func TestStreamerParseEventsDMLWithoutBegin(t *testing.T) {
input := []replication.BinlogEvent{
rotateEvent{},
formatEvent{},
queryEvent{query: replication.Query{
Database: "vt_test_keyspace",
SQL: "insert into vt_a(eid, id) values (1, 1) /* _stream vt_a (eid id ) (1 1 ); */"}},
xidEvent{},
}
events := make(chan replication.BinlogEvent)
want := []binlogdatapb.BinlogTransaction{
{
Statements: []*binlogdatapb.BinlogTransaction_Statement{
{Category: binlogdatapb.BinlogTransaction_Statement_BL_SET, Sql: []byte("SET TIMESTAMP=1407805592")},
{Category: binlogdatapb.BinlogTransaction_Statement_BL_DML, Sql: []byte("insert into vt_a(eid, id) values (1, 1) /* _stream vt_a (eid id ) (1 1 ); */")},
},
EventToken: &querypb.EventToken{
Timestamp: 1407805592,
Position: replication.EncodePosition(replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 0,
Server: 62344,
Sequence: 0x0d,
},
}),
},
},
{
Statements: nil,
EventToken: &querypb.EventToken{
Timestamp: 1407805592,
Position: replication.EncodePosition(replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 0,
Server: 62344,
Sequence: 0x0d,
},
}),
},
},
}
var got []binlogdatapb.BinlogTransaction
sendTransaction := func(trans *binlogdatapb.BinlogTransaction) error {
got = append(got, *trans)
return nil
}
bls := NewStreamer("vt_test_keyspace", nil, nil, replication.Position{}, 0, sendTransaction)
go sendTestEvents(events, input)
if _, err := bls.parseEvents(context.Background(), events); err != ErrServerEOF {
t.Errorf("unexpected error: %v", err)
}
if !reflect.DeepEqual(got, want) {
t.Errorf("binlogConnStreamer.parseEvents(): got %v, want %v", got, want)
}
}
// updateBlpCheckpoint returns a statement to update a value in the
// _vt.blp_checkpoint table.
func updateBlpCheckpoint(uid uint32, pos replication.Position, timeUpdated int64, txTimestamp int64) string {
if txTimestamp != 0 {
return fmt.Sprintf(
"UPDATE _vt.blp_checkpoint "+
"SET pos='%v', time_updated=%v, transaction_timestamp=%v "+
"WHERE source_shard_uid=%v",
replication.EncodePosition(pos), timeUpdated, txTimestamp, uid)
}
return fmt.Sprintf(
"UPDATE _vt.blp_checkpoint "+
"SET pos='%v', time_updated=%v "+
"WHERE source_shard_uid=%v",
replication.EncodePosition(pos), timeUpdated, uid)
}
// PromoteSlave makes the current tablet the master
func (agent *ActionAgent) PromoteSlave(ctx context.Context) (string, error) {
if err := agent.lock(ctx); err != nil {
return "", err
}
defer agent.unlock()
pos, err := agent.MysqlDaemon.PromoteSlave(agent.hookExtraEnv())
if err != nil {
return "", err
}
// If using semi-sync, we need to enable it before going read-write.
if *enableSemiSync {
if err := agent.enableSemiSync(true); err != nil {
return "", err
}
}
// Set the server read-write
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
if _, err := topotools.ChangeType(ctx, agent.TopoServer, agent.TabletAlias, topodatapb.TabletType_MASTER); err != nil {
return "", err
}
if err := agent.refreshTablet(ctx, "PromoteSlave"); err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// MasterPosition returns the master position
// Should be called under RPCWrap.
func (agent *ActionAgent) MasterPosition(ctx context.Context) (string, error) {
pos, err := agent.MysqlDaemon.MasterPosition()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// PromoteSlaveWhenCaughtUp waits for this slave to be caught up on
// replication up to the provided point, and then makes the slave the
// shard master.
func (agent *ActionAgent) PromoteSlaveWhenCaughtUp(ctx context.Context, position string) (string, error) {
pos, err := replication.DecodePosition(position)
if err != nil {
return "", err
}
if err := agent.MysqlDaemon.WaitMasterPos(ctx, pos); err != nil {
return "", err
}
pos, err = agent.MysqlDaemon.PromoteSlave(agent.hookExtraEnv())
if err != nil {
return "", err
}
// If using semi-sync, we need to enable it before going read-write.
if *enableSemiSync {
if err := agent.enableSemiSync(true); err != nil {
return "", err
}
}
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
if _, err := topotools.ChangeType(ctx, agent.TopoServer, agent.TabletAlias, topodatapb.TabletType_MASTER); err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// InitMaster breaks slaves replication, get the current MySQL replication
// position, insert a row in the reparent_journal table, and returns
// the replication position
func (agent *ActionAgent) InitMaster(ctx context.Context) (string, error) {
// we need to insert something in the binlogs, so we can get the
// current position. Let's just use the mysqlctl.CreateReparentJournal commands.
cmds := mysqlctl.CreateReparentJournal()
if err := agent.MysqlDaemon.ExecuteSuperQueryList(cmds); err != nil {
return "", err
}
// get the current replication position
pos, err := agent.MysqlDaemon.MasterPosition()
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 slaves to be able to commit
// transactions.
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
// Change our type to master if not already
if _, err := agent.TopoServer.UpdateTabletFields(ctx, agent.TabletAlias, func(tablet *topodatapb.Tablet) error {
tablet.Type = topodatapb.TabletType_MASTER
tablet.HealthMap = nil
return nil
}); err != nil {
return "", err
}
agent.initReplication = true
return replication.EncodePosition(pos), nil
}
// PromoteSlaveWhenCaughtUp waits for this slave to be caught up on
// replication up to the provided point, and then makes the slave the
// shard master.
func (agent *ActionAgent) PromoteSlaveWhenCaughtUp(ctx context.Context, position string) (string, error) {
pos, err := replication.DecodePosition(position)
if err != nil {
return "", err
}
// TODO(alainjobart) change the flavor API to take the context directly
// For now, extract the timeout from the context, or wait forever
var waitTimeout time.Duration
if deadline, ok := ctx.Deadline(); ok {
waitTimeout = deadline.Sub(time.Now())
if waitTimeout <= 0 {
waitTimeout = time.Millisecond
}
}
if err := agent.MysqlDaemon.WaitMasterPos(pos, waitTimeout); err != nil {
return "", err
}
pos, err = agent.MysqlDaemon.PromoteSlave(agent.hookExtraEnv())
if err != nil {
return "", err
}
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
if _, err := topotools.ChangeType(ctx, agent.TopoServer, agent.TabletAlias, topodatapb.TabletType_MASTER, topotools.ClearHealthMap); err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// DemoteMaster marks the server read-only, wait until it is done with
// its current transactions, and returns its master position.
// Should be called under RPCWrapLockAction.
func (agent *ActionAgent) DemoteMaster(ctx context.Context) (string, error) {
// Set the server read-only. Note all active connections are not
// affected.
if err := agent.MysqlDaemon.SetReadOnly(true); err != nil {
return "", err
}
// Now disallow queries, to make sure nobody is writing to the
// database.
tablet := agent.Tablet()
// We don't care if the QueryService state actually changed because we'll
// let vtgate keep serving read traffic from this master (see comment below).
if _ /* state changed */, err := agent.disallowQueries(tablet.Type, "DemoteMaster marks server rdonly"); err != nil {
return "", fmt.Errorf("disallowQueries failed: %v", err)
}
// If using semi-sync, we need to disable master-side.
if *enableSemiSync {
if err := agent.enableSemiSync(false); err != nil {
return "", err
}
}
pos, err := agent.MysqlDaemon.DemoteMaster()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
// There is no serving graph update - the master tablet will
// be replaced. Even though writes may fail, reads will
// succeed. It will be less noisy to simply leave the entry
// until we'll promote the master.
}
// RunBlpUntil runs the binlog player server until the position is reached,
// and returns the current mysql master replication position.
func (agent *ActionAgent) RunBlpUntil(ctx context.Context, bpl []*tabletmanagerdatapb.BlpPosition, waitTime time.Duration) (string, error) {
if agent.BinlogPlayerMap == nil {
return "", fmt.Errorf("No BinlogPlayerMap configured")
}
if err := agent.BinlogPlayerMap.RunUntil(ctx, bpl, waitTime); err != nil {
return "", err
}
pos, err := agent.MysqlDaemon.MasterPosition()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// InitMaster enables writes and returns the replication position.
func (agent *ActionAgent) InitMaster(ctx context.Context) (string, error) {
if err := agent.lock(ctx); err != nil {
return "", err
}
defer agent.unlock()
// Initializing as master implies undoing any previous "do not replicate".
agent.setSlaveStopped(false)
// we need to insert something in the binlogs, so we can get the
// current position. Let's just use the mysqlctl.CreateReparentJournal commands.
cmds := mysqlctl.CreateReparentJournal()
if err := agent.MysqlDaemon.ExecuteSuperQueryList(ctx, cmds); err != nil {
return "", err
}
// get the current replication position
pos, err := agent.MysqlDaemon.MasterPosition()
if err != nil {
return "", err
}
// If using semi-sync, we need to enable it before going read-write.
if *enableSemiSync {
if err := agent.enableSemiSync(true); 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 slaves to be able to commit transactions.
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
// Change our type to master if not already
if _, err := agent.TopoServer.UpdateTabletFields(ctx, agent.TabletAlias, func(tablet *topodatapb.Tablet) error {
tablet.Type = topodatapb.TabletType_MASTER
return nil
}); err != nil {
return "", err
}
// and refresh our state
agent.initReplication = true
if err := agent.refreshTablet(ctx, "InitMaster"); err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// StopSlaveMinimum will stop the slave after it reaches at least the
// provided position. Works both when Vitess manages
// replication or not (using hook if not).
func (agent *ActionAgent) StopSlaveMinimum(ctx context.Context, position string, waitTime time.Duration) (string, error) {
pos, err := replication.DecodePosition(position)
if err != nil {
return "", err
}
if err := agent.MysqlDaemon.WaitMasterPos(pos, waitTime); err != nil {
return "", err
}
if err := mysqlctl.StopSlave(agent.MysqlDaemon, agent.hookExtraEnv()); err != nil {
return "", err
}
pos, err = agent.MysqlDaemon.MasterPosition()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// PromoteSlave makes the current tablet the master
func (agent *ActionAgent) PromoteSlave(ctx context.Context) (string, error) {
pos, err := agent.MysqlDaemon.PromoteSlave(agent.hookExtraEnv())
if err != nil {
return "", err
}
// Set the server read-write
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
if _, err := topotools.ChangeType(ctx, agent.TopoServer, agent.TabletAlias, topodatapb.TabletType_MASTER, topotools.ClearHealthMap); err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
func TestStreamerParseEventsMariadbBeginGTID(t *testing.T) {
input := []replication.BinlogEvent{
mariadbRotateEvent,
mariadbFormatEvent,
mariadbBeginGTIDEvent,
mariadbInsertEvent,
mariadbXidEvent,
}
events := make(chan replication.BinlogEvent)
want := []binlogdatapb.BinlogTransaction{
{
Statements: []*binlogdatapb.BinlogTransaction_Statement{
{Category: binlogdatapb.BinlogTransaction_Statement_BL_SET, Charset: charset, Sql: []byte("SET TIMESTAMP=1409892744")},
{Category: binlogdatapb.BinlogTransaction_Statement_BL_DML, Charset: charset, Sql: []byte("insert into vt_insert_test(msg) values ('test 0') /* _stream vt_insert_test (id ) (null ); */")},
},
EventToken: &querypb.EventToken{
Timestamp: 1409892744,
Position: replication.EncodePosition(replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 0,
Server: 62344,
Sequence: 10,
},
}),
},
},
}
var got []binlogdatapb.BinlogTransaction
sendTransaction := func(trans *binlogdatapb.BinlogTransaction) error {
got = append(got, *trans)
return nil
}
bls := NewStreamer("vt_test_keyspace", nil, nil, replication.Position{}, 0, sendTransaction)
go sendTestEvents(events, input)
if _, err := bls.parseEvents(context.Background(), events); err != ErrServerEOF {
t.Errorf("unexpected error: %v", err)
}
if !reflect.DeepEqual(got, want) {
t.Errorf("binlogConnStreamer.parseEvents(): got %v, want %v", got, want)
}
}
// PromoteSlave makes the current tablet the master
func (agent *ActionAgent) PromoteSlave(ctx context.Context) (string, error) {
tablet, err := agent.TopoServer.GetTablet(ctx, agent.TabletAlias)
if err != nil {
return "", err
}
pos, err := agent.MysqlDaemon.PromoteSlave(agent.hookExtraEnv())
if err != nil {
return "", err
}
// Set the server read-write
if err := agent.MysqlDaemon.SetReadOnly(false); err != nil {
return "", err
}
return replication.EncodePosition(pos), agent.updateReplicationGraphForPromotedSlave(ctx, tablet)
}
func TestStreamerParseEventsMariadbStandaloneGTID(t *testing.T) {
input := []replication.BinlogEvent{
mariadbRotateEvent,
mariadbFormatEvent,
mariadbStandaloneGTIDEvent,
mariadbCreateEvent,
}
events := make(chan replication.BinlogEvent)
want := []binlogdatapb.BinlogTransaction{
{
Statements: []*binlogdatapb.BinlogTransaction_Statement{
{Category: binlogdatapb.BinlogTransaction_Statement_BL_SET, Charset: &binlogdatapb.Charset{Client: 8, Conn: 8, Server: 33}, Sql: []byte("SET TIMESTAMP=1409892744")},
{Category: binlogdatapb.BinlogTransaction_Statement_BL_DDL, Charset: &binlogdatapb.Charset{Client: 8, Conn: 8, Server: 33}, Sql: []byte("create table if not exists vt_insert_test (\nid bigint auto_increment,\nmsg varchar(64),\nprimary key (id)\n) Engine=InnoDB")},
},
EventToken: &querypb.EventToken{
Timestamp: 1409892744,
Position: replication.EncodePosition(replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 0,
Server: 62344,
Sequence: 9,
},
}),
},
},
}
var got []binlogdatapb.BinlogTransaction
sendTransaction := func(trans *binlogdatapb.BinlogTransaction) error {
got = append(got, *trans)
return nil
}
bls := NewStreamer("vt_test_keyspace", nil, nil, replication.Position{}, 0, sendTransaction)
go sendTestEvents(events, input)
if _, err := bls.parseEvents(context.Background(), events); err != ErrServerEOF {
t.Errorf("unexpected error: %v", err)
}
if !reflect.DeepEqual(got, want) {
t.Errorf("binlogConnStreamer.parseEvents(): got %v, want %v", got, want)
}
}
// StopSlaveMinimum will stop the slave after it reaches at least the
// provided position. Works both when Vitess manages
// replication or not (using hook if not).
func (agent *ActionAgent) StopSlaveMinimum(ctx context.Context, position string, waitTime time.Duration) (string, error) {
pos, err := replication.DecodePosition(position)
if err != nil {
return "", err
}
waitCtx, cancel := context.WithTimeout(ctx, waitTime)
defer cancel()
if err := agent.MysqlDaemon.WaitMasterPos(waitCtx, pos); err != nil {
return "", err
}
if err := agent.StopSlave(ctx); err != nil {
return "", err
}
pos, err = agent.MysqlDaemon.MasterPosition()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
}
// DemoteMaster marks the server read-only, wait until it is done with
// its current transactions, and returns its master position.
// Should be called under RPCWrapLockAction.
func (agent *ActionAgent) DemoteMaster(ctx context.Context) (string, error) {
// Set the server read-only. Note all active connections are not
// affected.
if err := agent.MysqlDaemon.SetReadOnly(true); err != nil {
return "", err
}
// Now stop the query service, to make sure nobody is writing to the
// database. This will in effect close the connection pools to the
// database.
tablet := agent.Tablet()
agent.disallowQueries(tablet.Tablet.Type, "DemoteMaster marks server rdonly")
pos, err := agent.MysqlDaemon.DemoteMaster()
if err != nil {
return "", err
}
return replication.EncodePosition(pos), nil
// There is no serving graph update - the master tablet will
// be replaced. Even though writes may fail, reads will
// succeed. It will be less noisy to simply leave the entry
// until well promote the master.
}
func TestMigrateServedTypes(t *testing.T) {
// TODO(b/26388813): Remove the next two lines once vtctl WaitForDrain is integrated in the vtctl MigrateServed* commands.
flag.Set("wait_for_drain_sleep_rdonly", "0s")
flag.Set("wait_for_drain_sleep_replica", "0s")
db := fakesqldb.Register()
ts := zktestserver.New(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient())
vp := NewVtctlPipe(t, ts)
defer vp.Close()
// create keyspace
if err := ts.CreateKeyspace(context.Background(), "ks", &topodatapb.Keyspace{
ShardingColumnName: "keyspace_id",
ShardingColumnType: topodatapb.KeyspaceIdType_UINT64,
}); err != nil {
t.Fatalf("CreateKeyspace failed: %v", err)
}
// create the source shard
sourceMaster := NewFakeTablet(t, wr, "cell1", 10, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "0"))
sourceReplica := NewFakeTablet(t, wr, "cell1", 11, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "0"))
sourceRdonly := NewFakeTablet(t, wr, "cell1", 12, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "0"))
// create the first destination shard
dest1Master := NewFakeTablet(t, wr, "cell1", 20, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "-80"))
dest1Replica := NewFakeTablet(t, wr, "cell1", 21, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "-80"))
dest1Rdonly := NewFakeTablet(t, wr, "cell1", 22, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "-80"))
// create the second destination shard
dest2Master := NewFakeTablet(t, wr, "cell1", 30, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "80-"))
dest2Replica := NewFakeTablet(t, wr, "cell1", 31, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "80-"))
dest2Rdonly := NewFakeTablet(t, wr, "cell1", 32, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "80-"))
// double check the shards have the right served types
checkShardServedTypes(t, ts, "0", 3)
checkShardServedTypes(t, ts, "-80", 0)
checkShardServedTypes(t, ts, "80-", 0)
// sourceRdonly will see the refresh
sourceRdonly.StartActionLoop(t, wr)
defer sourceRdonly.StopActionLoop(t)
// sourceReplica will see the refresh
sourceReplica.StartActionLoop(t, wr)
defer sourceReplica.StopActionLoop(t)
// sourceMaster will see the refresh, and has to respond to it
// also will be asked about its replication position.
sourceMaster.FakeMysqlDaemon.CurrentMasterPosition = replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 5,
Server: 456,
Sequence: 892,
},
}
sourceMaster.StartActionLoop(t, wr)
defer sourceMaster.StopActionLoop(t)
// dest1Rdonly will see the refresh
dest1Rdonly.StartActionLoop(t, wr)
defer dest1Rdonly.StopActionLoop(t)
// dest1Replica will see the refresh
dest1Replica.StartActionLoop(t, wr)
defer dest1Replica.StopActionLoop(t)
// dest1Master will see the refresh, and has to respond to it.
// It will also need to respond to WaitBlpPosition, saying it's already caught up.
dest1Master.FakeMysqlDaemon.FetchSuperQueryMap = map[string]*sqltypes.Result{
"SELECT pos, flags FROM _vt.blp_checkpoint WHERE source_shard_uid=0": {
Rows: [][]sqltypes.Value{
{
sqltypes.MakeString([]byte(replication.EncodePosition(sourceMaster.FakeMysqlDaemon.CurrentMasterPosition))),
sqltypes.MakeString([]byte("")),
},
},
},
}
dest1Master.StartActionLoop(t, wr)
defer dest1Master.StopActionLoop(t)
// dest2Rdonly will see the refresh
dest2Rdonly.StartActionLoop(t, wr)
defer dest2Rdonly.StopActionLoop(t)
// dest2Replica will see the refresh
dest2Replica.StartActionLoop(t, wr)
defer dest2Replica.StopActionLoop(t)
// dest2Master will see the refresh, and has to respond to it.
// It will also need to respond to WaitBlpPosition, saying it's already caught up.
dest2Master.FakeMysqlDaemon.FetchSuperQueryMap = map[string]*sqltypes.Result{
"SELECT pos, flags FROM _vt.blp_checkpoint WHERE source_shard_uid=0": {
Rows: [][]sqltypes.Value{
{
sqltypes.MakeString([]byte(replication.EncodePosition(sourceMaster.FakeMysqlDaemon.CurrentMasterPosition))),
sqltypes.MakeString([]byte("")),
},
},
},
}
dest2Master.StartActionLoop(t, wr)
defer dest2Master.StopActionLoop(t)
// migrate will error if the overlapping shards have no "SourceShard" entry
// and we cannot decide which shard is the source or the destination.
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "rdonly"}); err == nil || !strings.Contains(err.Error(), "' have a 'SourceShards' entry. Did you successfully run vtworker SplitClone before? Or did you already migrate the MASTER type?") {
t.Fatalf("MigrateServedType(rdonly) should fail if no 'SourceShards' entry is present: %v", err)
}
// simulate the clone, by fixing the dest shard record
checkShardSourceShards(t, ts, "-80", 0)
checkShardSourceShards(t, ts, "80-", 0)
if err := vp.Run([]string{"SourceShardAdd", "--key_range=-", "ks/-80", "0", "ks/0"}); err != nil {
t.Fatalf("SourceShardAdd failed: %v", err)
}
if err := vp.Run([]string{"SourceShardAdd", "--key_range=-", "ks/80-", "0", "ks/0"}); err != nil {
t.Fatalf("SourceShardAdd failed: %v", err)
}
checkShardSourceShards(t, ts, "-80", 1)
checkShardSourceShards(t, ts, "80-", 1)
// migrate rdonly over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "rdonly"}); err != nil {
t.Fatalf("MigrateServedType(rdonly) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 2)
checkShardServedTypes(t, ts, "-80", 1)
checkShardServedTypes(t, ts, "80-", 1)
checkShardSourceShards(t, ts, "-80", 1)
checkShardSourceShards(t, ts, "80-", 1)
// migrate replica over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "replica"}); err != nil {
t.Fatalf("MigrateServedType(replica) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 1)
checkShardServedTypes(t, ts, "-80", 2)
checkShardServedTypes(t, ts, "80-", 2)
checkShardSourceShards(t, ts, "-80", 1)
checkShardSourceShards(t, ts, "80-", 1)
// migrate master over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "master"}); err != nil {
t.Fatalf("MigrateServedType(master) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 0)
checkShardServedTypes(t, ts, "-80", 3)
checkShardServedTypes(t, ts, "80-", 3)
checkShardSourceShards(t, ts, "-80", 0)
checkShardSourceShards(t, ts, "80-", 0)
}
func TestMigrateServedTypes(t *testing.T) {
db := fakesqldb.Register()
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient())
vp := NewVtctlPipe(t, ts)
defer vp.Close()
// create keyspace
if err := ts.CreateKeyspace(context.Background(), "ks", &topodatapb.Keyspace{
ShardingColumnName: "keyspace_id",
ShardingColumnType: topodatapb.KeyspaceIdType_UINT64,
}); err != nil {
t.Fatalf("CreateKeyspace failed: %v", err)
}
// create the source shard
sourceMaster := NewFakeTablet(t, wr, "cell1", 10, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "0"))
sourceReplica := NewFakeTablet(t, wr, "cell1", 11, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "0"))
sourceRdonly := NewFakeTablet(t, wr, "cell1", 12, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "0"))
// create the first destination shard
dest1Master := NewFakeTablet(t, wr, "cell1", 20, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "-80"))
dest1Replica := NewFakeTablet(t, wr, "cell1", 21, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "-80"))
dest1Rdonly := NewFakeTablet(t, wr, "cell1", 22, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "-80"))
// create the second destination shard
dest2Master := NewFakeTablet(t, wr, "cell1", 30, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "ks", "80-"))
dest2Replica := NewFakeTablet(t, wr, "cell1", 31, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "ks", "80-"))
dest2Rdonly := NewFakeTablet(t, wr, "cell1", 32, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "ks", "80-"))
// double check the shards have the right served types
checkShardServedTypes(t, ts, "0", 3)
checkShardServedTypes(t, ts, "-80", 0)
checkShardServedTypes(t, ts, "80-", 0)
// sourceRdonly will see the refresh
sourceRdonly.StartActionLoop(t, wr)
defer sourceRdonly.StopActionLoop(t)
// sourceReplica will see the refresh
sourceReplica.StartActionLoop(t, wr)
defer sourceReplica.StopActionLoop(t)
// sourceMaster will see the refresh, and has to respond to it
// also will be asked about its replication position.
sourceMaster.FakeMysqlDaemon.CurrentMasterPosition = replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 5,
Server: 456,
Sequence: 892,
},
}
sourceMaster.StartActionLoop(t, wr)
defer sourceMaster.StopActionLoop(t)
// dest1Rdonly will see the refresh
dest1Rdonly.StartActionLoop(t, wr)
defer dest1Rdonly.StopActionLoop(t)
// dest1Replica will see the refresh
dest1Replica.StartActionLoop(t, wr)
defer dest1Replica.StopActionLoop(t)
// dest1Master will see the refresh, and has to respond to it.
// It will also need to respond to WaitBlpPosition, saying it's already caught up.
dest1Master.FakeMysqlDaemon.FetchSuperQueryMap = map[string]*sqltypes.Result{
"SELECT pos, flags FROM _vt.blp_checkpoint WHERE source_shard_uid=0": {
Rows: [][]sqltypes.Value{
{
sqltypes.MakeString([]byte(replication.EncodePosition(sourceMaster.FakeMysqlDaemon.CurrentMasterPosition))),
sqltypes.MakeString([]byte("")),
},
},
},
}
dest1Master.StartActionLoop(t, wr)
defer dest1Master.StopActionLoop(t)
// dest2Rdonly will see the refresh
dest2Rdonly.StartActionLoop(t, wr)
defer dest2Rdonly.StopActionLoop(t)
// dest2Replica will see the refresh
dest2Replica.StartActionLoop(t, wr)
defer dest2Replica.StopActionLoop(t)
// dest2Master will see the refresh, and has to respond to it.
// It will also need to respond to WaitBlpPosition, saying it's already caught up.
dest2Master.FakeMysqlDaemon.FetchSuperQueryMap = map[string]*sqltypes.Result{
"SELECT pos, flags FROM _vt.blp_checkpoint WHERE source_shard_uid=0": {
Rows: [][]sqltypes.Value{
{
sqltypes.MakeString([]byte(replication.EncodePosition(sourceMaster.FakeMysqlDaemon.CurrentMasterPosition))),
sqltypes.MakeString([]byte("")),
},
},
},
}
dest2Master.StartActionLoop(t, wr)
defer dest2Master.StopActionLoop(t)
// simulate the clone, by fixing the dest shard record
if err := vp.Run([]string{"SourceShardAdd", "--key_range=-", "ks/-80", "0", "ks/0"}); err != nil {
t.Fatalf("SourceShardAdd failed: %v", err)
}
if err := vp.Run([]string{"SourceShardAdd", "--key_range=-", "ks/80-", "0", "ks/0"}); err != nil {
t.Fatalf("SourceShardAdd failed: %v", err)
}
// migrate rdonly over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "rdonly"}); err != nil {
t.Fatalf("MigrateServedType(rdonly) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 2)
checkShardServedTypes(t, ts, "-80", 1)
checkShardServedTypes(t, ts, "80-", 1)
// migrate replica over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "replica"}); err != nil {
t.Fatalf("MigrateServedType(replica) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 1)
checkShardServedTypes(t, ts, "-80", 2)
checkShardServedTypes(t, ts, "80-", 2)
// migrate master over
if err := vp.Run([]string{"MigrateServedTypes", "ks/0", "master"}); err != nil {
t.Fatalf("MigrateServedType(master) failed: %v", err)
}
checkShardServedTypes(t, ts, "0", 0)
checkShardServedTypes(t, ts, "-80", 3)
checkShardServedTypes(t, ts, "80-", 3)
}
// ApplyBinlogEvents makes an RPC request to BinlogServer
// and processes the events. It will return nil if the provided context
// was canceled, or if we reached the stopping point.
// It will return io.EOF if the server stops sending us updates.
// It may return any other error it encounters.
func (blp *BinlogPlayer) ApplyBinlogEvents(ctx context.Context) error {
// Instantiate the throttler based on the configuration stored in the db.
maxTPS, maxReplicationLag, err := blp.readThrottlerSettings()
if err != nil {
log.Error(err)
return err
}
t, err := throttler.NewThrottler(
fmt.Sprintf("BinlogPlayer/%d", blp.uid), "transactions", 1 /* threadCount */, maxTPS, maxReplicationLag)
if err != nil {
err := fmt.Errorf("failed to instantiate throttler: %v", err)
log.Error(err)
return err
}
defer t.Close()
// Log the mode of operation and when the player stops.
if len(blp.tables) > 0 {
log.Infof("BinlogPlayer client %v for tables %v starting @ '%v', server: %v",
blp.uid,
blp.tables,
blp.position,
blp.tablet,
)
} else {
log.Infof("BinlogPlayer client %v for keyrange '%v-%v' starting @ '%v', server: %v",
blp.uid,
hex.EncodeToString(blp.keyRange.Start),
hex.EncodeToString(blp.keyRange.End),
blp.position,
blp.tablet,
)
}
if !blp.stopPosition.IsZero() {
// We need to stop at some point. Sanity check the point.
switch {
case blp.position.Equal(blp.stopPosition):
log.Infof("Not starting BinlogPlayer, we're already at the desired position %v", blp.stopPosition)
return nil
case blp.position.AtLeast(blp.stopPosition):
return fmt.Errorf("starting point %v greater than stopping point %v", blp.position, blp.stopPosition)
default:
log.Infof("Will stop player when reaching %v", blp.stopPosition)
}
}
clientFactory, ok := clientFactories[*binlogPlayerProtocol]
if !ok {
return fmt.Errorf("no binlog player client factory named %v", *binlogPlayerProtocol)
}
blplClient := clientFactory()
err = blplClient.Dial(blp.tablet, *BinlogPlayerConnTimeout)
if err != nil {
err := fmt.Errorf("error dialing binlog server: %v", err)
log.Error(err)
return err
}
defer blplClient.Close()
// Get the current charset of our connection, so we can ask the stream server
// to check that they match. The streamer will also only send per-statement
// charset data if that statement's charset is different from what we specify.
if dbClient, ok := blp.dbClient.(*DBClient); ok {
blp.defaultCharset, err = dbClient.dbConn.GetCharset()
if err != nil {
return fmt.Errorf("can't get charset to request binlog stream: %v", err)
}
log.Infof("original charset: %v", blp.defaultCharset)
blp.currentCharset = blp.defaultCharset
// Restore original charset when we're done.
defer func() {
// If the connection has been closed, there's no need to restore
// this connection-specific setting.
if dbClient.dbConn == nil {
return
}
log.Infof("restoring original charset %v", blp.defaultCharset)
if csErr := dbClient.dbConn.SetCharset(blp.defaultCharset); csErr != nil {
log.Errorf("can't restore original charset %v: %v", blp.defaultCharset, csErr)
}
}()
}
var stream BinlogTransactionStream
if len(blp.tables) > 0 {
stream, err = blplClient.StreamTables(ctx, replication.EncodePosition(blp.position), blp.tables, blp.defaultCharset)
} else {
stream, err = blplClient.StreamKeyRange(ctx, replication.EncodePosition(blp.position), blp.keyRange, blp.defaultCharset)
}
if err != nil {
err := fmt.Errorf("error sending streaming query to binlog server: %v", err)
log.Error(err)
return err
}
for {
// Block if we are throttled.
for {
backoff := t.Throttle(0 /* threadID */)
if backoff == throttler.NotThrottled {
break
}
// We don't bother checking for context cancellation here because the
// sleep will block only up to 1 second. (Usually, backoff is 1s / rate
// e.g. a rate of 1000 TPS results into a backoff of 1 ms.)
time.Sleep(backoff)
}
// get the response
response, err := stream.Recv()
if err != nil {
switch err {
case context.Canceled:
return nil
default:
// if the context is canceled, we
// return nil (some RPC
// implementations will remap the
// context error to their own errors)
select {
case <-ctx.Done():
if ctx.Err() == context.Canceled {
return nil
}
default:
}
return fmt.Errorf("Error received from Stream %v", err)
}
}
// process the transaction
for {
ok, err = blp.processTransaction(response)
if err != nil {
return fmt.Errorf("Error in processing binlog event %v", err)
}
if ok {
if !blp.stopPosition.IsZero() {
if blp.position.AtLeast(blp.stopPosition) {
log.Infof("Reached stopping position, done playing logs")
return nil
}
}
break
}
log.Infof("Retrying txn")
time.Sleep(1 * time.Second)
}
}
}
// ApplyBinlogEvents makes an RPC request to BinlogServer
// and processes the events. It will return nil if the provided context
// was canceled, or if we reached the stopping point.
// It will return io.EOF if the server stops sending us updates.
// It may return any other error it encounters.
func (blp *BinlogPlayer) ApplyBinlogEvents(ctx context.Context) error {
if len(blp.tables) > 0 {
log.Infof("BinlogPlayer client %v for tables %v starting @ '%v', server: %v",
blp.uid,
blp.tables,
blp.position,
blp.endPoint,
)
} else {
log.Infof("BinlogPlayer client %v for keyrange '%v-%v' starting @ '%v', server: %v",
blp.uid,
hex.EncodeToString(blp.keyRange.Start),
hex.EncodeToString(blp.keyRange.End),
blp.position,
blp.endPoint,
)
}
if !blp.stopPosition.IsZero() {
// We need to stop at some point. Sanity check the point.
switch {
case blp.position.Equal(blp.stopPosition):
log.Infof("Not starting BinlogPlayer, we're already at the desired position %v", blp.stopPosition)
return nil
case blp.position.AtLeast(blp.stopPosition):
return fmt.Errorf("starting point %v greater than stopping point %v", blp.position, blp.stopPosition)
default:
log.Infof("Will stop player when reaching %v", blp.stopPosition)
}
}
clientFactory, ok := clientFactories[*binlogPlayerProtocol]
if !ok {
return fmt.Errorf("no binlog player client factory named %v", *binlogPlayerProtocol)
}
blplClient := clientFactory()
err := blplClient.Dial(blp.endPoint, *BinlogPlayerConnTimeout)
if err != nil {
log.Errorf("Error dialing binlog server: %v", err)
return fmt.Errorf("error dialing binlog server: %v", err)
}
defer blplClient.Close()
// Get the current charset of our connection, so we can ask the stream server
// to check that they match. The streamer will also only send per-statement
// charset data if that statement's charset is different from what we specify.
if dbClient, ok := blp.dbClient.(*DBClient); ok {
blp.defaultCharset, err = dbClient.dbConn.GetCharset()
if err != nil {
return fmt.Errorf("can't get charset to request binlog stream: %v", err)
}
log.Infof("original charset: %v", blp.defaultCharset)
blp.currentCharset = blp.defaultCharset
// Restore original charset when we're done.
defer func() {
log.Infof("restoring original charset %v", blp.defaultCharset)
if csErr := dbClient.dbConn.SetCharset(blp.defaultCharset); csErr != nil {
log.Errorf("can't restore original charset %v: %v", blp.defaultCharset, csErr)
}
}()
}
var responseChan chan *pb.BinlogTransaction
var errFunc ErrFunc
if len(blp.tables) > 0 {
responseChan, errFunc, err = blplClient.StreamTables(ctx, replication.EncodePosition(blp.position), blp.tables, blp.defaultCharset)
} else {
responseChan, errFunc, err = blplClient.StreamKeyRange(ctx, replication.EncodePosition(blp.position), blp.keyspaceIDType, blp.keyRange, blp.defaultCharset)
}
if err != nil {
log.Errorf("Error sending streaming query to binlog server: %v", err)
return fmt.Errorf("error sending streaming query to binlog server: %v", err)
}
for response := range responseChan {
for {
ok, err = blp.processTransaction(response)
if err != nil {
return fmt.Errorf("Error in processing binlog event %v", err)
}
if ok {
if !blp.stopPosition.IsZero() {
if blp.position.AtLeast(blp.stopPosition) {
log.Infof("Reached stopping position, done playing logs")
return nil
}
}
break
}
log.Infof("Retrying txn")
time.Sleep(1 * time.Second)
}
}
switch err := errFunc(); err {
case nil:
return io.EOF
case context.Canceled:
return nil
default:
// if the context is canceled, we return nil (some RPC
// implementations will remap the context error to their own
// errors)
select {
case <-ctx.Done():
if ctx.Err() == context.Canceled {
return nil
}
default:
}
return fmt.Errorf("Error received from ServeBinlog %v", err)
}
}
// PopulateReparentJournal returns the SQL command to use to populate
// the _vt.reparent_journal table, as well as the time_created_ns
// value used.
func PopulateReparentJournal(timeCreatedNS int64, actionName, masterAlias string, pos replication.Position) string {
return fmt.Sprintf("INSERT INTO _vt.reparent_journal "+
"(time_created_ns, action_name, master_alias, replication_position) "+
"VALUES (%v, '%v', '%v', '%v')",
timeCreatedNS, actionName, masterAlias, replication.EncodePosition(pos))
}
func TestMigrateServedFrom(t *testing.T) {
ctx := context.Background()
db := fakesqldb.Register()
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, tmclient.NewTabletManagerClient())
vp := NewVtctlPipe(t, ts)
defer vp.Close()
// create the source keyspace tablets
sourceMaster := NewFakeTablet(t, wr, "cell1", 10, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "source", "0"))
sourceReplica := NewFakeTablet(t, wr, "cell1", 11, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "source", "0"))
sourceRdonly := NewFakeTablet(t, wr, "cell1", 12, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "source", "0"))
// create the destination keyspace, served form source
// double check it has all entries in map
if err := vp.Run([]string{"CreateKeyspace", "-served_from", "master:source,replica:source,rdonly:source", "dest"}); err != nil {
t.Fatalf("CreateKeyspace(dest) failed: %v", err)
}
ki, err := ts.GetKeyspace(ctx, "dest")
if err != nil {
t.Fatalf("GetKeyspace failed: %v", err)
}
if len(ki.ServedFroms) != 3 {
t.Fatalf("bad initial dest ServedFroms: %+v", ki.ServedFroms)
}
// create the destination keyspace tablets
destMaster := NewFakeTablet(t, wr, "cell1", 20, topodatapb.TabletType_MASTER, db,
TabletKeyspaceShard(t, "dest", "0"))
destReplica := NewFakeTablet(t, wr, "cell1", 21, topodatapb.TabletType_REPLICA, db,
TabletKeyspaceShard(t, "dest", "0"))
destRdonly := NewFakeTablet(t, wr, "cell1", 22, topodatapb.TabletType_RDONLY, db,
TabletKeyspaceShard(t, "dest", "0"))
// sourceRdonly will see the refresh
sourceRdonly.StartActionLoop(t, wr)
defer sourceRdonly.StopActionLoop(t)
// sourceReplica will see the refresh
sourceReplica.StartActionLoop(t, wr)
defer sourceReplica.StopActionLoop(t)
// sourceMaster will see the refresh, and has to respond to it
// also will be asked about its replication position.
sourceMaster.FakeMysqlDaemon.CurrentMasterPosition = replication.Position{
GTIDSet: replication.MariadbGTID{
Domain: 5,
Server: 456,
Sequence: 892,
},
}
sourceMaster.StartActionLoop(t, wr)
defer sourceMaster.StopActionLoop(t)
// destRdonly will see the refresh
destRdonly.StartActionLoop(t, wr)
defer destRdonly.StopActionLoop(t)
// destReplica will see the refresh
destReplica.StartActionLoop(t, wr)
defer destReplica.StopActionLoop(t)
// destMaster will see the refresh, and has to respond to it.
// It will also need to respond to WaitBlpPosition, saying it's already caught up.
destMaster.FakeMysqlDaemon.FetchSuperQueryMap = map[string]*sqltypes.Result{
"SELECT pos, flags FROM _vt.blp_checkpoint WHERE source_shard_uid=0": &sqltypes.Result{
Rows: [][]sqltypes.Value{
[]sqltypes.Value{
sqltypes.MakeString([]byte(replication.EncodePosition(sourceMaster.FakeMysqlDaemon.CurrentMasterPosition))),
sqltypes.MakeString([]byte("")),
},
},
},
}
destMaster.StartActionLoop(t, wr)
defer destMaster.StopActionLoop(t)
// simulate the clone, by fixing the dest shard record
if err := vp.Run([]string{"SourceShardAdd", "--tables", "gone1,gone2", "dest/0", "0", "source/0"}); err != nil {
t.Fatalf("SourceShardAdd failed: %v", err)
}
// migrate rdonly over
if err := vp.Run([]string{"MigrateServedFrom", "dest/0", "rdonly"}); err != nil {
t.Fatalf("MigrateServedFrom(rdonly) failed: %v", err)
}
// check it's gone from keyspace
ki, err = ts.GetKeyspace(ctx, "dest")
if err != nil {
t.Fatalf("GetKeyspace failed: %v", err)
}
if len(ki.ServedFroms) != 2 || ki.GetServedFrom(topodatapb.TabletType_RDONLY) != nil {
t.Fatalf("bad initial dest ServedFroms: %v", ki.ServedFroms)
}
// check the source shard has the right blacklisted tables
si, err := ts.GetShard(ctx, "source", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
if len(si.TabletControls) != 1 || !reflect.DeepEqual(si.TabletControls, []*topodatapb.Shard_TabletControl{
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_RDONLY,
BlacklistedTables: []string{"gone1", "gone2"},
},
}) {
t.Fatalf("rdonly type doesn't have right blacklisted tables")
}
// migrate replica over
if err := vp.Run([]string{"MigrateServedFrom", "dest/0", "replica"}); err != nil {
t.Fatalf("MigrateServedFrom(replica) failed: %v", err)
}
// check it's gone from keyspace
ki, err = ts.GetKeyspace(ctx, "dest")
if err != nil {
t.Fatalf("GetKeyspace failed: %v", err)
}
if len(ki.ServedFroms) != 1 || ki.GetServedFrom(topodatapb.TabletType_REPLICA) != nil {
t.Fatalf("bad initial dest ServedFrom: %+v", ki.ServedFroms)
}
// check the source shard has the right blacklisted tables
si, err = ts.GetShard(ctx, "source", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
if len(si.TabletControls) != 2 || !reflect.DeepEqual(si.TabletControls, []*topodatapb.Shard_TabletControl{
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_RDONLY,
BlacklistedTables: []string{"gone1", "gone2"},
},
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_REPLICA,
BlacklistedTables: []string{"gone1", "gone2"},
},
}) {
t.Fatalf("replica type doesn't have right blacklisted tables")
}
// migrate master over
if err := vp.Run([]string{"MigrateServedFrom", "dest/0", "master"}); err != nil {
t.Fatalf("MigrateServedFrom(master) failed: %v", err)
}
// make sure ServedFromMap is empty
ki, err = ts.GetKeyspace(ctx, "dest")
if err != nil {
t.Fatalf("GetKeyspace failed: %v", err)
}
if len(ki.ServedFroms) > 0 {
t.Fatalf("dest keyspace still is ServedFrom: %+v", ki.ServedFroms)
}
// check the source shard has the right blacklisted tables
si, err = ts.GetShard(ctx, "source", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
if len(si.TabletControls) != 3 || !reflect.DeepEqual(si.TabletControls, []*topodatapb.Shard_TabletControl{
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_RDONLY,
BlacklistedTables: []string{"gone1", "gone2"},
},
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_REPLICA,
BlacklistedTables: []string{"gone1", "gone2"},
},
&topodatapb.Shard_TabletControl{
TabletType: topodatapb.TabletType_MASTER,
BlacklistedTables: []string{"gone1", "gone2"},
},
}) {
t.Fatalf("master type doesn't have right blacklisted tables")
}
}
// parseEvents processes the raw binlog dump stream from the server, one event
// at a time, and groups them into transactions. It is called from within the
// service function launched by Stream().
//
// If the sendTransaction func returns io.EOF, parseEvents returns ErrClientEOF.
// If the events channel is closed, parseEvents returns ErrServerEOF.
// If the context is done, returns ctx.Err().
func (bls *Streamer) parseEvents(ctx context.Context, events <-chan replication.BinlogEvent) (replication.Position, error) {
var statements []*binlogdatapb.BinlogTransaction_Statement
var format replication.BinlogFormat
var gtid replication.GTID
var pos = bls.startPos
var autocommit = true
var err error
// A begin can be triggered either by a BEGIN query, or by a GTID_EVENT.
begin := func() {
if statements != nil {
// If this happened, it would be a legitimate error.
log.Errorf("BEGIN in binlog stream while still in another transaction; dropping %d statements: %v", len(statements), statements)
binlogStreamerErrors.Add("ParseEvents", 1)
}
statements = make([]*binlogdatapb.BinlogTransaction_Statement, 0, 10)
autocommit = false
}
// A commit can be triggered either by a COMMIT query, or by an XID_EVENT.
// Statements that aren't wrapped in BEGIN/COMMIT are committed immediately.
commit := func(timestamp uint32) error {
if int64(timestamp) >= bls.timestamp {
trans := &binlogdatapb.BinlogTransaction{
Statements: statements,
EventToken: &querypb.EventToken{
Timestamp: int64(timestamp),
Position: replication.EncodePosition(pos),
},
}
if err = bls.sendTransaction(trans); err != nil {
if err == io.EOF {
return ErrClientEOF
}
return fmt.Errorf("send reply error: %v", err)
}
}
statements = nil
autocommit = true
return nil
}
// Parse events.
for {
var ev replication.BinlogEvent
var ok bool
select {
case ev, ok = <-events:
if !ok {
// events channel has been closed, which means the connection died.
log.Infof("reached end of binlog event stream")
return pos, ErrServerEOF
}
case <-ctx.Done():
log.Infof("stopping early due to binlog Streamer service shutdown or client disconnect")
return pos, ctx.Err()
}
// Validate the buffer before reading fields from it.
if !ev.IsValid() {
return pos, fmt.Errorf("can't parse binlog event, invalid data: %#v", ev)
}
// We need to keep checking for FORMAT_DESCRIPTION_EVENT even after we've
// seen one, because another one might come along (e.g. on log rotate due to
// binlog settings change) that changes the format.
if ev.IsFormatDescription() {
format, err = ev.Format()
if err != nil {
return pos, fmt.Errorf("can't parse FORMAT_DESCRIPTION_EVENT: %v, event data: %#v", err, ev)
}
continue
}
// We can't parse anything until we get a FORMAT_DESCRIPTION_EVENT that
// tells us the size of the event header.
if format.IsZero() {
// The only thing that should come before the FORMAT_DESCRIPTION_EVENT
// is a fake ROTATE_EVENT, which the master sends to tell us the name
// of the current log file.
if ev.IsRotate() {
continue
}
return pos, fmt.Errorf("got a real event before FORMAT_DESCRIPTION_EVENT: %#v", ev)
}
// Strip the checksum, if any. We don't actually verify the checksum, so discard it.
ev, _, err = ev.StripChecksum(format)
if err != nil {
return pos, fmt.Errorf("can't strip checksum from binlog event: %v, event data: %#v", err, ev)
}
// Update the GTID if the event has one. The actual event type could be
// something special like GTID_EVENT (MariaDB, MySQL 5.6), or it could be
// an arbitrary event with a GTID in the header (Google MySQL).
if ev.HasGTID(format) {
gtid, err = ev.GTID(format)
if err != nil {
return pos, fmt.Errorf("can't get GTID from binlog event: %v, event data: %#v", err, ev)
}
pos = replication.AppendGTID(pos, gtid)
}
switch {
case ev.IsGTID(): // GTID_EVENT
if ev.IsBeginGTID(format) {
begin()
}
case ev.IsXID(): // XID_EVENT (equivalent to COMMIT)
if err = commit(ev.Timestamp()); err != nil {
return pos, err
}
case ev.IsIntVar(): // INTVAR_EVENT
name, value, err := ev.IntVar(format)
if err != nil {
return pos, fmt.Errorf("can't parse INTVAR_EVENT: %v, event data: %#v", err, ev)
}
statements = append(statements, &binlogdatapb.BinlogTransaction_Statement{
Category: binlogdatapb.BinlogTransaction_Statement_BL_SET,
Sql: []byte(fmt.Sprintf("SET %s=%d", name, value)),
})
case ev.IsRand(): // RAND_EVENT
seed1, seed2, err := ev.Rand(format)
if err != nil {
return pos, fmt.Errorf("can't parse RAND_EVENT: %v, event data: %#v", err, ev)
}
statements = append(statements, &binlogdatapb.BinlogTransaction_Statement{
Category: binlogdatapb.BinlogTransaction_Statement_BL_SET,
Sql: []byte(fmt.Sprintf("SET @@RAND_SEED1=%d, @@RAND_SEED2=%d", seed1, seed2)),
})
case ev.IsQuery(): // QUERY_EVENT
// Extract the query string and group into transactions.
q, err := ev.Query(format)
if err != nil {
return pos, fmt.Errorf("can't get query from binlog event: %v, event data: %#v", err, ev)
}
switch cat := getStatementCategory(q.SQL); cat {
case binlogdatapb.BinlogTransaction_Statement_BL_BEGIN:
begin()
case binlogdatapb.BinlogTransaction_Statement_BL_ROLLBACK:
// Rollbacks are possible under some circumstances. Since the stream
// client keeps track of its replication position by updating the set
// of GTIDs it's seen, we must commit an empty transaction so the client
// can update its position.
statements = nil
fallthrough
case binlogdatapb.BinlogTransaction_Statement_BL_COMMIT:
if err = commit(ev.Timestamp()); err != nil {
return pos, err
}
default: // BL_DDL, BL_DML, BL_SET, BL_UNRECOGNIZED
if q.Database != "" && q.Database != bls.dbname {
// Skip cross-db statements.
continue
}
setTimestamp := &binlogdatapb.BinlogTransaction_Statement{
Category: binlogdatapb.BinlogTransaction_Statement_BL_SET,
Sql: []byte(fmt.Sprintf("SET TIMESTAMP=%d", ev.Timestamp())),
}
statement := &binlogdatapb.BinlogTransaction_Statement{
Category: cat,
Sql: []byte(q.SQL),
}
// If the statement has a charset and it's different than our client's
// default charset, send it along with the statement.
// If our client hasn't told us its charset, always send it.
if bls.clientCharset == nil || (q.Charset != nil && *q.Charset != *bls.clientCharset) {
setTimestamp.Charset = q.Charset
statement.Charset = q.Charset
}
statements = append(statements, setTimestamp, statement)
if autocommit {
if err = commit(ev.Timestamp()); err != nil {
return pos, err
}
}
}
case ev.IsPreviousGTIDs(): // PREVIOUS_GTIDS_EVENT
// MySQL 5.6 only: The Binlogs contain an
// event that gives us all the previously
// applied commits. It is *not* an
// authoritative value, unless we started from
// the beginning of a binlog file.
if !bls.usePreviousGTIDs {
continue
}
newPos, err := ev.PreviousGTIDs(format)
if err != nil {
return pos, err
}
pos = newPos
if err = commit(ev.Timestamp()); err != nil {
return pos, err
}
}
}
}