Golang topo.ShardInfo类代码示例

func (wr *Wrangler) validateReplication(shardInfo *topo.ShardInfo, tabletMap map[topo.TabletAlias]*topo.TabletInfo, results chan<- vresult) {
	_, ok := tabletMap[shardInfo.MasterAlias]
	if !ok {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("master not in tablet map")}
		return
	}

	actionPath, err := wr.ai.GetSlaves(shardInfo.MasterAlias)
	if err != nil {
		results <- vresult{shardInfo.MasterAlias.String(), err}
		return
	}
	sa, err := wr.ai.WaitForCompletionReply(actionPath, wr.actionTimeout())
	if err != nil {
		results <- vresult{shardInfo.MasterAlias.String(), err}
		return
	}
	slaveAddrs := sa.(*tm.SlaveList).Addrs
	if len(slaveAddrs) == 0 {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("no slaves found")}
		return
	}

	// Some addresses don't resolve in all locations, just use IP address
	if err != nil {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("resolve slaves failed: %v", err)}
		return
	}

	tabletIpMap := make(map[string]*topo.Tablet)
	for _, tablet := range tabletMap {
		ipAddr, _, err := net.SplitHostPort(tablet.MysqlIpAddr)
		if err != nil {
			results <- vresult{tablet.Alias().String(), fmt.Errorf("bad mysql addr: %v %v", tablet.MysqlIpAddr, err)}
			continue
		}
		tabletIpMap[ipAddr] = tablet.Tablet
	}

	// See if every slave is in the replication graph.
	for _, slaveAddr := range slaveAddrs {
		if tabletIpMap[slaveAddr] == nil {
			results <- vresult{shardInfo.Keyspace() + "/" + shardInfo.ShardName(), fmt.Errorf("slave not in replication graph: %v (mysql instance without vttablet?)", slaveAddr)}
		}
	}

	// See if every entry in the replication graph is connected to the master.
	for _, tablet := range tabletMap {
		if !tablet.IsSlaveType() {
			continue
		}

		ipAddr, _, err := net.SplitHostPort(tablet.MysqlIpAddr)
		if err != nil {
			results <- vresult{tablet.Alias().String(), fmt.Errorf("bad mysql addr: %v", err)}
		} else if !strInList(slaveAddrs, ipAddr) {
			results <- vresult{tablet.Alias().String(), fmt.Errorf("slave not replicating: %v %q", ipAddr, slaveAddrs)}
		}
	}
}

// RefreshTablesByShard calls RefreshState on all the tables of a
// given type in a shard. It would work for the master, but the
// discovery wouldn't be very efficient.
func (wr *Wrangler) RefreshTablesByShard(si *topo.ShardInfo, tabletType topo.TabletType, cells []string) error {
	tabletMap, err := topo.GetTabletMapForShardByCell(wr.ts, si.Keyspace(), si.ShardName(), cells)
	switch err {
	case nil:
		// keep going
	case topo.ErrPartialResult:
		wr.Logger().Warningf("RefreshTablesByShard: got partial result for shard %v/%v, may not refresh all tablets everywhere", si.Keyspace(), si.ShardName())
	default:
		return err
	}

	// ignore errors in this phase
	wg := sync.WaitGroup{}
	for _, ti := range tabletMap {
		if ti.Type != tabletType {
			continue
		}

		wg.Add(1)
		go func(ti *topo.TabletInfo) {
			if err := wr.tmc.RefreshState(ti, wr.ActionTimeout()); err != nil {
				wr.Logger().Warningf("RefreshTablesByShard: failed to refresh %v: %v", ti.Alias, err)
			}
			wg.Done()
		}(ti)
	}
	wg.Wait()

	return nil
}

// replicaMigrateServedFrom handles the slave (replica, rdonly) migration.
func (wr *Wrangler) replicaMigrateServedFrom(ki *topo.KeyspaceInfo, sourceShard *topo.ShardInfo, destinationShard *topo.ShardInfo, servedType topo.TabletType, reverse bool, tables []string, ev *events.MigrateServedFrom) error {
	// Save the destination keyspace (its ServedFrom has been changed)
	event.DispatchUpdate(ev, "updating keyspace")
	if err := topo.UpdateKeyspace(wr.ts, ki); err != nil {
		return err
	}

	// Save the source shard (its blacklisted tables field has changed)
	event.DispatchUpdate(ev, "updating source shard")
	if sourceShard.BlacklistedTablesMap == nil {
		sourceShard.BlacklistedTablesMap = make(map[topo.TabletType][]string)
	}
	if reverse {
		delete(sourceShard.BlacklistedTablesMap, servedType)
	} else {
		sourceShard.BlacklistedTablesMap[servedType] = tables
	}
	if err := topo.UpdateShard(wr.ts, sourceShard); err != nil {
		return err
	}

	// Now refresh the source servers so they reload their
	// blacklisted table list
	event.DispatchUpdate(ev, "refreshing sources tablets state so they update their blacklisted tables")
	if err := wr.RefreshTablesByShard(sourceShard.Keyspace(), sourceShard.ShardName(), servedType); err != nil {
		return err
	}

	return nil
}

// InitTablet creates or updates a tablet. If no parent is specified
// in the tablet, and the tablet has a slave type, we will find the
// appropriate parent. If createShardAndKeyspace is true and the
// parent keyspace or shard don't exist, they will be created.  If
// allowUpdate is true, and a tablet with the same ID exists, just update it.
// If a tablet is created as master, and there is already a different
// master in the shard, allowMasterOverride must be set.
func (wr *Wrangler) InitTablet(ctx context.Context, tablet *topodatapb.Tablet, allowMasterOverride, createShardAndKeyspace, allowUpdate bool) error {
	if err := topo.TabletComplete(tablet); err != nil {
		return err
	}

	// get the shard, possibly creating it
	var err error
	var si *topo.ShardInfo

	if createShardAndKeyspace {
		// create the parent keyspace and shard if needed
		si, err = wr.ts.GetOrCreateShard(ctx, tablet.Keyspace, tablet.Shard)
	} else {
		si, err = wr.ts.GetShard(ctx, tablet.Keyspace, tablet.Shard)
		if err == topo.ErrNoNode {
			return fmt.Errorf("missing parent shard, use -parent option to create it, or CreateKeyspace / CreateShard")
		}
	}

	// get the shard, checks a couple things
	if err != nil {
		return fmt.Errorf("cannot get (or create) shard %v/%v: %v", tablet.Keyspace, tablet.Shard, err)
	}
	if !key.KeyRangeEqual(si.KeyRange, tablet.KeyRange) {
		return fmt.Errorf("shard %v/%v has a different KeyRange: %v != %v", tablet.Keyspace, tablet.Shard, si.KeyRange, tablet.KeyRange)
	}
	if tablet.Type == topodatapb.TabletType_MASTER && si.HasMaster() && !topoproto.TabletAliasEqual(si.MasterAlias, tablet.Alias) && !allowMasterOverride {
		return fmt.Errorf("creating this tablet would override old master %v in shard %v/%v, use allow_master_override flag", topoproto.TabletAliasString(si.MasterAlias), tablet.Keyspace, tablet.Shard)
	}

	// update the shard record if needed
	if err := wr.updateShardCellsAndMaster(ctx, si, tablet.Alias, tablet.Type, allowMasterOverride); err != nil {
		return err
	}

	err = wr.ts.CreateTablet(ctx, tablet)
	if err == topo.ErrNodeExists && allowUpdate {
		// Try to update then
		oldTablet, err := wr.ts.GetTablet(ctx, tablet.Alias)
		if err != nil {
			return fmt.Errorf("failed reading existing tablet %v: %v", topoproto.TabletAliasString(tablet.Alias), err)
		}

		// Check we have the same keyspace / shard, and if not,
		// require the allowDifferentShard flag.
		if oldTablet.Keyspace != tablet.Keyspace || oldTablet.Shard != tablet.Shard {
			return fmt.Errorf("old tablet has shard %v/%v. Cannot override with shard %v/%v. Delete and re-add tablet if you want to change the tablet's keyspace/shard", oldTablet.Keyspace, oldTablet.Shard, tablet.Keyspace, tablet.Shard)
		}

		*(oldTablet.Tablet) = *tablet
		if err := wr.ts.UpdateTablet(ctx, oldTablet); err != nil {
			return fmt.Errorf("failed updating tablet %v: %v", topoproto.TabletAliasString(tablet.Alias), err)
		}
	}
	return nil
}

func (zkts *Server) UpdateShard(si *topo.ShardInfo) error {
	shardPath := path.Join(globalKeyspacesPath, si.Keyspace(), "shards", si.ShardName())
	_, err := zkts.zconn.Set(shardPath, jscfg.ToJson(si.Shard), -1)
	if err != nil {
		if zookeeper.IsError(err, zookeeper.ZNONODE) {
			err = topo.ErrNoNode
		}
	}
	return err
}

func (wr *Wrangler) shardReplicationPositions(shardInfo *topo.ShardInfo) ([]*topo.TabletInfo, []*mysqlctl.ReplicationPosition, error) {
	// FIXME(msolomon) this assumes no hierarchical replication, which is currently the case.
	tabletMap, err := GetTabletMapForShard(wr.ts, shardInfo.Keyspace(), shardInfo.ShardName())
	if err != nil {
		return nil, nil, err
	}
	tablets := CopyMapValues(tabletMap, []*topo.TabletInfo{}).([]*topo.TabletInfo)
	positions, err := wr.tabletReplicationPositions(tablets)
	return tablets, positions, err
}

func (wr *Wrangler) shardReplicationStatuses(ctx context.Context, shardInfo *topo.ShardInfo) ([]*topo.TabletInfo, []*myproto.ReplicationStatus, error) {
	// FIXME(msolomon) this assumes no hierarchical replication, which is currently the case.
	tabletMap, err := wr.ts.GetTabletMapForShard(ctx, shardInfo.Keyspace(), shardInfo.ShardName())
	if err != nil {
		return nil, nil, err
	}
	tablets := topotools.CopyMapValues(tabletMap, []*topo.TabletInfo{}).([]*topo.TabletInfo)
	stats, err := wr.tabletReplicationStatuses(ctx, tablets)
	return tablets, stats, err
}

// updateShardCellsAndMaster will update the 'Cells' and possibly
// MasterAlias records for the shard, if needed.
func (wr *Wrangler) updateShardCellsAndMaster(ctx context.Context, si *topo.ShardInfo, tabletAlias topo.TabletAlias, tabletType topo.TabletType, force bool) error {
	// See if we need to update the Shard:
	// - add the tablet's cell to the shard's Cells if needed
	// - change the master if needed
	shardUpdateRequired := false
	if !si.HasCell(tabletAlias.Cell) {
		shardUpdateRequired = true
	}
	if tabletType == topo.TYPE_MASTER && si.MasterAlias != tabletAlias {
		shardUpdateRequired = true
	}
	if !shardUpdateRequired {
		return nil
	}

	actionNode := actionnode.UpdateShard()
	keyspace := si.Keyspace()
	shard := si.ShardName()
	lockPath, err := wr.lockShard(ctx, keyspace, shard, actionNode)
	if err != nil {
		return err
	}

	// re-read the shard with the lock
	si, err = wr.ts.GetShard(ctx, keyspace, shard)
	if err != nil {
		return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
	}

	// update it
	wasUpdated := false
	if !si.HasCell(tabletAlias.Cell) {
		si.Cells = append(si.Cells, tabletAlias.Cell)
		wasUpdated = true
	}
	if tabletType == topo.TYPE_MASTER && si.MasterAlias != tabletAlias {
		if !si.MasterAlias.IsZero() && !force {
			return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, fmt.Errorf("creating this tablet would override old master %v in shard %v/%v", si.MasterAlias, keyspace, shard))
		}
		si.MasterAlias = tabletAlias
		wasUpdated = true
	}

	if wasUpdated {
		// write it back
		if err := topo.UpdateShard(ctx, wr.ts, si); err != nil {
			return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
		}
	}

	// and unlock
	return wr.unlockShard(ctx, keyspace, shard, actionNode, lockPath, err)
}

func (tee *Tee) UpdateShard(si *topo.ShardInfo) error {
	if err := tee.primary.UpdateShard(si); err != nil {
		// failed on primary, not updating secondary
		return err
	}

	if err := tee.secondary.UpdateShard(si); err != nil {
		// not critical enough to fail
		log.Warningf("secondary.UpdateShard(%v,%v) failed: %v", si.Keyspace(), si.ShardName(), err)
	}
	return nil
}

// replicaMigrateServedFrom handles the slave (replica, rdonly) migration.
func (wr *Wrangler) replicaMigrateServedFrom(ctx context.Context, ki *topo.KeyspaceInfo, sourceShard *topo.ShardInfo, destinationShard *topo.ShardInfo, servedType topodatapb.TabletType, cells []string, reverse bool, tables []string, ev *events.MigrateServedFrom) error {
	// Save the destination keyspace (its ServedFrom has been changed)
	event.DispatchUpdate(ev, "updating keyspace")
	if err := wr.ts.UpdateKeyspace(ctx, ki); err != nil {
		return err
	}

	// Save the source shard (its blacklisted tables field has changed)
	event.DispatchUpdate(ev, "updating source shard")
	if err := sourceShard.UpdateSourceBlacklistedTables(servedType, cells, reverse, tables); err != nil {
		return fmt.Errorf("UpdateSourceBlacklistedTables(%v/%v) failed: %v", sourceShard.Keyspace(), sourceShard.ShardName(), err)
	}
	if err := wr.ts.UpdateShard(ctx, sourceShard); err != nil {
		return fmt.Errorf("UpdateShard(%v/%v) failed: %v", sourceShard.Keyspace(), sourceShard.ShardName(), err)
	}

	// Now refresh the source servers so they reload their
	// blacklisted table list
	event.DispatchUpdate(ev, "refreshing sources tablets state so they update their blacklisted tables")
	if err := wr.RefreshTablesByShard(ctx, sourceShard, servedType, cells); err != nil {
		return err
	}

	return nil
}

func (wr *Wrangler) validateReplication(shardInfo *topo.ShardInfo, tabletMap map[topo.TabletAlias]*topo.TabletInfo, results chan<- vresult) {
	masterTablet, ok := tabletMap[shardInfo.MasterAlias]
	if !ok {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("master not in tablet map")}
		return
	}

	slaveList, err := wr.ai.GetSlaves(masterTablet, wr.ActionTimeout())
	if err != nil {
		results <- vresult{shardInfo.MasterAlias.String(), err}
		return
	}
	if len(slaveList) == 0 {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("no slaves found")}
		return
	}

	// Some addresses don't resolve in all locations, just use IP address
	if err != nil {
		results <- vresult{shardInfo.MasterAlias.String(), fmt.Errorf("resolve slaves failed: %v", err)}
		return
	}

	tabletIpMap := make(map[string]*topo.Tablet)
	slaveIpMap := make(map[string]bool)
	for _, tablet := range tabletMap {
		tabletIpMap[normalizeIP(tablet.IPAddr)] = tablet.Tablet
	}

	// See if every slave is in the replication graph.
	for _, slaveAddr := range slaveList {
		if tabletIpMap[normalizeIP(slaveAddr)] == nil {
			results <- vresult{shardInfo.Keyspace() + "/" + shardInfo.ShardName(), fmt.Errorf("slave not in replication graph: %v (mysql instance without vttablet?)", slaveAddr)}
		}
		slaveIpMap[normalizeIP(slaveAddr)] = true
	}

	// See if every entry in the replication graph is connected to the master.
	for _, tablet := range tabletMap {
		if !tablet.IsSlaveType() {
			continue
		}

		if !slaveIpMap[normalizeIP(tablet.IPAddr)] {
			results <- vresult{tablet.Alias.String(), fmt.Errorf("slave not replicating: %v %q", tablet.IPAddr, slaveList)}
		}
	}
}

func (zkts *Server) UpdateShard(si *topo.ShardInfo) error {
	shardPath := path.Join(globalKeyspacesPath, si.Keyspace(), "shards", si.ShardName())
	_, err := zkts.zconn.Set(shardPath, jscfg.ToJson(si.Shard), -1)
	if err != nil {
		if zookeeper.IsError(err, zookeeper.ZNONODE) {
			err = topo.ErrNoNode
		}
		return err
	}

	event.Dispatch(&events.ShardChange{
		ShardInfo: *si,
		Status:    "updated",
	})
	return nil
}

func (wr *Wrangler) finishReparent(si *topo.ShardInfo, masterElect *topo.TabletInfo, majorityRestart, leaveMasterReadOnly bool) error {
	// If the majority of slaves restarted, move ahead.
	if majorityRestart {
		if leaveMasterReadOnly {
			wr.logger.Warningf("leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
		} else {
			wr.logger.Infof("marking master-elect read-write %v", masterElect.Alias)
			if err := wr.tmc.SetReadWrite(masterElect, wr.ActionTimeout()); err != nil {
				wr.logger.Warningf("master master-elect read-write failed, leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
			}
		}
	} else {
		wr.logger.Warningf("minority reparent, manual fixes are needed, leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
	}

	// save the new master in the shard info
	si.MasterAlias = masterElect.Alias
	if err := topo.UpdateShard(wr.ts, si); err != nil {
		wr.logger.Errorf("Failed to save new master into shard: %v", err)
		return err
	}

	// We rebuild all the cells, as we may have taken tablets in and
	// out of the graph.
	wr.logger.Infof("rebuilding shard serving graph data")
	_, err := topotools.RebuildShard(wr.logger, wr.ts, masterElect.Keyspace, masterElect.Shard, nil, wr.lockTimeout, interrupted)
	return err
}

func (wr *Wrangler) finishReparent(si *topo.ShardInfo, masterElect *topo.TabletInfo, majorityRestart, leaveMasterReadOnly bool) error {
	// If the majority of slaves restarted, move ahead.
	if majorityRestart {
		if leaveMasterReadOnly {
			log.Warningf("leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
		} else {
			log.Infof("marking master-elect read-write %v", masterElect.Alias)
			actionPath, err := wr.ai.SetReadWrite(masterElect.Alias)
			if err == nil {
				err = wr.ai.WaitForCompletion(actionPath, wr.actionTimeout())
			}
			if err != nil {
				log.Warningf("master master-elect read-write failed, leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
			}
		}
	} else {
		log.Warningf("minority reparent, manual fixes are needed, leaving master-elect read-only, change with: vtctl SetReadWrite %v", masterElect.Alias)
	}

	// save the new master in the shard info
	si.MasterAlias = masterElect.Alias
	if err := wr.ts.UpdateShard(si); err != nil {
		log.Errorf("Failed to save new master into shard: %v", err)
		return err
	}

	// We rebuild all the cells, as we may have taken tablets in and
	// out of the graph.
	log.Infof("rebuilding shard serving graph data")
	return topotools.RebuildShard(wr.ts, masterElect.Keyspace, masterElect.Shard, topotools.RebuildShardOptions{IgnorePartialResult: false}, wr.lockTimeout, interrupted)
}

// UpdateShard is part of the topo.Server interface
func (zkts *Server) UpdateShard(ctx context.Context, si *topo.ShardInfo, existingVersion int64) (int64, error) {
	shardPath := path.Join(globalKeyspacesPath, si.Keyspace(), "shards", si.ShardName())
	stat, err := zkts.zconn.Set(shardPath, jscfg.ToJSON(si.Shard), int(existingVersion))
	if err != nil {
		if zookeeper.IsError(err, zookeeper.ZNONODE) {
			err = topo.ErrNoNode
		}
		return -1, err
	}

	event.Dispatch(&events.ShardChange{
		ShardInfo: *si,
		Status:    "updated",
	})
	return int64(stat.Version()), nil
}