fix(master): stop goraft server on shutdown and bump raft to v1.2.1 (#11284)

MasterServer.Shutdown only stopped the Hashicorp raft implementation;
when using the default goraft backend, the raft event-loop goroutine
(leaderLoop/followerLoop) kept running after the master shut down. In
the in-process test harness this leaked goroutines across sequential
test runs, and a stale event occasionally reached a leader at term 0
and tripped the goraft "leader.elected.at.same.term" assertion,
crashing the whole test binary (CI run 34670959967, PR 11279).

Stop the goraft server in Shutdown() so its goroutines exit cleanly,
and bump seaweedfs/raft to v1.2.1 which replaces that assertion with a
graceful step-down to Follower instead of a panic.
This commit is contained in:
Chris Lu
2026-09-11 22:25:05 -07:00
committed by GitHub
parent 5a0e017457
commit c46f82d29a
3 changed files with 21 additions and 7 deletions
+1 -1
View File
@@ -65,7 +65,7 @@ require (
github.com/rcrowley/go-metrics v0.0.0-20201227073835-cf1acfcdf475 // indirect
github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
github.com/seaweedfs/goexif v2.0.0+incompatible
github.com/seaweedfs/raft v1.2.0
github.com/seaweedfs/raft v1.2.1
github.com/sirupsen/logrus v1.9.4 // indirect
github.com/spf13/afero v1.15.0 // indirect
github.com/spf13/cast v1.10.0 // indirect
+2 -2
View File
@@ -1821,8 +1821,8 @@ github.com/seaweedfs/go-fuse/v2 v2.9.4 h1:ACyloiuopdhRSjdLLeSWbsVaemMPskORaRF01T
github.com/seaweedfs/go-fuse/v2 v2.9.4/go.mod h1:zABdmWEa6A0bwaBeEOBUeUkGIZlxUhcdv+V1Dcc/U/I=
github.com/seaweedfs/goexif v2.0.0+incompatible h1:x8pckiT12QQhifwhDQpeISgDfsqmQ6VR4LFPQ64JRps=
github.com/seaweedfs/goexif v2.0.0+incompatible/go.mod h1:Oni780Z236sXpIQzk1XoJlTwqrJ02smEin9zQeff7Fk=
github.com/seaweedfs/raft v1.2.0 h1:Ez4Hw9ifBbTT7wg54DvGHBjw1vRlTb4roH0TKl0Oj9Y=
github.com/seaweedfs/raft v1.2.0/go.mod h1:fgs/rAVEzjQ7e04XMzG3eJhwZZRmBW+2uRtjakeCGeU=
github.com/seaweedfs/raft v1.2.1 h1:QgFl/aaPnagpUxYB6Bx+fFss1NyetVVcJmraMmnaQ5Q=
github.com/seaweedfs/raft v1.2.1/go.mod h1:fgs/rAVEzjQ7e04XMzG3eJhwZZRmBW+2uRtjakeCGeU=
github.com/secure-systems-lab/go-securesystemslib v0.11.0 h1:iuCR9kcMFD4QurdKrGvPLoKZLv9YvwPYVr0473BdtFs=
github.com/secure-systems-lab/go-securesystemslib v0.11.0/go.mod h1:+PMOTjUGwHj2vcZ+TFKlb1tXRbrdWE1LYDT5i9JC80Q=
github.com/sergi/go-diff v1.0.0/go.mod h1:0CfEIISq7TuYL3j771MWULgwwjU+GofnZX9QAmXWZgo=
+18 -4
View File
@@ -621,13 +621,27 @@ func (ms *MasterServer) missingRaftPeerName(peerAddress pb.ServerAddress) (strin
}
func (ms *MasterServer) Shutdown() {
if ms.Topo == nil || ms.Topo.HashicorpRaft == nil {
if ms.Topo == nil {
return
}
if ms.Topo.HashicorpRaft.State() == hashicorpRaft.Leader {
ms.Topo.HashicorpRaft.LeadershipTransfer()
ms.Topo.RaftServerAccessLock.RLock()
raftServer := ms.Topo.RaftServer
hashRaft := ms.Topo.HashicorpRaft
ms.Topo.RaftServerAccessLock.RUnlock()
if hashRaft != nil {
if hashRaft.State() == hashicorpRaft.Leader {
hashRaft.LeadershipTransfer()
}
hashRaft.Shutdown()
}
// Stop the goraft server too. Without this, the raft event loop
// goroutine (leaderLoop/followerLoop) keeps running after the master
// shuts down, leaking goroutines across test runs and occasionally
// processing stale events that trigger safety assertions.
if raftServer != nil {
raftServer.Stop()
}
ms.Topo.HashicorpRaft.Shutdown()
}
func (ms *MasterServer) Reload() {