.Unavailable, "the server stops accepting new RPCs")
+)
 
 // StreamError is an error that only affects one stream within a connection.
 type StreamError struct {
@@ -501,12 +551,25 @@ func ContextErr(err error) StreamError {
 
 // wait blocks until it can receive from ctx.Done, closing, or proceed.
 // If it receives from ctx.Done, it returns 0, the StreamError for ctx.Err.
+// If it receives from done, it returns 0, io.EOF if ctx is not done; otherwise
+// it return the StreamError for ctx.Err.
+// If it receives from goAway, it returns 0, ErrStreamDrain.
 // If it receives from closing, it returns 0, ErrConnClosing.
 // If it receives from proceed, it returns the received integer, nil.
-func wait(ctx context.Context, closing <-chan struct{}, proceed <-chan int) (int, error) {
+func wait(ctx context.Context, done, goAway, closing <-chan struct{}, proceed <-chan int) (int, error) {
 	select {
 	case <-ctx.Done():
 		return 0, ContextErr(ctx.Err())
+	case <-done:
+		// User cancellation has precedence.
+		select {
+		case <-ctx.Done():
+			return 0, ContextErr(ctx.Err())
+		default:
+		}
+		return 0, io.EOF
+	case <-goAway:
+		return 0, ErrStreamDrain
 	case <-closing:
 		return 0, ErrConnClosing
 	case i := <-proceed:

discovery/srv.go

@@ -53,8 +53,8 @@ func SRVGetCluster(name, dns string, defaultToken string, apurls types.URLs) (st
 			return err
 		}
 		for _, srv := range addrs {
-			target := strings.TrimSuffix(srv.Target, ".")
+			port := fmt.Sprintf("%d", srv.Port)
-			host := net.JoinHostPort(target, fmt.Sprintf("%d", srv.Port))
+			host := net.JoinHostPort(srv.Target, port)
 			tcpAddr, err := resolveTCPAddr("tcp", host)
 			if err != nil {
 				plog.Warningf("couldn't resolve host %s during SRV discovery", host)
@@ -70,8 +70,11 @@ func SRVGetCluster(name, dns string, defaultToken string, apurls types.URLs) (st
 				n = fmt.Sprintf("%d", tempName)
 				tempName += 1
 			}
-			stringParts = append(stringParts, fmt.Sprintf("%s=%s%s", n, prefix, host))
+			// SRV records have a trailing dot but URL shouldn't.
-			plog.Noticef("got bootstrap from DNS for %s at %s%s", service, prefix, host)
+			shortHost := strings.TrimSuffix(srv.Target, ".")
+			urlHost := net.JoinHostPort(shortHost, port)
+			stringParts = append(stringParts, fmt.Sprintf("%s=%s%s", n, prefix, urlHost))
+			plog.Noticef("got bootstrap from DNS for %s at %s%s", service, prefix, urlHost)
 		}
 		return nil
 	}

discovery/srv_test.go

@@ -17,6 +17,7 @@ package discovery
 import (
 	"errors"
 	"net"
+	"strings"
 	"testing"
 
 	"github.com/coreos/etcd/pkg/testutil"
@@ -29,11 +30,22 @@ func TestSRVGetCluster(t *testing.T) {
 	}()
 
 	name := "dnsClusterTest"
+	dns := map[string]string{
+		"1.example.com.:2480": "10.0.0.1:2480",
+		"2.example.com.:2480": "10.0.0.2:2480",
+		"3.example.com.:2480": "10.0.0.3:2480",
+		"4.example.com.:2380": "10.0.0.3:2380",
+	}
+	srvAll := []*net.SRV{
+		{Target: "1.example.com.", Port: 2480},
+		{Target: "2.example.com.", Port: 2480},
+		{Target: "3.example.com.", Port: 2480},
+	}
+
 	tests := []struct {
 		withSSL    []*net.SRV
 		withoutSSL []*net.SRV
 		urls       []string
-		dns        map[string]string
 
 		expected string
 	}{
@@ -41,61 +53,50 @@ func TestSRVGetCluster(t *testing.T) {
 			[]*net.SRV{},
 			[]*net.SRV{},
 			nil,
-			nil,
 
 			"",
 		},
 		{
-			[]*net.SRV{
+			srvAll,
-				{Target: "10.0.0.1", Port: 2480},
-				{Target: "10.0.0.2", Port: 2480},
-				{Target: "10.0.0.3", Port: 2480},
-			},
 			[]*net.SRV{},
 			nil,
+
+			"0=https://1.example.com:2480,1=https://2.example.com:2480,2=https://3.example.com:2480",
+		},
+		{
+			srvAll,
+			[]*net.SRV{{Target: "4.example.com.", Port: 2380}},
 			nil,
 
-			"0=https://10.0.0.1:2480,1=https://10.0.0.2:2480,2=https://10.0.0.3:2480",
+			"0=https://1.example.com:2480,1=https://2.example.com:2480,2=https://3.example.com:2480,3=http://4.example.com:2380",
 		},
 		{
-			[]*net.SRV{
+			srvAll,
-				{Target: "10.0.0.1", Port: 2480},
+			[]*net.SRV{{Target: "4.example.com.", Port: 2380}},
-				{Target: "10.0.0.2", Port: 2480},
-				{Target: "10.0.0.3", Port: 2480},
-			},
-			[]*net.SRV{
-				{Target: "10.0.0.1", Port: 2380},
-			},
-			nil,
-			nil,
-			"0=https://10.0.0.1:2480,1=https://10.0.0.2:2480,2=https://10.0.0.3:2480,3=http://10.0.0.1:2380",
-		},
-		{
-			[]*net.SRV{
-				{Target: "10.0.0.1", Port: 2480},
-				{Target: "10.0.0.2", Port: 2480},
-				{Target: "10.0.0.3", Port: 2480},
-			},
-			[]*net.SRV{
-				{Target: "10.0.0.1", Port: 2380},
-			},
 			[]string{"https://10.0.0.1:2480"},
-			nil,
+
-			"dnsClusterTest=https://10.0.0.1:2480,0=https://10.0.0.2:2480,1=https://10.0.0.3:2480,2=http://10.0.0.1:2380",
+			"dnsClusterTest=https://1.example.com:2480,0=https://2.example.com:2480,1=https://3.example.com:2480,2=http://4.example.com:2380",
 		},
 		// matching local member with resolved addr and return unresolved hostnames
 		{
-			[]*net.SRV{
+			srvAll,
-				{Target: "1.example.com.", Port: 2480},
-				{Target: "2.example.com.", Port: 2480},
-				{Target: "3.example.com.", Port: 2480},
-			},
 			nil,
 			[]string{"https://10.0.0.1:2480"},
-			map[string]string{"1.example.com:2480": "10.0.0.1:2480", "2.example.com:2480": "10.0.0.2:2480", "3.example.com:2480": "10.0.0.3:2480"},
 
 			"dnsClusterTest=https://1.example.com:2480,0=https://2.example.com:2480,1=https://3.example.com:2480",
 		},
+		// invalid
+	}
+
+	resolveTCPAddr = func(network, addr string) (*net.TCPAddr, error) {
+		if strings.Contains(addr, "10.0.0.") {
+			// accept IP addresses when resolving apurls
+			return net.ResolveTCPAddr(network, addr)
+		}
+		if dns[addr] == "" {
+			return nil, errors.New("missing dns record")
+		}
+		return net.ResolveTCPAddr(network, dns[addr])
 	}
 
 	for i, tt := range tests {
@@ -108,12 +109,6 @@ func TestSRVGetCluster(t *testing.T) {
 			}
 			return "", nil, errors.New("Unknown service in mock")
 		}
-		resolveTCPAddr = func(network, addr string) (*net.TCPAddr, error) {
-			if tt.dns == nil || tt.dns[addr] == "" {
-				return net.ResolveTCPAddr(network, addr)
-			}
-			return net.ResolveTCPAddr(network, tt.dns[addr])
-		}
 		urls := testutil.MustNewURLs(t, tt.urls)
 		str, token, err := SRVGetCluster(name, "example.com", "token", urls)
 		if err != nil {

e2e/ctl_v2_test.go

@@ -280,6 +280,42 @@ func TestCtlV2Backup(t *testing.T) { // For https://github.com/coreos/etcd/issue
 	}
 }
 
+func TestCtlV2AuthWithCommonName(t *testing.T) {
+	defer testutil.AfterTest(t)
+
+	copiedCfg := configClientTLS
+	copiedCfg.clientCertAuthEnabled = true
+
+	epc := setupEtcdctlTest(t, &copiedCfg, false)
+	defer func() {
+		if err := epc.Close(); err != nil {
+			t.Fatalf("error closing etcd processes (%v)", err)
+		}
+	}()
+
+	if err := etcdctlRoleAdd(epc, "testrole"); err != nil {
+		t.Fatalf("failed to add role (%v)", err)
+	}
+	if err := etcdctlRoleGrant(epc, "testrole", "--rw", "--path=/foo"); err != nil {
+		t.Fatalf("failed to grant role (%v)", err)
+	}
+	if err := etcdctlUserAdd(epc, "root", "123"); err != nil {
+		t.Fatalf("failed to add user (%v)", err)
+	}
+	if err := etcdctlUserAdd(epc, "Autogenerated CA", "123"); err != nil {
+		t.Fatalf("failed to add user (%v)", err)
+	}
+	if err := etcdctlUserGrant(epc, "Autogenerated CA", "testrole"); err != nil {
+		t.Fatalf("failed to grant role (%v)", err)
+	}
+	if err := etcdctlAuthEnable(epc); err != nil {
+		t.Fatalf("failed to enable auth (%v)", err)
+	}
+	if err := etcdctlSet(epc, "foo", "bar"); err != nil {
+		t.Fatalf("failed to write (%v)", err)
+	}
+}
+
 func etcdctlPrefixArgs(clus *etcdProcessCluster) []string {
 	endpoints := ""
 	if proxies := clus.proxies(); len(proxies) != 0 {
@@ -352,6 +388,13 @@ func etcdctlRoleAdd(clus *etcdProcessCluster, role string) error {
 	return spawnWithExpect(cmdArgs, role)
 }
 
+func etcdctlRoleGrant(clus *etcdProcessCluster, role string, perms ...string) error {
+	cmdArgs := append(etcdctlPrefixArgs(clus), "role", "grant")
+	cmdArgs = append(cmdArgs, perms...)
+	cmdArgs = append(cmdArgs, role)
+	return spawnWithExpect(cmdArgs, role)
+}
+
 func etcdctlRoleList(clus *etcdProcessCluster, expectedRole string) error {
 	cmdArgs := append(etcdctlPrefixArgs(clus), "role", "list")
 	return spawnWithExpect(cmdArgs, expectedRole)