maintainer: fix panic when maintainer into bootstrap after removing maintainer (#3222)

hongyunyan · hongyunyan · commit 1e7caec4b23b · 2025-11-20T15:20:11.000+08:00
close #3221
diff --git a/maintainer/maintainer.go b/maintainer/maintainer.go
@@ -888,7 +888,11 @@ func newDDLSpan(keyspaceID uint32, cfID common.ChangeFeedID, checkpointTs uint64
 }
 
 func (m *Maintainer) onBootstrapDone(cachedResp map[node.ID]*heartbeatpb.MaintainerBootstrapResponse) {
-	if cachedResp == nil {
+	// calCheckpointTs() skips advancing checkpoint when bootstrapped is false, so it won't call
+	// onNodeChanged() before FinishBootstrap succeeds. All other callers of onNodeChanged() and
+	// onMaintainerBootstrapResponse() run in the same event loop goroutine as onRemoveMaintainer(),
+	// hence guarding with m.removing is sufficient to avoid accessing a removed DDL span leading to panic.
+	if cachedResp == nil || m.removing.Load() {
 		return
 	}
 	isMySQLSinkCompatible, err := isMysqlCompatible(m.info.SinkURI)

Original file line number	Diff line number	Diff line change
`@@ -888,7 +888,11 @@ func newDDLSpan(keyspaceID uint32, cfID common.ChangeFeedID, checkpointTs uint64`
`888`	`888`	`}`
`889`	`889`
`890`	`890`	`func (m Maintainer) onBootstrapDone(cachedResp map[node.ID]heartbeatpb.MaintainerBootstrapResponse) {`
`891`		`- if cachedResp == nil {`
	`891`	`+ // calCheckpointTs() skips advancing checkpoint when bootstrapped is false, so it won't call`
	`892`	`+ // onNodeChanged() before FinishBootstrap succeeds. All other callers of onNodeChanged() and`
	`893`	`+ // onMaintainerBootstrapResponse() run in the same event loop goroutine as onRemoveMaintainer(),`
	`894`	`+ // hence guarding with m.removing is sufficient to avoid accessing a removed DDL span leading to panic.`
	`895`	`+ if cachedResp == nil \|\| m.removing.Load() {`
`892`	`896`	`return`
`893`	`897`	`}`
`894`	`898`	`isMySQLSinkCompatible, err := isMysqlCompatible(m.info.SinkURI)`