// Does a topo lookup for a single shard, and returns the tablet record of the master tablet.
func resolveDestinationShardMaster(ctx context.Context, keyspace, shard string, wr *wrangler.Wrangler) (*topo.TabletInfo, error) {
var ti *topo.TabletInfo
shortCtx, cancel := context.WithTimeout(ctx, *remoteActionsTimeout)
si, err := topo.GetShard(shortCtx, wr.TopoServer(), keyspace, shard)
cancel()
if err != nil {
return ti, fmt.Errorf("unable to resolve destination shard %v/%v", keyspace, shard)
}
if si.MasterAlias.IsZero() {
return ti, fmt.Errorf("no master in destination shard %v/%v", keyspace, shard)
}
wr.Logger().Infof("Found target master alias %v in shard %v/%v", si.MasterAlias, keyspace, shard)
shortCtx, cancel = context.WithTimeout(ctx, *remoteActionsTimeout)
ti, err = topo.GetTablet(shortCtx, wr.TopoServer(), si.MasterAlias)
cancel()
if err != nil {
return ti, fmt.Errorf("unable to get master tablet from alias %v in shard %v/%v",
si.MasterAlias, keyspace, shard)
}
return ti, nil
}
// changeCallback is run after every action that might
// have changed something in the tablet record.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *topo.Tablet) error {
span := trace.NewSpanFromContext(ctx)
span.StartLocal("ActionAgent.changeCallback")
defer span.Finish()
allowQuery := newTablet.IsRunningQueryService()
// Read the shard to get SourceShards / TabletControlMap if
// we're going to use it.
var shardInfo *topo.ShardInfo
var tabletControl *pbt.Shard_TabletControl
var blacklistedTables []string
var err error
if allowQuery {
shardInfo, err = topo.GetShard(ctx, agent.TopoServer, newTablet.Keyspace, newTablet.Shard)
if err != nil {
log.Errorf("Cannot read shard for this tablet %v, might have inaccurate SourceShards and TabletControls: %v", newTablet.Alias, err)
} else {
if newTablet.Type == topo.TYPE_MASTER {
allowQuery = len(shardInfo.SourceShards) == 0
}
if tc := shardInfo.GetTabletControl(topo.TabletTypeToProto(newTablet.Type)); tc != nil {
if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
if tc.DisableQueryService {
allowQuery = false
}
blacklistedTables = tc.BlacklistedTables
tabletControl = tc
}
}
}
}
// Read the keyspace on masters to get ShardingColumnType,
// for binlog replication, only if source shards are set.
var keyspaceInfo *topo.KeyspaceInfo
if newTablet.Type == topo.TYPE_MASTER && shardInfo != nil && len(shardInfo.SourceShards) > 0 {
keyspaceInfo, err = agent.TopoServer.GetKeyspace(ctx, newTablet.Keyspace)
if err != nil {
log.Errorf("Cannot read keyspace for this tablet %v: %v", newTablet.Alias, err)
keyspaceInfo = nil
}
}
if allowQuery {
// There are a few transitions when we need to restart the query service:
switch {
// If either InitMaster or InitSlave was called, because those calls
// (or a prior call to ResetReplication) may have silently broken the
// rowcache invalidator by executing RESET MASTER.
// Note that we don't care about fixing it after ResetReplication itself
// since that call breaks everything on purpose, and we don't expect
// anything to start working until either InitMaster or InitSlave.
case agent.initReplication:
agent.initReplication = false
agent.disallowQueries()
// Transitioning from replica to master, so clients that were already
// connected don't keep on using the master as replica or rdonly.
case newTablet.Type == topo.TYPE_MASTER && oldTablet.Type != topo.TYPE_MASTER:
agent.disallowQueries()
// Having different parameters for the query service.
// It needs to stop and restart with the new parameters.
// That includes:
// - changing KeyRange
// - changing the BlacklistedTables list
case (newTablet.KeyRange != oldTablet.KeyRange),
!reflect.DeepEqual(blacklistedTables, agent.BlacklistedTables()):
agent.disallowQueries()
}
if err := agent.allowQueries(newTablet, blacklistedTables); err != nil {
log.Errorf("Cannot start query service: %v", err)
}
} else {
agent.disallowQueries()
}
// save the tabletControl we've been using, so the background
// healthcheck makes the same decisions as we've been making.
agent.setTabletControl(tabletControl)
// update stream needs to be started or stopped too
if agent.DBConfigs != nil {
if topo.IsRunningUpdateStream(newTablet.Type) {
binlog.EnableUpdateStreamService(agent.DBConfigs.App.DbName, agent.MysqlDaemon)
} else {
binlog.DisableUpdateStreamService()
}
}
statsType.Set(string(newTablet.Type))
statsKeyspace.Set(newTablet.Keyspace)
statsShard.Set(newTablet.Shard)
statsKeyRangeStart.Set(string(newTablet.KeyRange.Start.Hex()))
statsKeyRangeEnd.Set(string(newTablet.KeyRange.End.Hex()))
// See if we need to start or stop any binlog player
if agent.BinlogPlayerMap != nil {
if newTablet.Type == topo.TYPE_MASTER {
agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, keyspaceInfo, shardInfo)
} else {
agent.BinlogPlayerMap.StopAllPlayersAndReset()
}
}
return nil
}
// changeCallback is run after every action that might
// have changed something in the tablet record.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *topo.Tablet) error {
span := trace.NewSpanFromContext(ctx)
span.StartLocal("ActionAgent.changeCallback")
defer span.Finish()
allowQuery := newTablet.IsRunningQueryService()
// Read the shard to get SourceShards / TabletControlMap if
// we're going to use it.
var shardInfo *topo.ShardInfo
var tabletControl *topo.TabletControl
var blacklistedTables []string
var err error
if allowQuery {
shardInfo, err = topo.GetShard(ctx, agent.TopoServer, newTablet.Keyspace, newTablet.Shard)
if err != nil {
log.Errorf("Cannot read shard for this tablet %v, might have inaccurate SourceShards and TabletControls: %v", newTablet.Alias, err)
} else {
if newTablet.Type == topo.TYPE_MASTER {
allowQuery = len(shardInfo.SourceShards) == 0
}
if tc, ok := shardInfo.TabletControlMap[newTablet.Type]; ok {
if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
if tc.DisableQueryService {
allowQuery = false
}
blacklistedTables = tc.BlacklistedTables
tabletControl = tc
}
}
}
}
// Read the keyspace on masters to get ShardingColumnType,
// for binlog replication, only if source shards are set.
var keyspaceInfo *topo.KeyspaceInfo
if newTablet.Type == topo.TYPE_MASTER && shardInfo != nil && len(shardInfo.SourceShards) > 0 {
keyspaceInfo, err = agent.TopoServer.GetKeyspace(ctx, newTablet.Keyspace)
if err != nil {
log.Errorf("Cannot read keyspace for this tablet %v: %v", newTablet.Alias, err)
keyspaceInfo = nil
}
}
if allowQuery {
// There are a few transitions when we're
// going to need to restart the query service:
// - transitioning from replica to master, so clients
// that were already connected don't keep on using
// the master as replica or rdonly.
// - having different parameters for the query
// service. It needs to stop and restart with the
// new parameters. That includes:
// - changing KeyRange
// - changing the BlacklistedTables list
if (newTablet.Type == topo.TYPE_MASTER &&
oldTablet.Type != topo.TYPE_MASTER) ||
(newTablet.KeyRange != oldTablet.KeyRange) ||
!reflect.DeepEqual(blacklistedTables, agent.BlacklistedTables()) {
agent.disallowQueries()
}
if err := agent.allowQueries(newTablet, blacklistedTables); err != nil {
log.Errorf("Cannot start query service: %v", err)
}
} else {
agent.disallowQueries()
}
// save the tabletControl we've been using, so the background
// healthcheck makes the same decisions as we've been making.
agent.setTabletControl(tabletControl)
// update stream needs to be started or stopped too
if agent.DBConfigs != nil {
if topo.IsRunningUpdateStream(newTablet.Type) {
binlog.EnableUpdateStreamService(agent.DBConfigs.App.DbName, agent.MysqlDaemon)
} else {
binlog.DisableUpdateStreamService()
}
}
statsType.Set(string(newTablet.Type))
statsKeyspace.Set(newTablet.Keyspace)
statsShard.Set(newTablet.Shard)
statsKeyRangeStart.Set(string(newTablet.KeyRange.Start.Hex()))
statsKeyRangeEnd.Set(string(newTablet.KeyRange.End.Hex()))
// See if we need to start or stop any binlog player
if agent.BinlogPlayerMap != nil {
if newTablet.Type == topo.TYPE_MASTER {
agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, keyspaceInfo, shardInfo)
} else {
agent.BinlogPlayerMap.StopAllPlayersAndReset()
}
}
return nil
}
// TabletExternallyReparented updates all topo records so the current
// tablet is the new master for this shard.
// Should be called under RPCWrapLock.
func (agent *ActionAgent) TabletExternallyReparented(ctx context.Context, externalID string) error {
startTime := time.Now()
// If there is a finalize step running, wait for it to finish or time out
// before checking the global shard record again.
if agent.finalizeReparentCtx != nil {
select {
case <-agent.finalizeReparentCtx.Done():
agent.finalizeReparentCtx = nil
case <-ctx.Done():
return ctx.Err()
}
}
tablet := agent.Tablet()
// Check the global shard record.
si, err := topo.GetShard(ctx, agent.TopoServer, tablet.Keyspace, tablet.Shard)
if err != nil {
log.Warningf("fastTabletExternallyReparented: failed to read global shard record for %v/%v: %v", tablet.Keyspace, tablet.Shard, err)
return err
}
if si.MasterAlias == tablet.Alias {
// We may get called on the current master even when nothing has changed.
// If the global shard record is already updated, it means we successfully
// finished a previous reparent to this tablet.
return nil
}
// Create a reusable Reparent event with available info.
ev := &events.Reparent{
ShardInfo: *si,
NewMaster: *tablet.Tablet,
OldMaster: topo.Tablet{Alias: si.MasterAlias, Type: topo.TYPE_MASTER},
ExternalID: externalID,
}
defer func() {
if err != nil {
event.DispatchUpdate(ev, "failed: "+err.Error())
}
}()
event.DispatchUpdate(ev, "starting external from tablet (fast)")
// Execute state change to master by force-updating only the local copy of the
// tablet record. The actual record in topo will be updated later.
log.Infof("fastTabletExternallyReparented: executing change callback for state change to MASTER")
oldTablet := *tablet.Tablet
newTablet := oldTablet
newTablet.Type = topo.TYPE_MASTER
newTablet.Health = nil
agent.setTablet(topo.NewTabletInfo(&newTablet, -1))
if err := agent.updateState(ctx, &oldTablet, "fastTabletExternallyReparented"); err != nil {
return fmt.Errorf("fastTabletExternallyReparented: failed to change tablet state to MASTER: %v", err)
}
// Directly write the new master endpoint in the serving graph.
// We will do a true rebuild in the background soon, but in the meantime,
// this will be enough for clients to re-resolve the new master.
event.DispatchUpdate(ev, "writing new master endpoint")
log.Infof("fastTabletExternallyReparented: writing new master endpoint to serving graph")
ep, err := tablet.EndPoint()
if err != nil {
return fmt.Errorf("fastTabletExternallyReparented: failed to generate EndPoint for tablet %v: %v", tablet.Alias, err)
}
err = topo.UpdateEndPoints(ctx, agent.TopoServer, tablet.Alias.Cell,
si.Keyspace(), si.ShardName(), topo.TYPE_MASTER,
&topo.EndPoints{Entries: []topo.EndPoint{*ep}}, -1)
if err != nil {
return fmt.Errorf("fastTabletExternallyReparented: failed to update master endpoint: %v", err)
}
externalReparentStats.Record("NewMasterVisible", startTime)
// Start the finalize stage with a background context, but connect the trace.
bgCtx, cancel := context.WithTimeout(agent.batchCtx, *finalizeReparentTimeout)
bgCtx = trace.CopySpan(bgCtx, ctx)
agent.finalizeReparentCtx = bgCtx
go func() {
err := agent.finalizeTabletExternallyReparented(bgCtx, si, ev)
cancel()
if err != nil {
log.Warningf("finalizeTabletExternallyReparented error: %v", err)
event.DispatchUpdate(ev, "failed: "+err.Error())
return
}
externalReparentStats.Record("FullRebuild", startTime)
}()
return nil
}
// CheckShard verifies the Shard operations work correctly
func CheckShard(ctx context.Context, t *testing.T, ts topo.Server) {
if err := ts.CreateKeyspace(ctx, "test_keyspace", &topo.Keyspace{}); err != nil {
t.Fatalf("CreateKeyspace: %v", err)
}
if err := topo.CreateShard(ctx, ts, "test_keyspace", "b0-c0"); err != nil {
t.Fatalf("CreateShard: %v", err)
}
if err := topo.CreateShard(ctx, ts, "test_keyspace", "b0-c0"); err != topo.ErrNodeExists {
t.Errorf("CreateShard called second time, got: %v", err)
}
// Delete shard and see if we can re-create it.
if err := ts.DeleteShard(ctx, "test_keyspace", "b0-c0"); err != nil {
t.Fatalf("DeleteShard: %v", err)
}
if err := topo.CreateShard(ctx, ts, "test_keyspace", "b0-c0"); err != nil {
t.Fatalf("CreateShard: %v", err)
}
// Delete ALL shards.
if err := ts.DeleteKeyspaceShards(ctx, "test_keyspace"); err != nil {
t.Fatalf("DeleteKeyspaceShards: %v", err)
}
if err := topo.CreateShard(ctx, ts, "test_keyspace", "b0-c0"); err != nil {
t.Fatalf("CreateShard: %v", err)
}
if _, err := topo.GetShard(ctx, ts, "test_keyspace", "666"); err != topo.ErrNoNode {
t.Errorf("GetShard(666): %v", err)
}
shardInfo, err := topo.GetShard(ctx, ts, "test_keyspace", "b0-c0")
if err != nil {
t.Errorf("GetShard: %v", err)
}
if want := newKeyRange("b0-c0"); shardInfo.KeyRange != want {
t.Errorf("shardInfo.KeyRange: want %v, got %v", want, shardInfo.KeyRange)
}
master := topo.TabletAlias{Cell: "ny", Uid: 1}
shardInfo.MasterAlias = master
shardInfo.KeyRange = newKeyRange("b0-c0")
shardInfo.ServedTypesMap = map[topo.TabletType]*topo.ShardServedType{
topo.TYPE_MASTER: &topo.ShardServedType{},
topo.TYPE_REPLICA: &topo.ShardServedType{Cells: []string{"c1"}},
topo.TYPE_RDONLY: &topo.ShardServedType{},
}
shardInfo.SourceShards = []topo.SourceShard{
topo.SourceShard{
Uid: 1,
Keyspace: "source_ks",
Shard: "b8-c0",
KeyRange: newKeyRange("b8-c0"),
Tables: []string{"table1", "table2"},
},
}
shardInfo.TabletControlMap = map[topo.TabletType]*topo.TabletControl{
topo.TYPE_MASTER: &topo.TabletControl{
Cells: []string{"c1", "c2"},
BlacklistedTables: []string{"black1", "black2"},
},
topo.TYPE_REPLICA: &topo.TabletControl{
DisableQueryService: true,
},
}
if err := topo.UpdateShard(ctx, ts, shardInfo); err != nil {
t.Errorf("UpdateShard: %v", err)
}
other := topo.TabletAlias{Cell: "ny", Uid: 82873}
_, err = topo.UpdateShardFields(ctx, ts, "test_keyspace", "b0-c0", func(shard *topo.Shard) error {
shard.MasterAlias = other
return nil
})
if err != nil {
t.Fatalf("UpdateShardFields error: %v", err)
}
si, err := topo.GetShard(ctx, ts, "test_keyspace", "b0-c0")
if err != nil {
t.Fatalf("GetShard: %v", err)
}
if si.MasterAlias != other {
t.Fatalf("shard.MasterAlias = %v, want %v", si.MasterAlias, other)
}
_, err = topo.UpdateShardFields(ctx, ts, "test_keyspace", "b0-c0", func(shard *topo.Shard) error {
shard.MasterAlias = master
return nil
})
if err != nil {
t.Fatalf("UpdateShardFields error: %v", err)
}
updatedShardInfo, err := topo.GetShard(ctx, ts, "test_keyspace", "b0-c0")
if err != nil {
t.Fatalf("GetShard: %v", err)
}
if eq, err := shardEqual(shardInfo.Shard, updatedShardInfo.Shard); err != nil {
t.Errorf("cannot compare shards: %v", err)
} else if !eq {
t.Errorf("put and got shards are not identical:\n%#v\n%#v", shardInfo.Shard, updatedShardInfo.Shard)
}
// test GetShardNames
shards, err := ts.GetShardNames(ctx, "test_keyspace")
if err != nil {
t.Errorf("GetShardNames: %v", err)
}
if len(shards) != 1 || shards[0] != "b0-c0" {
t.Errorf(`GetShardNames: want [ "b0-c0" ], got %v`, shards)
}
if _, err := ts.GetShardNames(ctx, "test_keyspace666"); err != topo.ErrNoNode {
t.Errorf("GetShardNames(666): %v", err)
}
// test ValidateShard
if err := ts.ValidateShard(ctx, "test_keyspace", "b0-c0"); err != nil {
t.Errorf("ValidateShard(test_keyspace, b0-c0) failed: %v", err)
}
}