Fix: convertBoolToSemiSyncAction method to account for all semi sync …

…actions (vitessio#13075)

* Fix convertBoolToSemiSyncAction method to account for all semi sync actions

Signed-off-by: William Lu <william.lu@shopify.com>

* Add VitessError VT09013

Signed-off-by: William Lu <william@tamedfox.ca>

---------

Signed-off-by: William Lu <william.lu@shopify.com>
Signed-off-by: William Lu <william@tamedfox.ca>
Signed-off-by: Arthur Schreiber <arthurschreiber@github.com>

go/cmd/vtcombo/main.go

@@ -345,3 +345,8 @@ func (mysqld *vtcomboMysqld) StopReplication(hookExtraEnv map[string]string) err
 func (mysqld *vtcomboMysqld) SetSemiSyncEnabled(source, replica bool) error {
 	return nil
 }
+
+// SemiSyncExtensionLoaded implements the MysqlDaemon interface
+func (mysqld *vtcomboMysqld) SemiSyncExtensionLoaded() (bool, error) {
+	return true, nil
+}

go/test/endtoend/reparent/newfeaturetest/reparent_test.go

@@ -17,6 +17,8 @@ limitations under the License.
 package newfeaturetest
 
 import (
+	"context"
+	"fmt"
 	"testing"
 
 	"github.com/stretchr/testify/require"
@@ -109,3 +111,38 @@ func TestTabletRestart(t *testing.T) {
 	err := tablets[1].VttabletProcess.Setup()
 	require.NoError(t, err)
 }
+
+// Tests ensures that ChangeTabletType works even when semi-sync plugins are not loaded.
+func TestChangeTypeWithoutSemiSync(t *testing.T) {
+	defer cluster.PanicHandler(t)
+	clusterInstance := utils.SetupReparentCluster(t, "none")
+	defer utils.TeardownCluster(clusterInstance)
+	tablets := clusterInstance.Keyspaces[0].Shards[0].Vttablets
+
+	ctx := context.Background()
+
+	primary, replica := tablets[0], tablets[1]
+
+	// Unload semi sync plugins
+	for _, tablet := range tablets[0:4] {
+		qr := utils.RunSQL(ctx, t, "select @@global.super_read_only", tablet)
+		result := fmt.Sprintf("%v", qr.Rows[0][0].ToString())
+		if result == "1" {
+			utils.RunSQL(ctx, t, "set global super_read_only = 0", tablet)
+		}
+
+		utils.RunSQL(ctx, t, "UNINSTALL PLUGIN rpl_semi_sync_slave;", tablet)
+		utils.RunSQL(ctx, t, "UNINSTALL PLUGIN rpl_semi_sync_master;", tablet)
+	}
+
+	utils.ValidateTopology(t, clusterInstance, true)
+	utils.CheckPrimaryTablet(t, clusterInstance, primary)
+
+	// Change replica's type to rdonly
+	err := clusterInstance.VtctlclientProcess.ExecuteCommand("ChangeTabletType", replica.Alias, "rdonly")
+	require.NoError(t, err)
+
+	// Change tablets type from rdonly back to replica
+	err = clusterInstance.VtctlclientProcess.ExecuteCommand("ChangeTabletType", replica.Alias, "replica")
+	require.NoError(t, err)
+}

go/vt/mysqlctl/fakemysqldaemon/fakemysqldaemon.go

@@ -663,6 +663,11 @@ func (fmd *FakeMysqlDaemon) SemiSyncClients() uint32 {
 	return 0
 }
 
+// SemiSyncExtensionLoaded is part of the MysqlDaemon interface.
+func (fmd *FakeMysqlDaemon) SemiSyncExtensionLoaded() (bool, error) {
+	return true, nil
+}
+
 // SemiSyncSettings is part of the MysqlDaemon interface.
 func (fmd *FakeMysqlDaemon) SemiSyncSettings() (timeout uint64, numReplicas uint32) {
 	return 10000000, 1

go/vt/mysqlctl/mysql_daemon.go

@@ -57,6 +57,7 @@ type MysqlDaemon interface {
 	GetGTIDPurged(ctx context.Context) (mysql.Position, error)
 	SetSemiSyncEnabled(source, replica bool) error
 	SemiSyncEnabled() (source, replica bool)
+	SemiSyncExtensionLoaded() (bool, error)
 	SemiSyncStatus() (source, replica bool)
 	SemiSyncClients() (count uint32)
 	SemiSyncSettings() (timeout uint64, numReplicas uint32)

go/vt/mysqlctl/replication.go

@@ -705,3 +705,16 @@ func (mysqld *Mysqld) SemiSyncReplicationStatus() (bool, error) {
 	}
 	return false, nil
 }
+
+// SemiSyncExtensionLoaded returns whether semi-sync plugins are loaded.
+func (mysqld *Mysqld) SemiSyncExtensionLoaded() (bool, error) {
+	qr, err := mysqld.FetchSuperQuery(context.Background(), "SELECT COUNT(*) > 0 AS plugin_loaded FROM information_schema.plugins WHERE plugin_name LIKE 'rpl_semi_sync%'")
+	if err != nil {
+		return false, err
+	}
+	pluginPresent, err := qr.Rows[0][0].ToBool()
+	if err != nil {
+		return false, err
+	}
+	return pluginPresent, nil
+}

go/vt/vterrors/code.go

@@ -67,6 +67,7 @@ var (
 	VT09008 = errorWithoutState("VT09008", vtrpcpb.Code_FAILED_PRECONDITION, "vexplain queries/all will actually run queries", "vexplain queries/all will actually run queries. `/*vt+ EXECUTE_DML_QUERIES */` must be set to run DML queries in vtexplain. Example: `vexplain /*vt+ EXECUTE_DML_QUERIES */ queries delete from t1`")
 	VT09009 = errorWithoutState("VT09009", vtrpcpb.Code_FAILED_PRECONDITION, "stream is supported only for primary tablet type, current type: %v", "Stream is only supported for primary tablets, please use a stream on those tablets.")
 	VT09010 = errorWithoutState("VT09010", vtrpcpb.Code_FAILED_PRECONDITION, "SHOW VITESS_THROTTLER STATUS works only on primary tablet", "SHOW VITESS_THROTTLER STATUS works only on primary tablet.")
+	VT09013 = errorWithoutState("VT09013", vtrpcpb.Code_FAILED_PRECONDITION, "semi-sync plugins are not loaded", "Durability policy wants Vitess to use semi-sync, but the MySQL instances don't have the semi-sync plugin loaded.")
 
 	VT10001 = errorWithoutState("VT10001", vtrpcpb.Code_ABORTED, "foreign key constraints are not allowed", "Foreign key constraints are not allowed, see https://vitess.io/blog/2021-06-15-online-ddl-why-no-fk/.")
 
@@ -123,6 +124,7 @@ var (
 		VT09008,
 		VT09009,
 		VT09010,
+		VT09013,
 		VT10001,
 		VT12001,
 		VT13001,

go/vt/vttablet/tabletmanager/rpc_actions.go

@@ -82,7 +82,13 @@ func (tm *TabletManager) ChangeType(ctx context.Context, tabletType topodatapb.T
 		return err
 	}
 	defer tm.unlock()
-	return tm.changeTypeLocked(ctx, tabletType, DBActionNone, convertBoolToSemiSyncAction(semiSync))
+
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return err
+	}
+
+	return tm.changeTypeLocked(ctx, tabletType, DBActionNone, semiSyncAction)
 }
 
 // ChangeType changes the tablet type
@@ -142,9 +148,23 @@ func (tm *TabletManager) RunHealthCheck(ctx context.Context) {
 	tm.QueryServiceControl.BroadcastHealth()
 }
 
-func convertBoolToSemiSyncAction(semiSync bool) SemiSyncAction {
-	if semiSync {
-		return SemiSyncActionSet
+func (tm *TabletManager) convertBoolToSemiSyncAction(semiSync bool) (SemiSyncAction, error) {
+	semiSyncExtensionLoaded, err := tm.MysqlDaemon.SemiSyncExtensionLoaded()
+	if err != nil {
+		return SemiSyncActionNone, err
+	}
+
+	if semiSyncExtensionLoaded {
+		if semiSync {
+			return SemiSyncActionSet, nil
+		} else {
+			return SemiSyncActionUnset, nil
+		}
+	} else {
+		if semiSync {
+			return SemiSyncActionNone, vterrors.VT09013()
+		} else {
+			return SemiSyncActionNone, nil
+		}
 	}
-	return SemiSyncActionUnset
 }

go/vt/vttablet/tabletmanager/rpc_replication.go

@@ -247,7 +247,12 @@ func (tm *TabletManager) StartReplication(ctx context.Context, semiSync bool) er
 	}
 	defer tm.unlock()
 
-	if err := tm.fixSemiSync(tm.Tablet().Type, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return err
+	}
+
+	if err := tm.fixSemiSync(tm.Tablet().Type, semiSyncAction); err != nil {
 		return err
 	}
 	return tm.MysqlDaemon.StartReplication(tm.hookExtraEnv())
@@ -321,16 +326,21 @@ func (tm *TabletManager) InitPrimary(ctx context.Context, semiSync bool) (string
 		return "", err
 	}
 
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return "", err
+	}
+
 	// Set the server read-write, from now on we can accept real
 	// client writes. Note that if semi-sync replication is enabled,
 	// we'll still need some replicas to be able to commit transactions.
-	if err := tm.changeTypeLocked(ctx, topodatapb.TabletType_PRIMARY, DBActionSetReadWrite, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	if err := tm.changeTypeLocked(ctx, topodatapb.TabletType_PRIMARY, DBActionSetReadWrite, semiSyncAction); err != nil {
 		return "", err
 	}
 
 	// Enforce semi-sync after changing the tablet type to PRIMARY. Otherwise, the
 	// primary will hang while trying to create the database.
-	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, semiSyncAction); err != nil {
 		return "", err
 	}
 
@@ -360,11 +370,16 @@ func (tm *TabletManager) InitReplica(ctx context.Context, parent *topodatapb.Tab
 	}
 	defer tm.unlock()
 
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return err
+	}
+
 	// If we were a primary type, switch our type to replica.  This
 	// is used on the old primary when using InitShardPrimary with
 	// -force, and the new primary is different from the old primary.
 	if tm.Tablet().Type == topodatapb.TabletType_PRIMARY {
-		if err := tm.changeTypeLocked(ctx, topodatapb.TabletType_REPLICA, DBActionNone, convertBoolToSemiSyncAction(semiSync)); err != nil {
+		if err := tm.changeTypeLocked(ctx, topodatapb.TabletType_REPLICA, DBActionNone, semiSyncAction); err != nil {
 			return err
 		}
 	}
@@ -385,7 +400,7 @@ func (tm *TabletManager) InitReplica(ctx context.Context, parent *topodatapb.Tab
 	if tt == topodatapb.TabletType_PRIMARY {
 		tt = topodatapb.TabletType_REPLICA
 	}
-	if err := tm.fixSemiSync(tt, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	if err := tm.fixSemiSync(tt, semiSyncAction); err != nil {
 		return err
 	}
 
@@ -525,8 +540,13 @@ func (tm *TabletManager) UndoDemotePrimary(ctx context.Context, semiSync bool) e
 	}
 	defer tm.unlock()
 
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return err
+	}
+
 	// If using semi-sync, we need to enable source-side.
-	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, semiSyncAction); err != nil {
 		return err
 	}
 
@@ -583,9 +603,14 @@ func (tm *TabletManager) SetReplicationSource(ctx context.Context, parentAlias *
 	}
 	defer tm.unlock()
 
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return err
+	}
+
 	// setReplicationSourceLocked also fixes the semi-sync. In case the tablet type is primary it assumes that it will become a replica if SetReplicationSource
 	// is called, so we always call fixSemiSync with a non-primary tablet type. This will always set the source side replication to false.
-	return tm.setReplicationSourceLocked(ctx, parentAlias, timeCreatedNS, waitPosition, forceStartReplication, convertBoolToSemiSyncAction(semiSync))
+	return tm.setReplicationSourceLocked(ctx, parentAlias, timeCreatedNS, waitPosition, forceStartReplication, semiSyncAction)
 }
 
 func (tm *TabletManager) setReplicationSourceRepairReplication(ctx context.Context, parentAlias *topodatapb.TabletAlias, timeCreatedNS int64, waitPosition string, forceStartReplication bool) (err error) {
@@ -839,8 +864,13 @@ func (tm *TabletManager) PromoteReplica(ctx context.Context, semiSync bool) (str
 		return "", err
 	}
 
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(semiSync)
+	if err != nil {
+		return "", err
+	}
+
 	// If using semi-sync, we need to enable it before going read-write.
-	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, convertBoolToSemiSyncAction(semiSync)); err != nil {
+	if err := tm.fixSemiSync(topodatapb.TabletType_PRIMARY, semiSyncAction); err != nil {
 		return "", err
 	}
 

go/vt/vttablet/tabletmanager/tm_init.go

@@ -872,8 +872,15 @@ func (tm *TabletManager) initializeReplication(ctx context.Context, tabletType t
 	}
 	// If using semi-sync, we need to enable it before connecting to primary.
 	// We should set the correct type, since it is used in replica semi-sync
+
 	tablet.Type = tabletType
-	if err := tm.fixSemiSync(tabletType, convertBoolToSemiSyncAction(reparentutil.IsReplicaSemiSync(durability, currentPrimary.Tablet, tablet))); err != nil {
+
+	semiSyncAction, err := tm.convertBoolToSemiSyncAction(reparentutil.IsReplicaSemiSync(durability, currentPrimary.Tablet, tablet))
+	if err != nil {
+		return nil, err
+	}
+
+	if err := tm.fixSemiSync(tabletType, semiSyncAction); err != nil {
 		return nil, err
 	}