The grpc-proxy test logic was assuming that the context associated to client is closed,
while in practice all tests called client.Close() without explicit context close.
The current testing strategy is complicated 2 fold:
- grpc proxy works like man-in-the middle of each Connection issues
from integration tests and its lifetime is bound to the connection.
- both connections (client -> proxy, and proxy -> etcd-server) are
represented by the same ClientV3 object instance (with substituted
implementations of KV or watcher).
The fix splits context representing proxy from context representing proxy -> etcd-server connection,
thus allowing cancelation of the proxy context.
Currently, the grpc gateway is enabled by default when run the etcd server
using command line configuration. However, when provide config through
config file, the grpc gateway is disabled by default. We should either use
the same approach or at least document this.
related #https://github.com/etcd-io/etcd/issues/12093
This CL fixes:
COVERDIR=./coverage PASSES="build_cov" && go test --tags cov -v ./tests/e2e
and is part of the effort to make:
COVERDIR=coverage PASSES="build_cov cov" ./test
fully pass.
The args passed to ./bin/etcd_test and ./bin/etcdctl_test binaries were
mismatched. The protocol of passing the arguments using
environment variables has been replaces with proper passing of flags.
How the measurement of coverage by e2e tests works:
1. COVERDIR=./coverage PASSES="build_cov" are generating
./bin/etcd_test and ./bin/etcdctl_test binaries.
2. These binaries are tests (as coverage can be computed only for
tests) [see ./main_test.go ./etcdctl/main_test.go], but this tests are
running the main logic of the server and uppon termination (or SIGTERM
signal) are writting proper .coverprofile files in the $COVERDIR folder.
The binaries used to take arguments using env variables, but its not
needed any longer. The binaries can consume any command line arguments
that either test (so --test.fooo) or the original binary can consume.
3. The tests/e2e (when compiled with the --tags cov) are starting the
_test binaries instead of the original binaries, such that the coverage
is being collected.
We have following communication schema:
client --- 1 ---> grpc-proxy --- 2 --- > etcd-server
There are 2 sets of flags/certs in grpc proxy [ https://github.com/etcd-io/etcd/blob/master/etcdmain/grpc_proxy.go#L140 ]:
A. (cert-file, key-file, trusted-ca-file, auto-tls) this are controlling [1] so client to proxy connection and in particular they are describing proxy public identity.
B. (cert,key, cacert ) - these are controlling [2] so what's the identity that proxy uses to make connections to the etcd-server.
If 2 (B.) contains certificate with CN and etcd-server is running with --client-cert-auth=true, the CN can be used as identity of 'client' from service perspective. This is permission escalation, that we should forbid.
If 1 (A.) contains certificate with CN - it should be considered perfectly valid. The server can (should) have full identity.
So only --cert flag (and not --cert-file flag) should be validated for empty CN.
os.MkdirAll creates directory before umask so make sure that a desired
permission is set after creating a directory with MkdirAll. Use the
existing TouchDirAll function which checks for permission if dir is already
exist and when create a new dir.
This makes it possible to run an etcd node for testing and development
without placing lots of load on the file system.
Fixes#11930.
Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
This change makes the etcd package compatible with the existing Go
ecosystem for module versioning.
Used this tool to update package imports:
https://github.com/KSubedi/gomove
Currently grpc-proxy doesn't config keep alive related options, so it
will use the default values provided by the underlay gprc library. If
clients uses a keep alive ping interval smaller than server's default
minTime, connections between server and clients will be closed and
reopened frequently.
To make it easier to root-cause when /health check fails.
For example, we are using load balancer to health check
each etcd instance, and when one etcd node gets terminated,
it's hard to tell whether etcd "server" was really failing
or client (or load balancer") failed to reach the etcd cluster
which is also failure in load balancer health check.
Signed-off-by: Gyuho Lee <leegyuho@amazon.com>
Version flag is meant for displaying version and it's a boolean.
It's for an internal use, and we may not want users to set
it in the environment.
Fixes # 11210
Piotr Tabor