Fixes for avoidance of hosts taking backup in PRS & ERS

go/vt/vtctl/reparentutil/replication.go

@@ -216,9 +216,6 @@ func stopReplicationAndBuildStatusMaps(
 		logger.Infof("getting replication position from %v", alias)
 
 		stopReplicationStatus, err := tmc.StopReplicationAndGetStatus(groupCtx, tabletInfo.Tablet, replicationdatapb.StopReplicationMode_IOTHREADONLY)
-		m.Lock()
-		res.tabletsBackupState[alias] = stopReplicationStatus.GetBackupRunning()
-		m.Unlock()
 		if err != nil {
 			sqlErr, isSQLErr := sqlerror.NewSQLErrorFromError(err).(*sqlerror.SQLError)
 			if isSQLErr && sqlErr != nil && sqlErr.Number() == sqlerror.ERNotReplica {
@@ -242,6 +239,20 @@ func stopReplicationAndBuildStatusMaps(
 				err = vterrors.Wrapf(err, "error when getting replication status for alias %v: %v", alias, err)
 			}
 		} else {
+			isTakingBackup := false
+
+			// Prefer the most up-to-date information regarding whether the tablet is taking a backup from the After
+			// replication status, but fall back to the Before status if After is nil.
+			if stopReplicationStatus.After != nil {
+				isTakingBackup = stopReplicationStatus.After.BackupRunning
+			} else if stopReplicationStatus.Before != nil {
+				isTakingBackup = stopReplicationStatus.Before.BackupRunning
+			}
+
+			m.Lock()
+			res.tabletsBackupState[alias] = isTakingBackup
+			m.Unlock()
+
 			var sqlThreadRunning bool
 			// Check if the sql thread was running for the tablet
 			sqlThreadRunning, err = SQLThreadWasRunning(stopReplicationStatus)

go/vt/vttablet/grpctmserver/server.go

@@ -354,7 +354,7 @@ func (s *server) MysqlHostMetrics(ctx context.Context, request *tabletmanagerdat
 func (s *server) ReplicationStatus(ctx context.Context, request *tabletmanagerdatapb.ReplicationStatusRequest) (response *tabletmanagerdatapb.ReplicationStatusResponse, err error) {
 	defer s.tm.HandleRPCPanic(ctx, "ReplicationStatus", request, response, false /*verbose*/, &err)
 	ctx = callinfo.GRPCCallInfo(ctx)
-	response = &tabletmanagerdatapb.ReplicationStatusResponse{BackupRunning: s.tm.IsBackupRunning()}
+	response = &tabletmanagerdatapb.ReplicationStatusResponse{}
 	status, err := s.tm.ReplicationStatus(ctx)
 	if err == nil {
 		response.Status = status
@@ -638,8 +638,6 @@ func (s *server) StopReplicationAndGetStatus(ctx context.Context, request *table
 		response.Status = statusResponse.Status
 	}
 
-	response.BackupRunning = s.tm.IsBackupRunning()
-
 	return response, err
 }
 

go/vt/vttablet/tabletmanager/rpc_replication.go

@@ -46,7 +46,11 @@ func (tm *TabletManager) ReplicationStatus(ctx context.Context) (*replicationdat
 	if err != nil {
 		return nil, err
 	}
-	return replication.ReplicationStatusToProto(status), nil
+
+	protoStatus := replication.ReplicationStatusToProto(status)
+	protoStatus.BackupRunning = tm.IsBackupRunning()
+
+	return protoStatus, nil
 }
 
 // FullStatus returns the full status of MySQL including the replication information, semi-sync information, GTID information among others
@@ -893,6 +897,7 @@ func (tm *TabletManager) StopReplicationAndGetStatus(ctx context.Context, stopRe
 		return StopReplicationAndGetStatusResponse{}, vterrors.Wrap(err, "before status failed")
 	}
 	before := replication.ReplicationStatusToProto(rs)
+	before.BackupRunning = tm.IsBackupRunning()
 
 	if stopReplicationMode == replicationdatapb.StopReplicationMode_IOTHREADONLY {
 		if !rs.IOHealthy() {
@@ -939,6 +944,7 @@ func (tm *TabletManager) StopReplicationAndGetStatus(ctx context.Context, stopRe
 		}, vterrors.Wrap(err, "acquiring replication status failed")
 	}
 	after := replication.ReplicationStatusToProto(rsAfter)
+	after.BackupRunning = tm.IsBackupRunning()
 
 	rs.Position = rsAfter.Position
 	rs.RelayLogPosition = rsAfter.RelayLogPosition

proto/replicationdata.proto

@@ -66,7 +66,6 @@ message Configuration {
 message StopReplicationStatus {
   replicationdata.Status before = 1;
   replicationdata.Status after = 2;
-  bool backup_running = 3;
 }
 
 // StopReplicationMode is used to provide controls over how replication is stopped.

proto/tabletmanagerdata.proto

@@ -360,7 +360,6 @@ message ReplicationStatusRequest {
 
 message ReplicationStatusResponse {
   replicationdata.Status status = 1;
-  bool backup_running = 2;
 }
 
 message PrimaryStatusRequest {
@@ -549,7 +548,6 @@ message StopReplicationAndGetStatusResponse {
 
   // Status represents the replication status call right before, and right after telling the replica to stop.
   replicationdata.StopReplicationStatus status = 2;
-  bool backup_running = 3;
 }
 
 message PromoteReplicaRequest {