Exemple #1
0
// changeCallback is run after every action that might
// have changed something in the tablet record.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *pbt.Tablet) error {
	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("ActionAgent.changeCallback")
	defer span.Finish()

	allowQuery := topo.IsRunningQueryService(newTablet.Type)

	// 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
	var disallowQueryReason string
	if allowQuery {
		shardInfo, err = agent.TopoServer.GetShard(ctx, 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 == pbt.TabletType_MASTER {
				if len(shardInfo.SourceShards) > 0 {
					allowQuery = false
					disallowQueryReason = "old master is still in shard info"
				}
			}
			if tc := shardInfo.GetTabletControl(newTablet.Type); tc != nil {
				if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
					if tc.DisableQueryService {
						allowQuery = false
						disallowQueryReason = "query service disabled by tablet control"
					}
					blacklistedTables = tc.BlacklistedTables
					tabletControl = tc
				}
			}
		}
	} else {
		disallowQueryReason = fmt.Sprintf("not a serving tablet type(%v)", newTablet.Type)
	}

	// Read the keyspace on masters to get ShardingColumnType,
	// for binlog replication, only if source shards are set.
	var keyspaceInfo *topo.KeyspaceInfo
	if newTablet.Type == pbt.TabletType_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.stopQueryService("initialize replication")

		// 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 == pbt.TabletType_MASTER && oldTablet.Type != pbt.TabletType_MASTER:
			agent.stopQueryService("tablet promoted to master")

		// 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.stopQueryService("keyrange/blacklistedtables changed")
		}

		if err := agent.allowQueries(newTablet, blacklistedTables); err != nil {
			log.Errorf("Cannot start query service: %v", err)
		}
	} else {
		agent.stopQueryService(disallowQueryReason)
	}

	// 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(strings.ToLower(newTablet.Type.String()))
	statsKeyspace.Set(newTablet.Keyspace)
	statsShard.Set(newTablet.Shard)
	if newTablet.KeyRange != nil {
		statsKeyRangeStart.Set(hex.EncodeToString(newTablet.KeyRange.Start))
		statsKeyRangeEnd.Set(hex.EncodeToString(newTablet.KeyRange.End))
	} else {
		statsKeyRangeStart.Set("")
		statsKeyRangeEnd.Set("")
	}

	// See if we need to start or stop any binlog player
	if agent.BinlogPlayerMap != nil {
		if newTablet.Type == pbt.TabletType_MASTER {
			agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, keyspaceInfo, shardInfo)
		} else {
			agent.BinlogPlayerMap.StopAllPlayersAndReset()
		}
	}
	return nil
}
Exemple #2
0
// changeCallback is run after every action that might
// have changed something in the tablet record or in the topology.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *pbt.Tablet) error {
	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("ActionAgent.changeCallback")
	defer span.Finish()

	allowQuery := topo.IsRunningQueryService(newTablet.Type)

	// 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
	var disallowQueryReason string
	if allowQuery {
		shardInfo, err = agent.TopoServer.GetShard(ctx, 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 == pbt.TabletType_MASTER {
				if len(shardInfo.SourceShards) > 0 {
					allowQuery = false
					disallowQueryReason = "master tablet with filtered replication on"
				}
			}
			if tc := shardInfo.GetTabletControl(newTablet.Type); tc != nil {
				if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
					if tc.DisableQueryService {
						allowQuery = false
						disallowQueryReason = "query service disabled by tablet control"
					}
					blacklistedTables = tc.BlacklistedTables
					tabletControl = tc
				}
			}
		}
	} else {
		disallowQueryReason = fmt.Sprintf("not a serving tablet type(%v)", newTablet.Type)
	}

	if allowQuery {
		if err := agent.allowQueries(newTablet, blacklistedTables); err != nil {
			log.Errorf("Cannot start query service: %v", err)
		}
	} else {
		agent.disallowQueries(newTablet, disallowQueryReason)
	}

	// 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 topo.IsRunningUpdateStream(newTablet.Type) {
		agent.UpdateStream.Enable()
	} else {
		agent.UpdateStream.Disable()
	}

	statsType.Set(strings.ToLower(newTablet.Type.String()))
	statsKeyspace.Set(newTablet.Keyspace)
	statsShard.Set(newTablet.Shard)
	if newTablet.KeyRange != nil {
		statsKeyRangeStart.Set(hex.EncodeToString(newTablet.KeyRange.Start))
		statsKeyRangeEnd.Set(hex.EncodeToString(newTablet.KeyRange.End))
	} else {
		statsKeyRangeStart.Set("")
		statsKeyRangeEnd.Set("")
	}

	// See if we need to start or stop any binlog player
	if agent.BinlogPlayerMap != nil {
		if newTablet.Type == pbt.TabletType_MASTER {
			// Read the keyspace on masters to get
			// ShardingColumnType, for binlog replication,
			// only if source shards are set.
			var keyspaceInfo *topo.KeyspaceInfo
			if shardInfo != nil && len(shardInfo.SourceShards) > 0 {
				keyspaceInfo, err = agent.TopoServer.GetKeyspace(ctx, newTablet.Keyspace)
				if err != nil {
					keyspaceInfo = nil
				}
			}
			agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, keyspaceInfo, shardInfo)
		} else {
			agent.BinlogPlayerMap.StopAllPlayersAndReset()
		}
	}
	return nil
}
Exemple #3
0
// changeCallback is run after every action that might
// have changed something in the tablet record or in the topology.
//
// It owns making changes to the BinlogPlayerMap. The input for this is the
// tablet type (has to be master), and the shard's SourceShards.
//
// It owns updating the blacklisted tables.
//
// It owns updating the stats record for 'TabletType'.
//
// It owns starting and stopping the update stream service.
//
// It owns reading the TabletControl for the current tablet, and storing it.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *topodatapb.Tablet) {
	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("ActionAgent.changeCallback")
	defer span.Finish()

	allowQuery := topo.IsRunningQueryService(newTablet.Type)
	broadcastHealth := false
	runUpdateStream := allowQuery

	// Read the shard to get SourceShards / TabletControlMap if
	// we're going to use it.
	var shardInfo *topo.ShardInfo
	var err error
	var disallowQueryReason string
	var blacklistedTables []string
	updateBlacklistedTables := true
	if allowQuery {
		shardInfo, err = agent.TopoServer.GetShard(ctx, 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)
			updateBlacklistedTables = false
		} else {
			if newTablet.Type == topodatapb.TabletType_MASTER {
				if len(shardInfo.SourceShards) > 0 {
					allowQuery = false
					disallowQueryReason = "master tablet with filtered replication on"
				}
			}
			if tc := shardInfo.GetTabletControl(newTablet.Type); tc != nil {
				if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
					if tc.DisableQueryService {
						allowQuery = false
						disallowQueryReason = "TabletControl.DisableQueryService set"
					}
					blacklistedTables = tc.BlacklistedTables
				}
			}
		}
	} else {
		disallowQueryReason = fmt.Sprintf("not a serving tablet type(%v)", newTablet.Type)
	}
	agent.setServicesDesiredState(disallowQueryReason, runUpdateStream)
	if updateBlacklistedTables {
		if err := agent.loadBlacklistRules(newTablet, blacklistedTables); err != nil {
			// FIXME(alainjobart) how to handle this error?
			log.Errorf("Cannot update blacklisted tables rule: %v", err)
		} else {
			agent.setBlacklistedTables(blacklistedTables)
		}
	}

	if allowQuery {
		// Query service should be running.
		if oldTablet.Type == topodatapb.TabletType_REPLICA &&
			newTablet.Type == topodatapb.TabletType_MASTER {
			// When promoting from replica to master, allow both master and replica
			// queries to be served during gracePeriod.
			if _, err := agent.QueryServiceControl.SetServingType(newTablet.Type,
				true, []topodatapb.TabletType{oldTablet.Type}); err == nil {
				// If successful, broadcast to vtgate and then wait.
				agent.broadcastHealth()
				time.Sleep(*gracePeriod)
			} else {
				log.Errorf("Can't start query service for MASTER+REPLICA mode: %v", err)
			}
		}

		if stateChanged, err := agent.QueryServiceControl.SetServingType(newTablet.Type, true, nil); err == nil {
			// If the state changed, broadcast to vtgate.
			// (e.g. this happens when the tablet was already master, but it just
			// changed from NOT_SERVING to SERVING due to
			// "vtctl MigrateServedFrom ... master".)
			if stateChanged {
				broadcastHealth = true
			}
		} else {
			runUpdateStream = false
			log.Errorf("Cannot start query service: %v", err)
		}
	} else {
		// Query service should be stopped.
		if topo.IsSubjectToLameduck(oldTablet.Type) &&
			newTablet.Type == topodatapb.TabletType_SPARE &&
			*gracePeriod > 0 {
			// When a non-MASTER serving type is going SPARE,
			// put query service in lameduck during gracePeriod.
			agent.lameduck(disallowQueryReason)
		}

		log.Infof("Disabling query service on type change, reason: %v", disallowQueryReason)
		if stateChanged, err := agent.QueryServiceControl.SetServingType(newTablet.Type, false, nil); err == nil {
			// If the state changed, broadcast to vtgate.
			// (e.g. this happens when the tablet was already master, but it just
			// changed from SERVING to NOT_SERVING because filtered replication was
			// enabled.)
			if stateChanged {
				broadcastHealth = true
			}
		} else {
			log.Errorf("SetServingType(serving=false) failed: %v", err)
		}
	}

	// update stream needs to be started or stopped too
	if topo.IsRunningUpdateStream(newTablet.Type) && runUpdateStream {
		agent.UpdateStream.Enable()
	} else {
		agent.UpdateStream.Disable()
	}

	// upate the stats to our current type
	if agent.exportStats {
		agent.statsTabletType.Set(topoproto.TabletTypeLString(newTablet.Type))
	}

	// See if we need to start or stop any binlog player
	if agent.BinlogPlayerMap != nil {
		if newTablet.Type == topodatapb.TabletType_MASTER {
			agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, shardInfo)
		} else {
			agent.BinlogPlayerMap.StopAllPlayersAndReset()
		}
	}

	// Broadcast health changes to vtgate immediately.
	if broadcastHealth {
		agent.broadcastHealth()
	}
}
Exemple #4
0
// changeCallback is run after every action that might
// have changed something in the tablet record or in the topology.
//
// It owns making changes to the BinlogPlayerMap. The input for this is the
// tablet type (has to be master), and the shard's SourceShards.
//
// It owns updating the blacklisted tables.
//
// It owns updating the stats record for 'TabletType'.
//
// It owns starting and stopping the update stream service.
//
// It owns reading the TabletControl for the current tablet, and storing it.
func (agent *ActionAgent) changeCallback(ctx context.Context, oldTablet, newTablet *topodatapb.Tablet) error {
	span := trace.NewSpanFromContext(ctx)
	span.StartLocal("ActionAgent.changeCallback")
	defer span.Finish()

	allowQuery := topo.IsRunningQueryService(newTablet.Type)

	// Read the shard to get SourceShards / TabletControlMap if
	// we're going to use it.
	var shardInfo *topo.ShardInfo
	var tabletControl *topodatapb.Shard_TabletControl
	var err error
	var disallowQueryReason string
	var blacklistedTables []string
	updateBlacklistedTables := true
	if allowQuery {
		shardInfo, err = agent.TopoServer.GetShard(ctx, 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)
			updateBlacklistedTables = false
		} else {
			if newTablet.Type == topodatapb.TabletType_MASTER {
				if len(shardInfo.SourceShards) > 0 {
					allowQuery = false
					disallowQueryReason = "master tablet with filtered replication on"
				}
			}
			if tc := shardInfo.GetTabletControl(newTablet.Type); tc != nil {
				if topo.InCellList(newTablet.Alias.Cell, tc.Cells) {
					if tc.DisableQueryService {
						allowQuery = false
						disallowQueryReason = "query service disabled by tablet control"
					}
					blacklistedTables = tc.BlacklistedTables
					tabletControl = tc
				}
			}
		}
	} else {
		disallowQueryReason = fmt.Sprintf("not a serving tablet type(%v)", newTablet.Type)
	}
	if updateBlacklistedTables {
		if err := agent.loadBlacklistRules(newTablet, blacklistedTables); err != nil {
			// FIXME(alainjobart) how to handle this error?
			log.Errorf("Cannot update blacklisted tables rule: %v", err)
		}
	}

	if allowQuery {
		// Query service should be running.
		if oldTablet.Type == topodatapb.TabletType_REPLICA &&
			newTablet.Type == topodatapb.TabletType_MASTER {
			// When promoting from replica to master, allow both master and replica
			// queries to be served during gracePeriod.
			if err := agent.QueryServiceControl.SetServingType(newTablet.Type, true,
				[]topodatapb.TabletType{oldTablet.Type}); err != nil {
				log.Errorf("Can't start query service for MASTER+REPLICA mode: %v", err)
			} else {
				// If successful, broadcast to vtgate and then wait.
				agent.broadcastHealth()
				time.Sleep(*gracePeriod)
			}
		}
		if err := agent.allowQueries(newTablet.Type); err != nil {
			log.Errorf("Cannot start query service: %v", err)
		}
	} else {
		// Query service should be stopped.
		if (oldTablet.Type == topodatapb.TabletType_REPLICA ||
			oldTablet.Type == topodatapb.TabletType_RDONLY) &&
			newTablet.Type == topodatapb.TabletType_SPARE {
			// When a non-MASTER serving type is going SPARE,
			// put query service in lameduck during gracePeriod.
			agent.enterLameduck(disallowQueryReason)
			agent.broadcastHealth()
			time.Sleep(*gracePeriod)
		}
		agent.disallowQueries(newTablet.Type, disallowQueryReason)
	}

	// 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 topo.IsRunningUpdateStream(newTablet.Type) {
		agent.UpdateStream.Enable()
	} else {
		agent.UpdateStream.Disable()
	}

	// upate the stats to our current type
	if agent.exportStats {
		agent.statsTabletType.Set(strings.ToLower(newTablet.Type.String()))
	}

	// See if we need to start or stop any binlog player
	if agent.BinlogPlayerMap != nil {
		if newTablet.Type == topodatapb.TabletType_MASTER {
			agent.BinlogPlayerMap.RefreshMap(agent.batchCtx, newTablet, shardInfo)
		} else {
			agent.BinlogPlayerMap.StopAllPlayersAndReset()
		}
	}
	return nil
}