// rpcCallTablet wil execute the RPC on the remote server.
func (client *GoRpcTabletManagerClient) rpcCallTablet(ctx context.Context, tablet *topo.TabletInfo, name string, args, reply interface{}) error {
// create the RPC client, using ctx.Deadline if set, or no timeout.
var connectTimeout time.Duration
deadline, ok := ctx.Deadline()
if ok {
connectTimeout = deadline.Sub(time.Now())
if connectTimeout < 0 {
return fmt.Errorf("Timeout connecting to TabletManager.%v on %v", name, tablet.Alias)
}
}
rpcClient, err := bsonrpc.DialHTTP("tcp", tablet.Addr(), connectTimeout, nil)
if err != nil {
return fmt.Errorf("RPC error for %v: %v", tablet.Alias, err.Error())
}
defer rpcClient.Close()
// use the context Done() channel. Will handle context timeout.
call := rpcClient.Go(ctx, "TabletManager."+name, args, reply, nil)
select {
case <-ctx.Done():
return fmt.Errorf("Timeout waiting for TabletManager.%v to %v", name, tablet.Alias)
case <-call.Done:
if call.Error != nil {
return fmt.Errorf("Remote error for %v: %v", tablet.Alias, call.Error.Error())
} else {
return nil
}
}
}
func (client *GoRpcTabletManagerClient) Snapshot(ctx context.Context, tablet *topo.TabletInfo, sa *actionnode.SnapshotArgs, waitTime time.Duration) (<-chan *logutil.LoggerEvent, tmclient.SnapshotReplyFunc, error) {
rpcClient, err := bsonrpc.DialHTTP("tcp", tablet.Addr(), waitTime, nil)
if err != nil {
return nil, nil, err
}
logstream := make(chan *logutil.LoggerEvent, 10)
rpcstream := make(chan *gorpcproto.SnapshotStreamingReply, 10)
result := &actionnode.SnapshotReply{}
c := rpcClient.StreamGo("TabletManager.Snapshot", sa, rpcstream)
go func() {
for ssr := range rpcstream {
if ssr.Log != nil {
logstream <- ssr.Log
}
if ssr.Result != nil {
*result = *ssr.Result
}
}
close(logstream)
rpcClient.Close()
}()
return logstream, func() (*actionnode.SnapshotReply, error) {
return result, c.Error
}, nil
}
// runSqlCommands will send the sql commands to the remote tablet.
func runSqlCommands(wr *wrangler.Wrangler, ti *topo.TabletInfo, commands []string, abort chan struct{}, disableBinLogs bool) error {
for _, command := range commands {
command, err := fillStringTemplate(command, map[string]string{"DatabaseName": ti.DbName()})
if err != nil {
return fmt.Errorf("fillStringTemplate failed: %v", err)
}
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
_, err = wr.TabletManagerClient().ExecuteFetch(ctx, ti, command, 0, false, disableBinLogs)
if err != nil {
return err
}
cancel()
// check on abort
select {
case <-abort:
return nil
default:
break
}
}
return nil
}
// updateReplicationGraphForPromotedSlave makes sure the newly promoted slave
// is correctly represented in the replication graph
func (agent *ActionAgent) updateReplicationGraphForPromotedSlave(ctx context.Context, tablet *topo.TabletInfo) error {
// Update tablet regardless - trend towards consistency.
tablet.State = topo.STATE_READ_WRITE
tablet.Type = topo.TYPE_MASTER
tablet.Parent.Cell = ""
tablet.Parent.Uid = topo.NO_TABLET
tablet.Health = nil
err := topo.UpdateTablet(ctx, agent.TopoServer, tablet)
if err != nil {
return err
}
// NOTE(msolomon) A serving graph update is required, but in
// order for the shard to be consistent the old master must be
// scrapped first. That is externally coordinated by the
// wrangler reparent action.
// Insert the new tablet location in the replication graph now that
// we've updated the tablet.
err = topo.UpdateTabletReplicationData(ctx, agent.TopoServer, tablet.Tablet)
if err != nil && err != topo.ErrNodeExists {
return err
}
return nil
}
func (wr *Wrangler) checkMasterElect(ti *topo.TabletInfo) error {
// Check the master-elect is fit for duty - call out for hardware checks.
// if the server was already serving live traffic, it's probably good
if ti.IsInServingGraph() {
return nil
}
return wr.ExecuteOptionalTabletInfoHook(ti, hook.NewSimpleHook("preflight_serving_type"))
}
func newTabletNodeFromTabletInfo(ti *topo.TabletInfo) *TabletNode {
if err := ti.ValidatePortmap(); err != nil {
log.Errorf("ValidatePortmap(%v): %v", ti.Alias, err)
}
return &TabletNode{
Host: ti.Hostname,
Port: ti.Portmap["vt"],
Alias: ti.Alias,
}
}
// applySqlShard applies a given SQL change on a given tablet alias. It allows executing arbitrary
// SQL statements, but doesn't return any results, so it's only useful for SQL statements
// that would be run for their effects (e.g., CREATE).
// It works by applying the SQL statement on the shard's master tablet with replication turned on.
// Thus it should be used only for changes that can be applies on a live instance without causing issues;
// it shouldn't be used for anything that will require a pivot.
// The SQL statement string is expected to have {{.DatabaseName}} in place of the actual db name.
func (wr *Wrangler) applySqlShard(tabletInfo *topo.TabletInfo, change string) error {
filledChange, err := fillStringTemplate(change, map[string]string{"DatabaseName": tabletInfo.DbName()})
if err != nil {
return fmt.Errorf("fillStringTemplate failed: %v", err)
}
ctx, cancel := context.WithTimeout(wr.ctx, 30*time.Second)
defer cancel()
// Need to make sure that we enable binlog, since we're only applying the statement on masters.
_, err = wr.tmc.ExecuteFetch(ctx, tabletInfo, filledChange, 0, false, false)
return err
}
func (client *GoRpcTabletManagerClient) Restore(ctx context.Context, tablet *topo.TabletInfo, sa *actionnode.RestoreArgs, waitTime time.Duration) (<-chan *logutil.LoggerEvent, tmclient.ErrFunc, error) {
rpcClient, err := bsonrpc.DialHTTP("tcp", tablet.Addr(), waitTime, nil)
if err != nil {
return nil, nil, err
}
logstream := make(chan *logutil.LoggerEvent, 10)
c := rpcClient.StreamGo("TabletManager.Restore", sa, logstream)
return logstream, func() error {
rpcClient.Close()
return c.Error
}, nil
}
func (zkts *Server) UpdateTablet(tablet *topo.TabletInfo, existingVersion int64) (int64, error) {
zkTabletPath := TabletPathForAlias(tablet.Alias)
stat, err := zkts.zconn.Set(zkTabletPath, tablet.Json(), int(existingVersion))
if err != nil {
if zookeeper.IsError(err, zookeeper.ZBADVERSION) {
err = topo.ErrBadVersion
} else if zookeeper.IsError(err, zookeeper.ZNONODE) {
err = topo.ErrNoNode
}
return 0, err
}
event.Dispatch(&events.TabletChange{
Tablet: *tablet.Tablet,
Status: "updated",
})
return int64(stat.Version()), nil
}
// executeFetchLoop loops over the provided insertChannel
// and sends the commands to the provided tablet.
func executeFetchLoop(wr *wrangler.Wrangler, ti *topo.TabletInfo, insertChannel chan string, abort chan struct{}, disableBinLogs bool) error {
for {
select {
case cmd, ok := <-insertChannel:
if !ok {
// no more to read, we're done
return nil
}
cmd = "INSERT INTO `" + ti.DbName() + "`." + cmd
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
_, err := wr.TabletManagerClient().ExecuteFetch(ctx, ti, cmd, 0, false, disableBinLogs)
if err != nil {
return fmt.Errorf("ExecuteFetch failed: %v", err)
}
cancel()
case <-abort:
// FIXME(alainjobart): note this select case
// could be starved here, and we might miss
// the abort in some corner cases.
return nil
}
}
}
// change a tablet type to RESTORE and set all the other arguments.
// from now on, we can go to:
// - back to IDLE if we don't use the tablet at all (after for instance
// a successful ReserveForRestore but a failed Snapshot)
// - to SCRAP if something in the process on the target host fails
// - to SPARE if the clone works
func (agent *ActionAgent) changeTypeToRestore(ctx context.Context, tablet, sourceTablet *topo.TabletInfo, parentAlias topo.TabletAlias, keyRange key.KeyRange) error {
// run the optional preflight_assigned hook
hk := hook.NewSimpleHook("preflight_assigned")
topotools.ConfigureTabletHook(hk, agent.TabletAlias)
if err := hk.ExecuteOptional(); err != nil {
return err
}
// change the type
tablet.Parent = parentAlias
tablet.Keyspace = sourceTablet.Keyspace
tablet.Shard = sourceTablet.Shard
tablet.Type = topo.TYPE_RESTORE
tablet.KeyRange = keyRange
tablet.DbNameOverride = sourceTablet.DbNameOverride
if err := topo.UpdateTablet(ctx, agent.TopoServer, tablet); err != nil {
return err
}
// and create the replication graph items
return topo.UpdateTabletReplicationData(ctx, agent.TopoServer, tablet.Tablet)
}
// findChunks returns an array of chunks to use for splitting up a table
// into multiple data chunks. It only works for tables with a primary key
// (and the primary key first column is an integer type).
// The array will always look like:
// "", "value1", "value2", ""
// A non-split tablet will just return:
// "", ""
func findChunks(wr *wrangler.Wrangler, ti *topo.TabletInfo, td *myproto.TableDefinition, minTableSizeForSplit uint64, sourceReaderCount int) ([]string, error) {
result := []string{"", ""}
// eliminate a few cases we don't split tables for
if len(td.PrimaryKeyColumns) == 0 {
// no primary key, what can we do?
return result, nil
}
if td.DataLength < minTableSizeForSplit {
// table is too small to split up
return result, nil
}
// get the min and max of the leading column of the primary key
query := fmt.Sprintf("SELECT MIN(%v), MAX(%v) FROM %v.%v", td.PrimaryKeyColumns[0], td.PrimaryKeyColumns[0], ti.DbName(), td.Name)
ctx, cancel := context.WithTimeout(context.TODO(), 30*time.Second)
qr, err := wr.TabletManagerClient().ExecuteFetch(ctx, ti, query, 1, true, false)
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks: %v", td.Name, err)
return result, nil
}
cancel()
if len(qr.Rows) != 1 {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot get min and max", td.Name)
return result, nil
}
if qr.Rows[0][0].IsNull() || qr.Rows[0][1].IsNull() {
wr.Logger().Infof("Not splitting table %v into multiple chunks, min or max is NULL: %v %v", td.Name, qr.Rows[0][0], qr.Rows[0][1])
return result, nil
}
switch qr.Fields[0].Type {
case mproto.VT_TINY, mproto.VT_SHORT, mproto.VT_LONG, mproto.VT_LONGLONG, mproto.VT_INT24:
minNumeric := sqltypes.MakeNumeric(qr.Rows[0][0].Raw())
maxNumeric := sqltypes.MakeNumeric(qr.Rows[0][1].Raw())
if qr.Rows[0][0].Raw()[0] == '-' {
// signed values, use int64
min, err := minNumeric.ParseInt64()
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert min: %v %v", td.Name, minNumeric, err)
return result, nil
}
max, err := maxNumeric.ParseInt64()
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert max: %v %v", td.Name, maxNumeric, err)
return result, nil
}
interval := (max - min) / int64(sourceReaderCount)
if interval == 0 {
wr.Logger().Infof("Not splitting table %v into multiple chunks, interval=0: %v %v", td.Name, max, min)
return result, nil
}
result = make([]string, sourceReaderCount+1)
result[0] = ""
result[sourceReaderCount] = ""
for i := int64(1); i < int64(sourceReaderCount); i++ {
result[i] = fmt.Sprintf("%v", min+interval*i)
}
return result, nil
}
// unsigned values, use uint64
min, err := minNumeric.ParseUint64()
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert min: %v %v", td.Name, minNumeric, err)
return result, nil
}
max, err := maxNumeric.ParseUint64()
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert max: %v %v", td.Name, maxNumeric, err)
return result, nil
}
interval := (max - min) / uint64(sourceReaderCount)
if interval == 0 {
wr.Logger().Infof("Not splitting table %v into multiple chunks, interval=0: %v %v", td.Name, max, min)
return result, nil
}
result = make([]string, sourceReaderCount+1)
result[0] = ""
result[sourceReaderCount] = ""
for i := uint64(1); i < uint64(sourceReaderCount); i++ {
result[i] = fmt.Sprintf("%v", min+interval*i)
}
return result, nil
case mproto.VT_FLOAT, mproto.VT_DOUBLE:
min, err := strconv.ParseFloat(qr.Rows[0][0].String(), 64)
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert min: %v %v", td.Name, qr.Rows[0][0], err)
return result, nil
}
max, err := strconv.ParseFloat(qr.Rows[0][1].String(), 64)
if err != nil {
wr.Logger().Infof("Not splitting table %v into multiple chunks, cannot convert max: %v %v", td.Name, qr.Rows[0][1].String(), err)
return result, nil
}
interval := (max - min) / float64(sourceReaderCount)
if interval == 0 {
wr.Logger().Infof("Not splitting table %v into multiple chunks, interval=0: %v %v", td.Name, max, min)
return result, nil
}
result = make([]string, sourceReaderCount+1)
result[0] = ""
result[sourceReaderCount] = ""
for i := 1; i < sourceReaderCount; i++ {
result[i] = fmt.Sprintf("%v", min+interval*float64(i))
}
return result, nil
}
wr.Logger().Infof("Not splitting table %v into multiple chunks, primary key not numeric", td.Name)
return result, nil
}