fix: enabling semisync; feat: errno in logs (#103)

* fix: enabling semisync; feat: errno in logs * fix: issues * fix: issues
yandex · May 27, 2024 · 8173179 · 8173179
1 parent d74946f
commit 8173179
Show file tree

Hide file tree

Showing 3 changed files with 44 additions and 5 deletions.
diff --git a/internal/app/app.go b/internal/app/app.go
@@ -320,7 +320,17 @@ func (app *App) checkRecovery() {
 	app.logger.Infof("recovery: local node %s has GTIDs %s", localNode.Host(), sstatus.GetExecutedGtidSet())
 
 	if isSlavePermanentlyLost(sstatus, mgtids) {
-		app.logger.Errorf("recovery: local node %s is NOT behind the master %s, need RESETUP", localNode.Host(), masterNode)
+		rp, err := localNode.GetReplicaStatus()
+		if err == nil {
+			if rp.GetLastError() != "" {
+				app.logger.Errorf("recovery: local node %s has error: %s", localNode.Host(), rp.GetLastError())
+			}
+			if rp.GetLastIOError() != "" {
+				app.logger.Errorf("recovery: local node %s has IO error: %s", localNode.Host(), rp.GetLastIOError())
+			}
+		} else {
+			app.logger.Errorf("recovery: local node %s is NOT behind the master %s, need RESETUP", localNode.Host(), masterNode)
+		}
 		app.writeResetupFile("")
 	} else {
 		readOnly, _, err := localNode.IsReadOnly()
@@ -1067,9 +1077,9 @@ func (app *App) enableSemiSyncOnSlave(host string) error {
 		app.logger.Errorf("failed to enable semi_sync_slave on %s: %s", host, err)
 		return err
 	}
-	err = node.RestartSlaveIOThread()
+	err = node.RestartReplica()
 	if err != nil {
-		app.logger.Errorf("failed restart slave io thread after set semi_sync_slave on %s: %s", host, err)
+		app.logger.Errorf("failed restart replication after set semi_sync_slave on %s: %s", host, err)
 		return err
 	}
 	return nil

diff --git a/internal/mysql/data.go b/internal/mysql/data.go
@@ -68,9 +68,11 @@ type SlaveStatusStruct struct {
 	ReadMasterLogPos int64           `db:"Read_Master_Log_Pos"`
 	SlaveIORunning   string          `db:"Slave_IO_Running"`
 	SlaveSQLRunning  string          `db:"Slave_SQL_Running"`
+	LastError        string          `db:"Last_Error"`
 	RetrievedGtidSet string          `db:"Retrieved_Gtid_Set"`
 	ExecutedGtidSet  string          `db:"Executed_Gtid_Set"`
 	LastIOErrno      int             `db:"Last_IO_Errno"`
+	LastIOError      string          `db:"Last_IO_Error"`
 	LastSQLErrno     int             `db:"Last_SQL_Errno"`
 	Lag              sql.NullFloat64 `db:"Seconds_Behind_Master"`
 }
@@ -83,9 +85,11 @@ type ReplicaStatusStruct struct {
 	ReadSourceLogPos  int64           `db:"Read_Source_Log_Pos"`
 	ReplicaIORunning  string          `db:"Replica_IO_Running"`
 	ReplicaSQLRunning string          `db:"Replica_SQL_Running"`
+	LastError         string          `db:"Last_Error"`
 	RetrievedGtidSet  string          `db:"Retrieved_Gtid_Set"`
 	ExecutedGtidSet   string          `db:"Executed_Gtid_Set"`
 	LastIOErrno       int             `db:"Last_IO_Errno"`
+	LastIOError       string          `db:"Last_IO_Error"`
 	LastSQLErrno      int             `db:"Last_SQL_Errno"`
 	Lag               sql.NullFloat64 `db:"Seconds_Behind_Source"`
 }
@@ -95,12 +99,14 @@ type ReplicaStatus interface {
 	ReplicationSQLRunning() bool
 	ReplicationRunning() bool
 	ReplicationState() string
+	GetLastError() string
 	GetMasterHost() string
 	GetMasterLogFile() string
 	GetReadMasterLogPos() int64
 	GetExecutedGtidSet() string
 	GetRetrievedGtidSet() string
 	GetLastIOErrno() int
+	GetLastIOError() string
 	GetLastSQLErrno() int
 	GetReplicationLag() sql.NullFloat64
 }
@@ -144,6 +150,14 @@ func (ss *SlaveStatusStruct) GetLastIOErrno() int {
 	return ss.LastIOErrno
 }
 
+func (ss *SlaveStatusStruct) GetLastError() string {
+	return ss.LastError
+}
+
+func (ss *SlaveStatusStruct) GetLastIOError() string {
+	return ss.LastIOError
+}
+
 func (ss *SlaveStatusStruct) GetLastSQLErrno() int {
 	return ss.LastSQLErrno
 }
@@ -211,7 +225,13 @@ func (ss *ReplicaStatusStruct) GetReplicationLag() sql.NullFloat64 {
 	return ss.Lag
 }
 
-//GetReplicationLag
+func (ss *ReplicaStatusStruct) GetLastError() string {
+	return ss.LastError
+}
+
+func (ss *ReplicaStatusStruct) GetLastIOError() string {
+	return ss.LastIOError
+}
 
 // ReplicationState ...
 func (ss *SlaveStatusStruct) ReplicationState() string {

diff --git a/internal/mysql/node.go b/internal/mysql/node.go
@@ -722,6 +722,15 @@ func (n *Node) StartSlave() error {
 	})
 }
 
+// RestartReplica restart replication
+func (n *Node) RestartReplica() error {
+	err := n.StopSlave()
+	if err != nil {
+		return err
+	}
+	return n.StartSlave()
+}
+
 // StopSlaveIOThread stops IO replication thread
 func (n *Node) StopSlaveIOThread() error {
 	return n.execMogrify(queryStopSlaveIOThread, map[string]interface{}{
@@ -736,7 +745,7 @@ func (n *Node) StartSlaveIOThread() error {
 	})
 }
 
-// RestartSlaveIOThread stops IO replication thread
+// RestartSlaveIOThread restart IO replication thread
 func (n *Node) RestartSlaveIOThread() error {
 	err := n.StopSlaveIOThread()
 	if err != nil {