Use fresh shard state in offline node repair

yandex · Jan 8, 2024 · c912f52 · c912f52
1 parent b3001c6
commit c912f52
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 3 deletions.
diff --git a/internal/app/candidate.go b/internal/app/candidate.go
@@ -52,7 +52,7 @@ func (app *App) stateCandidate() appState {
 		app.logger.Error("Candidate: failed to get current master from DCS", "error", err)
 		return stateCandidate
 	}
-	app.repairLocalNode(shardState, master)
+	app.repairLocalNode(master)
 
 	if app.dcs.AcquireLock(pathManagerLock) {
 		return stateManager

diff --git a/internal/app/manager.go b/internal/app/manager.go
@@ -65,7 +65,7 @@ func (app *App) stateManager() appState {
 		return stateMaintenance
 	}
 
-	app.repairLocalNode(shardState, master)
+	app.repairLocalNode(master)
 
 	var switchover Switchover
 	if err := app.dcs.Get(pathCurrentSwitch, &switchover); err == nil {

diff --git a/internal/app/repair.go b/internal/app/repair.go
@@ -107,7 +107,7 @@ func (app *App) repairReplica(node *redis.Node, masterState, state *HostState, m
 	}
 }
 
-func (app *App) repairLocalNode(shardState map[string]*HostState, master string) {
+func (app *App) repairLocalNode(master string) {
 	local := app.shard.Local()
 
 	offline, err := local.IsOffline(app.ctx)
@@ -132,6 +132,11 @@ func (app *App) repairLocalNode(shardState map[string]*HostState, master string)
 		return
 	}
 
+	shardState, err := app.getShardStateFromDB()
+	if err != nil {
+		app.logger.Error("Local repair: unable to get actual shard state", "error", err)
+		return
+	}
 	state, ok := shardState[local.FQDN()]
 	if !ok {
 		app.logger.Error("Local repair: unable to find local node in shard state")