Merge pull request #616 from xiangli-cmu/add_snapshot_post
fix(server/transporter.go) fix the short timeout problem of sending snapshotrelease-0.4
commit
c8e65c1315
|
@ -15,6 +15,10 @@ import (
|
||||||
"github.com/coreos/etcd/log"
|
"github.com/coreos/etcd/log"
|
||||||
)
|
)
|
||||||
|
|
||||||
|
const (
|
||||||
|
snapshotTimeout = time.Second * 120
|
||||||
|
)
|
||||||
|
|
||||||
// Transporter layer for communication between raft nodes
|
// Transporter layer for communication between raft nodes
|
||||||
type transporter struct {
|
type transporter struct {
|
||||||
requestTimeout time.Duration
|
requestTimeout time.Duration
|
||||||
|
@ -24,6 +28,8 @@ type transporter struct {
|
||||||
|
|
||||||
client *http.Client
|
client *http.Client
|
||||||
transport *httpclient.Transport
|
transport *httpclient.Transport
|
||||||
|
snapshotClient *http.Client
|
||||||
|
snapshotTransport *httpclient.Transport
|
||||||
}
|
}
|
||||||
|
|
||||||
type dialer func(network, addr string) (net.Conn, error)
|
type dialer func(network, addr string) (net.Conn, error)
|
||||||
|
@ -43,9 +49,23 @@ func NewTransporter(followersStats *raftFollowersStats, serverStats *raftServerS
|
||||||
ReadWriteTimeout: responseHeaderTimeout,
|
ReadWriteTimeout: responseHeaderTimeout,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// Sending snapshot might take a long time so we use a different HTTP transporter
|
||||||
|
// Timeout is set to 120s (Around 100MB if the bandwidth is 10Mbits/s)
|
||||||
|
// This timeout is not calculated by heartbeat time.
|
||||||
|
// TODO(xiangl1) we can actually calculate the max bandwidth if we know
|
||||||
|
// average RTT.
|
||||||
|
// It should be equal to (TCP max window size/RTT).
|
||||||
|
sTr := &httpclient.Transport{
|
||||||
|
ConnectTimeout: dialTimeout,
|
||||||
|
RequestTimeout: snapshotTimeout,
|
||||||
|
ReadWriteTimeout: snapshotTimeout,
|
||||||
|
}
|
||||||
|
|
||||||
t := transporter{
|
t := transporter{
|
||||||
client: &http.Client{Transport: tr},
|
client: &http.Client{Transport: tr},
|
||||||
transport: tr,
|
transport: tr,
|
||||||
|
snapshotClient: &http.Client{Transport: sTr},
|
||||||
|
snapshotTransport: sTr,
|
||||||
requestTimeout: requestTimeout,
|
requestTimeout: requestTimeout,
|
||||||
followersStats: followersStats,
|
followersStats: followersStats,
|
||||||
serverStats: serverStats,
|
serverStats: serverStats,
|
||||||
|
@ -58,6 +78,9 @@ func NewTransporter(followersStats *raftFollowersStats, serverStats *raftServerS
|
||||||
func (t *transporter) SetTLSConfig(tlsConf tls.Config) {
|
func (t *transporter) SetTLSConfig(tlsConf tls.Config) {
|
||||||
t.transport.TLSClientConfig = &tlsConf
|
t.transport.TLSClientConfig = &tlsConf
|
||||||
t.transport.DisableCompression = true
|
t.transport.DisableCompression = true
|
||||||
|
|
||||||
|
t.snapshotTransport.TLSClientConfig = &tlsConf
|
||||||
|
t.snapshotTransport.DisableCompression = true
|
||||||
}
|
}
|
||||||
|
|
||||||
// Sends AppendEntries RPCs to a peer when the server is the leader.
|
// Sends AppendEntries RPCs to a peer when the server is the leader.
|
||||||
|
@ -191,7 +214,7 @@ func (t *transporter) SendSnapshotRecoveryRequest(server raft.Server, peer *raft
|
||||||
u, _ := t.registry.PeerURL(peer.Name)
|
u, _ := t.registry.PeerURL(peer.Name)
|
||||||
log.Debugf("Send Snapshot Recovery from %s to %s", server.Name(), u)
|
log.Debugf("Send Snapshot Recovery from %s to %s", server.Name(), u)
|
||||||
|
|
||||||
resp, _, err := t.Post(fmt.Sprintf("%s/snapshotRecovery", u), &b)
|
resp, err := t.PostSnapshot(fmt.Sprintf("%s/snapshotRecovery", u), &b)
|
||||||
|
|
||||||
if err != nil {
|
if err != nil {
|
||||||
log.Debugf("Cannot send Snapshot Recovery to %s : %s", u, err)
|
log.Debugf("Cannot send Snapshot Recovery to %s : %s", u, err)
|
||||||
|
@ -224,3 +247,9 @@ func (t *transporter) Get(urlStr string) (*http.Response, *http.Request, error)
|
||||||
resp, err := t.client.Do(req)
|
resp, err := t.client.Do(req)
|
||||||
return resp, req, err
|
return resp, req, err
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// PostSnapshot posts a json format snapshot to the given url
|
||||||
|
// The underlying HTTP transport has a minute level timeout
|
||||||
|
func (t *transporter) PostSnapshot(url string, body io.Reader) (*http.Response, error) {
|
||||||
|
return t.snapshotClient.Post(url, "application/json", body)
|
||||||
|
}
|
||||||
|
|
Loading…
Reference in New Issue