func main() {
flag.Parse()
args := flag.Args()
installSignalHandlers()
servenv.Init()
defer servenv.Close()
ts := topo.GetServer()
defer topo.CloseServers()
// the logger will be replaced when we start a job
wr = wrangler.New(logutil.NewConsoleLogger(), ts, 30*time.Second, 30*time.Second)
if len(args) == 0 {
// interactive mode, initialize the web UI to chose a command
initInteractiveMode()
} else {
// single command mode, just runs it
runCommand(args)
}
initStatusHandling()
servenv.RunDefault()
}
func (ar *ActionRepository) ApplyTabletAction(actionName string, tabletAlias topo.TabletAlias, r *http.Request) *ActionResult {
result := &ActionResult{Name: actionName, Parameters: tabletAlias.String()}
action, ok := ar.tabletActions[actionName]
if !ok {
result.error("Unknown tablet action")
return result
}
// check the role
if action.role != "" {
if err := acl.CheckAccessHTTP(r, action.role); err != nil {
result.error("Access denied")
return result
}
}
// run the action
wr := wrangler.New(logutil.NewConsoleLogger(), ar.ts, *actionTimeout, *lockTimeout)
output, err := action.method(wr, tabletAlias, r)
if err != nil {
result.error(err.Error())
return result
}
result.Output = output
return result
}
// ExecuteVtctlCommand is the server side method that will execute the query,
// and stream the results.
func (s *VtctlServer) ExecuteVtctlCommand(context context.Context, query *gorpcproto.ExecuteVtctlCommandArgs, sendReply func(interface{}) error) error {
// create a logger, send the result back to the caller
logstream := logutil.NewChannelLogger(10)
logger := logutil.NewTeeLogger(logstream, logutil.NewConsoleLogger())
// send logs to the caller
wg := sync.WaitGroup{}
wg.Add(1)
go func() {
for e := range logstream {
// Note we don't interrupt the loop here, as
// we still need to flush and finish the
// command, even if the channel to the client
// has been broken. We'll just keep trying.
sendReply(&e)
}
wg.Done()
}()
// create the wrangler
wr := wrangler.New(logger, s.ts, query.ActionTimeout, query.LockTimeout)
// execute the command
err := vtctl.RunCommand(wr, query.Args)
// close the log channel, and wait for them all to be sent
close(logstream)
wg.Wait()
return err
}
func init() {
logger := logutil.NewConsoleLogger()
flag.CommandLine.SetOutput(logutil.NewLoggerWriter(logger))
flag.Usage = func() {
logger.Printf("Usage: %s [global parameters] command [command parameters]\n", os.Args[0])
logger.Printf("\nThe global optional parameters are:\n")
flag.PrintDefaults()
logger.Printf("\nThe commands are listed below, sorted by group. Use '%s <command> -h' for more help.\n\n", os.Args[0])
vtctl.PrintAllCommands(logger)
}
}
// rebuildShardIfNeeded will rebuild the serving graph if we need to
func (agent *ActionAgent) rebuildShardIfNeeded(tablet *topo.TabletInfo, targetTabletType topo.TabletType) error {
if topo.IsInServingGraph(targetTabletType) {
// TODO: interrupted may need to be a global one closed when we exit
interrupted := make(chan struct{})
// no need to take the shard lock in this case
if _, err := topotools.RebuildShard(context.TODO(), logutil.NewConsoleLogger(), agent.TopoServer, tablet.Keyspace, tablet.Shard, []string{tablet.Alias.Cell}, agent.LockTimeout, interrupted); err != nil {
return fmt.Errorf("topotools.RebuildShard returned an error: %v", err)
}
}
return nil
}
func snapshotCmd(mysqld *mysqlctl.Mysqld, subFlags *flag.FlagSet, args []string) {
concurrency := subFlags.Int("concurrency", 4, "how many compression jobs to run simultaneously")
subFlags.Parse(args)
if subFlags.NArg() != 1 {
log.Fatalf("Command snapshot requires <db name>")
}
filename, _, _, err := mysqld.CreateSnapshot(logutil.NewConsoleLogger(), subFlags.Arg(0), tabletAddr, false, *concurrency, false, nil)
if err != nil {
log.Fatalf("snapshot failed: %v", err)
} else {
log.Infof("manifest location: %v", filename)
}
}
func snapshotSourceStartCmd(mysqld *mysqlctl.Mysqld, subFlags *flag.FlagSet, args []string) {
concurrency := subFlags.Int("concurrency", 4, "how many checksum jobs to run simultaneously")
subFlags.Parse(args)
if subFlags.NArg() != 1 {
log.Fatalf("Command snapshotsourcestart requires <db name>")
}
filename, slaveStartRequired, readOnly, err := mysqld.CreateSnapshot(logutil.NewConsoleLogger(), subFlags.Arg(0), tabletAddr, false, *concurrency, true, nil)
if err != nil {
log.Fatalf("snapshot failed: %v", err)
} else {
log.Infof("manifest location: %v", filename)
log.Infof("slave start required: %v", slaveStartRequired)
log.Infof("read only: %v", readOnly)
}
}
func restoreCmd(mysqld *mysqlctl.Mysqld, subFlags *flag.FlagSet, args []string) {
dontWaitForSlaveStart := subFlags.Bool("dont_wait_for_slave_start", false, "won't wait for replication to start (useful when restoring from master server)")
fetchConcurrency := subFlags.Int("fetch_concurrency", 3, "how many files to fetch simultaneously")
fetchRetryCount := subFlags.Int("fetch_retry_count", 3, "how many times to retry a failed transfer")
subFlags.Parse(args)
if subFlags.NArg() != 1 {
log.Fatalf("Command restore requires <snapshot manifest file>")
}
rs, err := mysqlctl.ReadSnapshotManifest(subFlags.Arg(0))
if err == nil {
err = mysqld.RestoreFromSnapshot(logutil.NewConsoleLogger(), rs, *fetchConcurrency, *fetchRetryCount, *dontWaitForSlaveStart, nil)
}
if err != nil {
log.Fatalf("restore failed: %v", err)
}
}
func TestCopySchemaShard(t *testing.T) {
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
sourceMaster := NewFakeTablet(t, wr, "cell1", 0,
topo.TYPE_MASTER, TabletKeyspaceShard(t, "ks", "-80"))
sourceRdonly := NewFakeTablet(t, wr, "cell1", 1,
topo.TYPE_RDONLY, TabletKeyspaceShard(t, "ks", "-80"),
TabletParent(sourceMaster.Tablet.Alias))
destinationMaster := NewFakeTablet(t, wr, "cell1", 10,
topo.TYPE_MASTER, TabletKeyspaceShard(t, "ks", "-40"))
// one destination RdOnly, so we know that schema copies propogate from masters
destinationRdonly := NewFakeTablet(t, wr, "cell1", 11,
topo.TYPE_RDONLY, TabletKeyspaceShard(t, "ks", "-40"),
TabletParent(destinationMaster.Tablet.Alias))
for _, ft := range []*FakeTablet{sourceMaster, sourceRdonly, destinationMaster, destinationRdonly} {
ft.StartActionLoop(t, wr)
defer ft.StopActionLoop(t)
}
sourceRdonly.FakeMysqlDaemon.Schema = &myproto.SchemaDefinition{
DatabaseSchema: "CREATE DATABASE `{{.DatabaseName}}` /*!40100 DEFAULT CHARACTER SET utf8 */",
TableDefinitions: []*myproto.TableDefinition{
&myproto.TableDefinition{
Name: "table1",
Schema: "CREATE TABLE `resharding1` (\n `id` bigint(20) NOT NULL AUTO_INCREMENT,\n `msg` varchar(64) DEFAULT NULL,\n `keyspace_id` bigint(20) unsigned NOT NULL,\n PRIMARY KEY (`id`),\n KEY `by_msg` (`msg`)\n) ENGINE=InnoDB DEFAULT CHARSET=utf8",
Type: myproto.TABLE_BASE_TABLE,
},
&myproto.TableDefinition{
Name: "view1",
Schema: "CREATE TABLE `view1` (\n `id` bigint(20) NOT NULL AUTO_INCREMENT,\n `msg` varchar(64) DEFAULT NULL,\n `keyspace_id` bigint(20) unsigned NOT NULL,\n PRIMARY KEY (`id`),\n KEY `by_msg` (`msg`)\n) ENGINE=InnoDB DEFAULT CHARSET=utf8",
Type: myproto.TABLE_VIEW,
},
},
}
destinationMaster.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t)
destinationRdonly.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t)
if err := wr.CopySchemaShard(sourceRdonly.Tablet.Alias, nil, nil, true, "ks", "-40"); err != nil {
t.Fatalf("CopySchemaShard failed: %v", err)
}
}
// TestTabletExternallyReparentedWithDifferentMysqlPort makes sure
// that if mysql is restarted on the master-elect tablet and has a different
// port, we pick it up correctly.
func TestTabletExternallyReparentedWithDifferentMysqlPort(t *testing.T) {
ts := zktopo.NewTestServer(t, []string{"cell1"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
// Create an old master, a new master, two good slaves, one bad slave
oldMaster := NewFakeTablet(t, wr, "cell1", 0, topo.TYPE_MASTER)
newMaster := NewFakeTablet(t, wr, "cell1", 1, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave := NewFakeTablet(t, wr, "cell1", 2, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
// Now we're restarting mysql on a different port, 3301->3303
// but without updating the Tablet record in topology.
// On the elected master, we will respond to
// TABLET_ACTION_SLAVE_WAS_PROMOTED, so we need a MysqlDaemon
// that returns no master, and the new port (as returned by mysql)
newMaster.FakeMysqlDaemon.MasterAddr = ""
newMaster.FakeMysqlDaemon.MysqlPort = 3303
newMaster.StartActionLoop(t, wr)
defer newMaster.StopActionLoop(t)
// On the old master, we will only respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED and point to the new mysql port
oldMaster.FakeMysqlDaemon.MasterAddr = "101.0.0.1:3303"
oldMaster.StartActionLoop(t, wr)
defer oldMaster.StopActionLoop(t)
// On the good slaves, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED and point to the new mysql port
goodSlave.FakeMysqlDaemon.MasterAddr = "101.0.0.1:3303"
goodSlave.StartActionLoop(t, wr)
defer goodSlave.StopActionLoop(t)
// This tests the good case, where everything works as planned
t.Logf("TabletExternallyReparented(new master) expecting success")
tmc := tmclient.NewTabletManagerClient()
ti, err := ts.GetTablet(newMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := tmc.TabletExternallyReparented(wr.Context(), ti, ""); err != nil {
t.Fatalf("TabletExternallyReparented(replica) failed: %v", err)
}
}
func (ar *ActionRepository) ApplyKeyspaceAction(actionName, keyspace string, r *http.Request) *ActionResult {
result := &ActionResult{Name: actionName, Parameters: keyspace}
action, ok := ar.keyspaceActions[actionName]
if !ok {
result.error("Unknown keyspace action")
return result
}
wr := wrangler.New(logutil.NewConsoleLogger(), ar.ts, *actionTimeout, *lockTimeout)
output, err := action(wr, keyspace, r)
if err != nil {
result.error(err.Error())
return result
}
result.Output = output
return result
}
func main() {
flag.Parse()
servenv.Init()
ts := topo.GetServer()
scheduler, err := janitor.New(*keyspace, *shard, ts, wrangler.New(logutil.NewConsoleLogger(), ts, *actionTimeout, *lockTimeout), *sleepTime)
if err != nil {
log.Fatalf("janitor.New: %v", err)
}
if len(activeModules)+len(dryRunModules) < 1 {
log.Fatal("no modules to run specified")
}
scheduler.Enable(activeModules)
scheduler.EnableDryRun(dryRunModules)
go scheduler.Run()
servenv.RunDefault()
}
// setAndStartWorker will set the current worker.
// We always log to both memory logger (for display on the web) and
// console logger (for records / display of command line worker).
func setAndStartWorker(wrk worker.Worker) (chan struct{}, error) {
currentWorkerMutex.Lock()
defer currentWorkerMutex.Unlock()
if currentWorker != nil {
return nil, fmt.Errorf("A worker is already in progress: %v", currentWorker)
}
currentWorker = wrk
currentMemoryLogger = logutil.NewMemoryLogger()
currentDone = make(chan struct{})
wr.SetLogger(logutil.NewTeeLogger(currentMemoryLogger, logutil.NewConsoleLogger()))
// one go function runs the worker, closes 'done' when done
go func() {
log.Infof("Starting worker...")
wrk.Run()
close(currentDone)
}()
return currentDone, nil
}
func (ar *ActionRepository) ApplyShardAction(actionName, keyspace, shard string, r *http.Request) *ActionResult {
// if the shard name contains a '-', we assume it's the
// name for a ranged based shard, so we lower case it.
if strings.Contains(shard, "-") {
shard = strings.ToLower(shard)
}
result := &ActionResult{Name: actionName, Parameters: keyspace + "/" + shard}
action, ok := ar.shardActions[actionName]
if !ok {
result.error("Unknown shard action")
return result
}
wr := wrangler.New(logutil.NewConsoleLogger(), ar.ts, *actionTimeout, *lockTimeout)
output, err := action(wr, keyspace, shard, r)
if err != nil {
result.error(err.Error())
return result
}
result.Output = output
return result
}
func main() {
defer exit.RecoverAll()
defer logutil.Flush()
flag.Parse()
args := flag.Args()
if len(args) == 0 {
flag.Usage()
exit.Return(1)
}
action := args[0]
installSignalHandlers()
startMsg := fmt.Sprintf("USER=%v SUDO_USER=%v %v", os.Getenv("USER"), os.Getenv("SUDO_USER"), strings.Join(os.Args, " "))
if syslogger, err := syslog.New(syslog.LOG_INFO, "vtctl "); err == nil {
syslogger.Info(startMsg)
} else {
log.Warningf("cannot connect to syslog: %v", err)
}
topoServer := topo.GetServer()
defer topo.CloseServers()
wr := wrangler.New(logutil.NewConsoleLogger(), topoServer, *waitTime, *lockWaitTimeout)
err := vtctl.RunCommand(wr, args)
switch err {
case vtctl.ErrUnknownCommand:
flag.Usage()
exit.Return(1)
case nil:
// keep going
default:
log.Errorf("action failed: %v %v", action, err)
exit.Return(255)
}
}
func TestTabletExternallyReparented(t *testing.T) {
ctx := context.Background()
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
// Create an old master, a new master, two good slaves, one bad slave
oldMaster := NewFakeTablet(t, wr, "cell1", 0, topo.TYPE_MASTER)
newMaster := NewFakeTablet(t, wr, "cell1", 1, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave1 := NewFakeTablet(t, wr, "cell1", 2, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave2 := NewFakeTablet(t, wr, "cell2", 3, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
badSlave := NewFakeTablet(t, wr, "cell1", 4, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
// Add a new Cell to the Shard, that doesn't map to any read topo cell,
// to simulate a data center being unreachable.
si, err := ts.GetShard("test_keyspace", "0")
if err != nil {
t.Fatalf("GetShard failed: %v", err)
}
si.Cells = append(si.Cells, "cell666")
if err := topo.UpdateShard(ctx, ts, si); err != nil {
t.Fatalf("UpdateShard failed: %v", err)
}
// Slightly unrelated test: make sure we can find the tablets
// even with a datacenter being down.
tabletMap, err := topo.GetTabletMapForShardByCell(ctx, ts, "test_keyspace", "0", []string{"cell1"})
if err != nil {
t.Fatalf("GetTabletMapForShardByCell should have worked but got: %v", err)
}
master, err := topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != nil || master != oldMaster.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
slave1, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave1.Tablet.IPAddr, "vt", goodSlave1.Tablet.Portmap["vt"])
if err != nil || slave1 != goodSlave1.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(slave1) failed: %v %v", err, master)
}
slave2, err := topotools.FindTabletByIPAddrAndPort(tabletMap, goodSlave2.Tablet.IPAddr, "vt", goodSlave2.Tablet.Portmap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(slave2) worked: %v %v", err, slave2)
}
// Make sure the master is not exported in other cells
tabletMap, err = topo.GetTabletMapForShardByCell(ctx, ts, "test_keyspace", "0", []string{"cell2"})
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != topo.ErrNoNode {
t.Fatalf("FindTabletByIPAddrAndPort(master) worked in cell2: %v %v", err, master)
}
tabletMap, err = topo.GetTabletMapForShard(ctx, ts, "test_keyspace", "0")
if err != topo.ErrPartialResult {
t.Fatalf("GetTabletMapForShard should have returned ErrPartialResult but got: %v", err)
}
master, err = topotools.FindTabletByIPAddrAndPort(tabletMap, oldMaster.Tablet.IPAddr, "vt", oldMaster.Tablet.Portmap["vt"])
if err != nil || master != oldMaster.Tablet.Alias {
t.Fatalf("FindTabletByIPAddrAndPort(master) failed: %v %v", err, master)
}
// On the elected master, we will respond to
// TABLET_ACTION_SLAVE_WAS_PROMOTED
newMaster.FakeMysqlDaemon.MasterAddr = ""
newMaster.StartActionLoop(t, wr)
defer newMaster.StopActionLoop(t)
// On the old master, we will only respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED.
oldMaster.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
oldMaster.StartActionLoop(t, wr)
defer oldMaster.StopActionLoop(t)
// On the good slaves, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED.
goodSlave1.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
goodSlave1.StartActionLoop(t, wr)
defer goodSlave1.StopActionLoop(t)
goodSlave2.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
goodSlave2.StartActionLoop(t, wr)
defer goodSlave2.StopActionLoop(t)
// On the bad slave, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED with bad data.
badSlave.FakeMysqlDaemon.MasterAddr = "234.0.0.1:3301"
badSlave.StartActionLoop(t, wr)
defer badSlave.StopActionLoop(t)
// First test: reparent to the same master, make sure it works
// as expected.
tmc := tmclient.NewTabletManagerClient()
ti, err := ts.GetTablet(oldMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := tmc.TabletExternallyReparented(wr.Context(), ti, ""); err != nil {
t.Fatalf("TabletExternallyReparented(same master) should have worked")
}
// Second test: reparent to a replica, and pretend the old
// master is still good to go.
// This tests a bad case; the new designated master is a slave,
// but we should do what we're told anyway
ti, err = ts.GetTablet(goodSlave1.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := tmc.TabletExternallyReparented(wr.Context(), ti, ""); err != nil {
t.Fatalf("TabletExternallyReparented(slave) error: %v", err)
}
// This tests the good case, where everything works as planned
t.Logf("TabletExternallyReparented(new master) expecting success")
ti, err = ts.GetTablet(newMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := tmc.TabletExternallyReparented(wr.Context(), ti, ""); err != nil {
t.Fatalf("TabletExternallyReparented(replica) failed: %v", err)
}
// Now double-check the serving graph is good.
// Should only have one good replica left.
addrs, err := ts.GetEndPoints("cell1", "test_keyspace", "0", topo.TYPE_REPLICA)
if err != nil {
t.Fatalf("GetEndPoints failed at the end: %v", err)
}
if len(addrs.Entries) != 1 {
t.Fatalf("GetEndPoints has too many entries: %v", addrs)
}
}
func main() {
flag.Parse()
servenv.Init()
defer servenv.Close()
templateLoader = NewTemplateLoader(*templateDir, dummyTemplate, *debug)
ts = topo.GetServer()
defer topo.CloseServers()
wr := wrangler.New(logutil.NewConsoleLogger(), ts, 30*time.Second, 30*time.Second)
actionRepo = NewActionRepository(ts)
// keyspace actions
actionRepo.RegisterKeyspaceAction("ValidateKeyspace",
func(wr *wrangler.Wrangler, keyspace string, r *http.Request) (string, error) {
return "", wr.ValidateKeyspace(keyspace, false)
})
actionRepo.RegisterKeyspaceAction("ValidateSchemaKeyspace",
func(wr *wrangler.Wrangler, keyspace string, r *http.Request) (string, error) {
return "", wr.ValidateSchemaKeyspace(keyspace, nil, false)
})
actionRepo.RegisterKeyspaceAction("ValidateVersionKeyspace",
func(wr *wrangler.Wrangler, keyspace string, r *http.Request) (string, error) {
return "", wr.ValidateVersionKeyspace(keyspace)
})
actionRepo.RegisterKeyspaceAction("ValidatePermissionsKeyspace",
func(wr *wrangler.Wrangler, keyspace string, r *http.Request) (string, error) {
return "", wr.ValidatePermissionsKeyspace(keyspace)
})
// shard actions
actionRepo.RegisterShardAction("ValidateShard",
func(wr *wrangler.Wrangler, keyspace, shard string, r *http.Request) (string, error) {
return "", wr.ValidateShard(keyspace, shard, false)
})
actionRepo.RegisterShardAction("ValidateSchemaShard",
func(wr *wrangler.Wrangler, keyspace, shard string, r *http.Request) (string, error) {
return "", wr.ValidateSchemaShard(keyspace, shard, nil, false)
})
actionRepo.RegisterShardAction("ValidateVersionShard",
func(wr *wrangler.Wrangler, keyspace, shard string, r *http.Request) (string, error) {
return "", wr.ValidateVersionShard(keyspace, shard)
})
actionRepo.RegisterShardAction("ValidatePermissionsShard",
func(wr *wrangler.Wrangler, keyspace, shard string, r *http.Request) (string, error) {
return "", wr.ValidatePermissionsShard(keyspace, shard)
})
// tablet actions
actionRepo.RegisterTabletAction("Ping", "",
func(wr *wrangler.Wrangler, tabletAlias topo.TabletAlias, r *http.Request) (string, error) {
ti, err := wr.TopoServer().GetTablet(tabletAlias)
if err != nil {
return "", err
}
return "", wr.TabletManagerClient().Ping(wr.Context(), ti)
})
actionRepo.RegisterTabletAction("ScrapTablet", acl.ADMIN,
func(wr *wrangler.Wrangler, tabletAlias topo.TabletAlias, r *http.Request) (string, error) {
// refuse to scrap tablets that are not spare
ti, err := wr.TopoServer().GetTablet(tabletAlias)
if err != nil {
return "", err
}
if ti.Type != topo.TYPE_SPARE {
return "", fmt.Errorf("Can only scrap spare tablets")
}
return "", wr.Scrap(tabletAlias, false, false)
})
actionRepo.RegisterTabletAction("ScrapTabletForce", acl.ADMIN,
func(wr *wrangler.Wrangler, tabletAlias topo.TabletAlias, r *http.Request) (string, error) {
// refuse to scrap tablets that are not spare
ti, err := wr.TopoServer().GetTablet(tabletAlias)
if err != nil {
return "", err
}
if ti.Type != topo.TYPE_SPARE {
return "", fmt.Errorf("Can only scrap spare tablets")
}
return "", wr.Scrap(tabletAlias, true, false)
})
actionRepo.RegisterTabletAction("DeleteTablet", acl.ADMIN,
func(wr *wrangler.Wrangler, tabletAlias topo.TabletAlias, r *http.Request) (string, error) {
return "", wr.DeleteTablet(tabletAlias)
})
// toplevel index
http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
templateLoader.ServeTemplate("index.html", indexContent, w, r)
})
// keyspace actions
http.HandleFunc("/keyspace_actions", func(w http.ResponseWriter, r *http.Request) {
if err := r.ParseForm(); err != nil {
httpError(w, "cannot parse form: %s", err)
return
}
action := r.FormValue("action")
if action == "" {
http.Error(w, "no action provided", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "no keyspace provided", http.StatusBadRequest)
return
}
result := actionRepo.ApplyKeyspaceAction(action, keyspace, r)
templateLoader.ServeTemplate("action.html", result, w, r)
})
// shard actions
http.HandleFunc("/shard_actions", func(w http.ResponseWriter, r *http.Request) {
if err := r.ParseForm(); err != nil {
httpError(w, "cannot parse form: %s", err)
return
}
action := r.FormValue("action")
if action == "" {
http.Error(w, "no action provided", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "no keyspace provided", http.StatusBadRequest)
return
}
shard := r.FormValue("shard")
if shard == "" {
http.Error(w, "no shard provided", http.StatusBadRequest)
return
}
result := actionRepo.ApplyShardAction(action, keyspace, shard, r)
templateLoader.ServeTemplate("action.html", result, w, r)
})
// tablet actions
http.HandleFunc("/tablet_actions", func(w http.ResponseWriter, r *http.Request) {
if err := r.ParseForm(); err != nil {
httpError(w, "cannot parse form: %s", err)
return
}
action := r.FormValue("action")
if action == "" {
http.Error(w, "no action provided", http.StatusBadRequest)
return
}
alias := r.FormValue("alias")
if alias == "" {
http.Error(w, "no alias provided", http.StatusBadRequest)
return
}
tabletAlias, err := topo.ParseTabletAliasString(alias)
if err != nil {
http.Error(w, "bad alias provided", http.StatusBadRequest)
return
}
result := actionRepo.ApplyTabletAction(action, tabletAlias, r)
templateLoader.ServeTemplate("action.html", result, w, r)
})
// topology server
http.HandleFunc("/dbtopo", func(w http.ResponseWriter, r *http.Request) {
if err := r.ParseForm(); err != nil {
httpError(w, "cannot parse form: %s", err)
return
}
result := DbTopologyResult{}
topology, err := topotools.DbTopology(context.TODO(), wr.TopoServer())
if err != nil {
result.Error = err.Error()
} else {
result.Topology = topology
}
templateLoader.ServeTemplate("dbtopo.html", result, w, r)
})
// serving graph
http.HandleFunc("/serving_graph/", func(w http.ResponseWriter, r *http.Request) {
parts := strings.Split(r.URL.Path, "/")
cell := parts[len(parts)-1]
if cell == "" {
cells, err := ts.GetKnownCells()
if err != nil {
httpError(w, "cannot get known cells: %v", err)
return
} else {
templateLoader.ServeTemplate("serving_graph_cells.html", cells, w, r)
}
return
}
servingGraph := topotools.DbServingGraph(wr.TopoServer(), cell)
templateLoader.ServeTemplate("serving_graph.html", servingGraph, w, r)
})
// redirects for explorers
http.HandleFunc("/explorers/redirect", func(w http.ResponseWriter, r *http.Request) {
if err := r.ParseForm(); err != nil {
httpError(w, "cannot parse form: %s", err)
return
}
var explorer Explorer
switch len(explorers) {
case 0:
http.Error(w, "no explorer configured", http.StatusInternalServerError)
return
case 1:
for _, ex := range explorers {
explorer = ex
}
default:
explorerName := r.FormValue("explorer")
var ok bool
explorer, ok = explorers[explorerName]
if !ok {
http.Error(w, "bad explorer name", http.StatusBadRequest)
return
}
}
var target string
switch r.FormValue("type") {
case "keyspace":
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetKeyspacePath(keyspace)
case "shard":
keyspace, shard := r.FormValue("keyspace"), r.FormValue("shard")
if keyspace == "" || shard == "" {
http.Error(w, "keyspace and shard are obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetShardPath(keyspace, shard)
case "srv_keyspace":
cell := r.FormValue("cell")
if cell == "" {
http.Error(w, "cell is obligatory for this redirect", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetSrvKeyspacePath(cell, keyspace)
case "srv_shard":
cell := r.FormValue("cell")
if cell == "" {
http.Error(w, "cell is obligatory for this redirect", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
shard := r.FormValue("shard")
if shard == "" {
http.Error(w, "shard is obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetSrvShardPath(cell, keyspace, shard)
case "srv_type":
cell := r.FormValue("cell")
if cell == "" {
http.Error(w, "cell is obligatory for this redirect", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
shard := r.FormValue("shard")
if shard == "" {
http.Error(w, "shard is obligatory for this redirect", http.StatusBadRequest)
return
}
tabletType := r.FormValue("tablet_type")
if tabletType == "" {
http.Error(w, "tablet_type is obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetSrvTypePath(cell, keyspace, shard, topo.TabletType(tabletType))
case "tablet":
aliasName := r.FormValue("alias")
if aliasName == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
alias, err := topo.ParseTabletAliasString(aliasName)
if err != nil {
http.Error(w, "bad tablet alias", http.StatusBadRequest)
return
}
target = explorer.GetTabletPath(alias)
case "replication":
cell := r.FormValue("cell")
if cell == "" {
http.Error(w, "cell is obligatory for this redirect", http.StatusBadRequest)
return
}
keyspace := r.FormValue("keyspace")
if keyspace == "" {
http.Error(w, "keyspace is obligatory for this redirect", http.StatusBadRequest)
return
}
shard := r.FormValue("shard")
if shard == "" {
http.Error(w, "shard is obligatory for this redirect", http.StatusBadRequest)
return
}
target = explorer.GetReplicationSlaves(cell, keyspace, shard)
default:
http.Error(w, "bad redirect type", http.StatusBadRequest)
return
}
http.Redirect(w, r, target, http.StatusFound)
})
servenv.RunDefault()
}
func TestTabletExternallyReparentedFailedOldMaster(t *testing.T) {
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
// Create an old master, a new master, two good slaves
oldMaster := NewFakeTablet(t, wr, "cell1", 0, topo.TYPE_MASTER)
newMaster := NewFakeTablet(t, wr, "cell1", 1, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
goodSlave := NewFakeTablet(t, wr, "cell1", 2, topo.TYPE_REPLICA,
TabletParent(oldMaster.Tablet.Alias))
// Reparent to a replica, and pretend the old master is not responding
// On the elected master, we will respond to
// TABLET_ACTION_SLAVE_WAS_PROMOTED
newMaster.FakeMysqlDaemon.MasterAddr = ""
newMaster.StartActionLoop(t, wr)
defer newMaster.StopActionLoop(t)
// On the old master, we will only get a
// TABLET_ACTION_SLAVE_WAS_RESTARTED call, let's just not
// respond to it at all
// On the good slave, we will respond to
// TABLET_ACTION_SLAVE_WAS_RESTARTED.
goodSlave.FakeMysqlDaemon.MasterAddr = newMaster.Tablet.MysqlIpAddr()
goodSlave.StartActionLoop(t, wr)
defer goodSlave.StopActionLoop(t)
// The reparent should work as expected here
t.Logf("TabletExternallyReparented(new master) expecting success")
tmc := tmclient.NewTabletManagerClient()
ti, err := ts.GetTablet(newMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet failed: %v", err)
}
if err := tmc.TabletExternallyReparented(wr.Context(), ti, ""); err != nil {
t.Fatalf("TabletExternallyReparented(replica) failed: %v", err)
}
// Now double-check the serving graph is good.
// Should only have one good replica left.
addrs, err := ts.GetEndPoints("cell1", "test_keyspace", "0", topo.TYPE_REPLICA)
if err != nil {
t.Fatalf("GetEndPoints failed at the end: %v", err)
}
if len(addrs.Entries) != 1 {
t.Fatalf("GetEndPoints has too many entries: %v", addrs)
}
// check the old master was converted to spare
tablet, err := ts.GetTablet(oldMaster.Tablet.Alias)
if err != nil {
t.Fatalf("GetTablet(%v) failed: %v", oldMaster.Tablet.Alias, err)
}
if tablet.Type != topo.TYPE_SPARE {
t.Fatalf("old master should be spare but is: %v", tablet.Type)
}
if tablet.Parent != newMaster.Tablet.Alias {
t.Fatalf("old master has the wrong master, got %v expected %v", tablet.Parent, newMaster.Tablet.Alias)
}
}
func testSplitClone(t *testing.T, strategy string) {
ts := zktopo.NewTestServer(t, []string{"cell1", "cell2"})
wr := wrangler.New(logutil.NewConsoleLogger(), ts, time.Minute, time.Second)
sourceMaster := testlib.NewFakeTablet(t, wr, "cell1", 0,
topo.TYPE_MASTER, testlib.TabletKeyspaceShard(t, "ks", "-80"))
sourceRdonly1 := testlib.NewFakeTablet(t, wr, "cell1", 1,
topo.TYPE_RDONLY, testlib.TabletKeyspaceShard(t, "ks", "-80"),
testlib.TabletParent(sourceMaster.Tablet.Alias))
sourceRdonly2 := testlib.NewFakeTablet(t, wr, "cell1", 2,
topo.TYPE_RDONLY, testlib.TabletKeyspaceShard(t, "ks", "-80"),
testlib.TabletParent(sourceMaster.Tablet.Alias))
leftMaster := testlib.NewFakeTablet(t, wr, "cell1", 10,
topo.TYPE_MASTER, testlib.TabletKeyspaceShard(t, "ks", "-40"))
leftRdonly := testlib.NewFakeTablet(t, wr, "cell1", 11,
topo.TYPE_RDONLY, testlib.TabletKeyspaceShard(t, "ks", "-40"),
testlib.TabletParent(leftMaster.Tablet.Alias))
rightMaster := testlib.NewFakeTablet(t, wr, "cell1", 20,
topo.TYPE_MASTER, testlib.TabletKeyspaceShard(t, "ks", "40-80"))
rightRdonly := testlib.NewFakeTablet(t, wr, "cell1", 21,
topo.TYPE_RDONLY, testlib.TabletKeyspaceShard(t, "ks", "40-80"),
testlib.TabletParent(rightMaster.Tablet.Alias))
for _, ft := range []*testlib.FakeTablet{sourceMaster, sourceRdonly1, sourceRdonly2, leftMaster, leftRdonly, rightMaster, rightRdonly} {
ft.StartActionLoop(t, wr)
defer ft.StopActionLoop(t)
}
// add the topo and schema data we'll need
if err := topo.CreateShard(ts, "ks", "80-"); err != nil {
t.Fatalf("CreateShard(\"-80\") failed: %v", err)
}
if err := wr.SetKeyspaceShardingInfo("ks", "keyspace_id", key.KIT_UINT64, 4, false); err != nil {
t.Fatalf("SetKeyspaceShardingInfo failed: %v", err)
}
if err := wr.RebuildKeyspaceGraph("ks", nil); err != nil {
t.Fatalf("RebuildKeyspaceGraph failed: %v", err)
}
gwrk, err := NewSplitCloneWorker(wr, "cell1", "ks", "-80", nil, strategy, 10 /*sourceReaderCount*/, 4 /*destinationPackCount*/, 1 /*minTableSizeForSplit*/, 10 /*destinationWriterCount*/)
if err != nil {
t.Errorf("Worker creation failed: %v", err)
}
wrk := gwrk.(*SplitCloneWorker)
for _, sourceRdonly := range []*testlib.FakeTablet{sourceRdonly1, sourceRdonly2} {
sourceRdonly.FakeMysqlDaemon.Schema = &myproto.SchemaDefinition{
DatabaseSchema: "",
TableDefinitions: []*myproto.TableDefinition{
&myproto.TableDefinition{
Name: "table1",
Columns: []string{"id", "msg", "keyspace_id"},
PrimaryKeyColumns: []string{"id"},
Type: myproto.TABLE_BASE_TABLE,
// This informs how many rows we can pack into a single insert
DataLength: 2048,
},
},
}
sourceRdonly.FakeMysqlDaemon.DbaConnectionFactory = SourceRdonlyFactory(t)
sourceRdonly.FakeMysqlDaemon.CurrentSlaveStatus = &myproto.ReplicationStatus{
Position: myproto.ReplicationPosition{
GTIDSet: myproto.MariadbGTID{Domain: 12, Server: 34, Sequence: 5678},
},
}
sourceRdonly.RpcServer.Register(&SqlQuery{t: t})
}
// We read 100 source rows. sourceReaderCount is set to 10, so
// we'll have 100/10=10 rows per table chunk.
// destinationPackCount is set to 4, so we take 4 source rows
// at once. So we'll process 4 + 4 + 2 rows to get to 10.
// That means 3 insert statements on each target (each
// containing half of the rows, i.e. 2 + 2 + 1 rows). So 3 * 10
// = 30 insert statements on each destination.
leftMaster.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t, 30)
leftRdonly.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t, 30)
rightMaster.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t, 30)
rightRdonly.FakeMysqlDaemon.DbaConnectionFactory = DestinationsFactory(t, 30)
wrk.Run()
status := wrk.StatusAsText()
t.Logf("Got status: %v", status)
if wrk.err != nil || wrk.state != stateSCDone {
t.Errorf("Worker run failed")
}
}
// Operate on restore tablet.
// Check that the SnapshotManifest is valid and the master has not changed.
// Shutdown mysqld.
// Load the snapshot from source tablet.
// Restart mysqld and replication.
// Put tablet into the replication graph as a spare.
// Should be called under RpcWrapLockAction.
func (agent *ActionAgent) Restore(ctx context.Context, args *actionnode.RestoreArgs, logger logutil.Logger) error {
// read our current tablet, verify its state
tablet, err := agent.TopoServer.GetTablet(agent.TabletAlias)
if err != nil {
return err
}
if args.WasReserved {
if tablet.Type != topo.TYPE_RESTORE {
return fmt.Errorf("expected restore type, not %v", tablet.Type)
}
} else {
if tablet.Type != topo.TYPE_IDLE {
return fmt.Errorf("expected idle type, not %v", tablet.Type)
}
}
// read the source tablet, compute args.SrcFilePath if default
sourceTablet, err := agent.TopoServer.GetTablet(args.SrcTabletAlias)
if err != nil {
return err
}
if strings.ToLower(args.SrcFilePath) == "default" {
args.SrcFilePath = path.Join(mysqlctl.SnapshotURLPath, mysqlctl.SnapshotManifestFile)
}
// read the parent tablet, verify its state
parentTablet, err := agent.TopoServer.GetTablet(args.ParentAlias)
if err != nil {
return err
}
if parentTablet.Type != topo.TYPE_MASTER && parentTablet.Type != topo.TYPE_SNAPSHOT_SOURCE {
return fmt.Errorf("restore expected master or snapshot_source parent: %v %v", parentTablet.Type, args.ParentAlias)
}
// read & unpack the manifest
sm := new(mysqlctl.SnapshotManifest)
if err := fetchAndParseJsonFile(sourceTablet.Addr(), args.SrcFilePath, sm); err != nil {
return err
}
if !args.WasReserved {
if err := agent.changeTypeToRestore(ctx, tablet, sourceTablet, parentTablet.Alias, sourceTablet.KeyRange); err != nil {
return err
}
}
// create the loggers: tee to console and source
l := logutil.NewTeeLogger(logutil.NewConsoleLogger(), logger)
// do the work
if err := agent.Mysqld.RestoreFromSnapshot(l, sm, args.FetchConcurrency, args.FetchRetryCount, args.DontWaitForSlaveStart, agent.hookExtraEnv()); err != nil {
log.Errorf("RestoreFromSnapshot failed (%v), scrapping", err)
if err := topotools.Scrap(agent.TopoServer, agent.TabletAlias, false); err != nil {
log.Errorf("Failed to Scrap after failed RestoreFromSnapshot: %v", err)
}
return err
}
// reload the schema
agent.ReloadSchema(ctx)
// change to TYPE_SPARE, we're done!
return topotools.ChangeType(agent.TopoServer, agent.TabletAlias, topo.TYPE_SPARE, nil, true)
}
// Snapshot takes a db snapshot
// Should be called under RpcWrapLockAction.
func (agent *ActionAgent) Snapshot(ctx context.Context, args *actionnode.SnapshotArgs, logger logutil.Logger) (*actionnode.SnapshotReply, error) {
// update our type to TYPE_BACKUP
tablet, err := agent.TopoServer.GetTablet(agent.TabletAlias)
if err != nil {
return nil, err
}
originalType := tablet.Type
// ForceMasterSnapshot: Normally a master is not a viable tablet
// to snapshot. However, there are degenerate cases where you need
// to override this, for instance the initial clone of a new master.
if tablet.Type == topo.TYPE_MASTER && args.ForceMasterSnapshot {
// In this case, we don't bother recomputing the serving graph.
// All queries will have to fail anyway.
log.Infof("force change type master -> backup")
// There is a legitimate reason to force in the case of a single
// master.
tablet.Tablet.Type = topo.TYPE_BACKUP
err = topo.UpdateTablet(ctx, agent.TopoServer, tablet)
} else {
err = topotools.ChangeType(agent.TopoServer, tablet.Alias, topo.TYPE_BACKUP, make(map[string]string), true /*runHooks*/)
}
if err != nil {
return nil, err
}
// let's update our internal state (stop query service and other things)
if err := agent.refreshTablet(ctx, "snapshotStart"); err != nil {
return nil, fmt.Errorf("failed to update state before snaphost: %v", err)
}
// create the loggers: tee to console and source
l := logutil.NewTeeLogger(logutil.NewConsoleLogger(), logger)
// now we can run the backup
filename, slaveStartRequired, readOnly, returnErr := agent.Mysqld.CreateSnapshot(l, tablet.DbName(), tablet.Addr(), false, args.Concurrency, args.ServerMode, agent.hookExtraEnv())
// and change our type to the appropriate value
newType := originalType
if returnErr != nil {
log.Errorf("snapshot failed, restoring tablet type back to %v: %v", newType, returnErr)
} else {
if args.ServerMode {
log.Infof("server mode specified, switching tablet to snapshot_source mode")
newType = topo.TYPE_SNAPSHOT_SOURCE
} else {
log.Infof("change type back after snapshot: %v", newType)
}
}
if tablet.Parent.Uid == topo.NO_TABLET && args.ForceMasterSnapshot && newType != topo.TYPE_SNAPSHOT_SOURCE {
log.Infof("force change type backup -> master: %v", tablet.Alias)
tablet.Tablet.Type = topo.TYPE_MASTER
err = topo.UpdateTablet(ctx, agent.TopoServer, tablet)
} else {
err = topotools.ChangeType(agent.TopoServer, tablet.Alias, newType, nil, true /*runHooks*/)
}
if err != nil {
// failure in changing the topology type is probably worse,
// so returning that (we logged the snapshot error anyway)
returnErr = err
}
// if anything failed, don't return anything
if returnErr != nil {
return nil, returnErr
}
// it all worked, return the required information
sr := &actionnode.SnapshotReply{
ManifestPath: filename,
SlaveStartRequired: slaveStartRequired,
ReadOnly: readOnly,
}
if tablet.Parent.Uid == topo.NO_TABLET {
// If this is a master, this will be the new parent.
// FIXME(msolomon) this doesn't work in hierarchical replication.
sr.ParentAlias = tablet.Alias
} else {
sr.ParentAlias = tablet.Parent
}
return sr, nil
}
// tabletExternallyReparentedLocked is called with the shard lock.
// It returns if agent.refreshTablet should be called, and the error.
// Note both are set independently (can have both true and an error).
func (agent *ActionAgent) tabletExternallyReparentedLocked(ctx context.Context, externalID string, interrupted chan struct{}) (bool, error) {
// re-read the tablet record to be sure we have the latest version
tablet, err := agent.TopoServer.GetTablet(agent.TabletAlias)
if err != nil {
return false, err
}
// read the shard, make sure again the master is not already good.
shardInfo, err := agent.TopoServer.GetShard(tablet.Keyspace, tablet.Shard)
if err != nil {
return false, err
}
if shardInfo.MasterAlias == tablet.Alias {
log.Infof("TabletExternallyReparented: tablet became the master before we get the lock?")
return false, nil
}
log.Infof("TabletExternallyReparented called and we're not the master, doing the work")
// Read the tablets, make sure the master elect is known to the shard
// (it's this tablet, so it better be!).
// Note we will keep going with a partial tablet map, which usually
// happens when a cell is not reachable. After these checks, the
// guarantees we'll have are:
// - global cell is reachable (we just locked and read the shard)
// - the local cell that contains the new master is reachable
// (as we're going to check the new master is in the list)
// That should be enough.
tabletMap, err := topo.GetTabletMapForShard(ctx, agent.TopoServer, tablet.Keyspace, tablet.Shard)
switch err {
case nil:
// keep going
case topo.ErrPartialResult:
log.Warningf("Got topo.ErrPartialResult from GetTabletMapForShard, may need to re-init some tablets")
default:
return false, err
}
masterElectTablet, ok := tabletMap[tablet.Alias]
if !ok {
return false, fmt.Errorf("this master-elect tablet %v not found in replication graph %v/%v %v", tablet.Alias, tablet.Keyspace, tablet.Shard, topotools.MapKeys(tabletMap))
}
// Create reusable Reparent event with available info
ev := &events.Reparent{
ShardInfo: *shardInfo,
NewMaster: *tablet.Tablet,
ExternalID: externalID,
}
if oldMasterTablet, ok := tabletMap[shardInfo.MasterAlias]; ok {
ev.OldMaster = *oldMasterTablet.Tablet
}
defer func() {
if err != nil {
event.DispatchUpdate(ev, "failed: "+err.Error())
}
}()
// sort the tablets, and handle them
slaveTabletMap, masterTabletMap := topotools.SortedTabletMap(tabletMap)
event.DispatchUpdate(ev, "starting external from tablet")
// We fix the new master in the replication graph.
// Note after this call, we may have changed the tablet record,
// so we will always return true, so the tablet record is re-read
// by the agent.
event.DispatchUpdate(ev, "mark ourself as new master")
err = agent.updateReplicationGraphForPromotedSlave(ctx, tablet)
if err != nil {
// This suggests we can't talk to topo server. This is bad.
return true, fmt.Errorf("updateReplicationGraphForPromotedSlave failed: %v", err)
}
// Once this tablet is promoted, remove it from our maps
delete(slaveTabletMap, tablet.Alias)
delete(masterTabletMap, tablet.Alias)
// Then fix all the slaves, including the old master. This
// last step is very likely to time out for some tablets (one
// random guy is dead, the old master is dead, ...). We
// execute them all in parallel until we get to
// wr.ActionTimeout(). After this, no other action with a
// timeout is executed, so even if we got to the timeout,
// we're still good.
event.DispatchUpdate(ev, "restarting slaves")
logger := logutil.NewConsoleLogger()
tmc := tmclient.NewTabletManagerClient()
topotools.RestartSlavesExternal(agent.TopoServer, logger, slaveTabletMap, masterTabletMap, masterElectTablet.Alias, func(ti *topo.TabletInfo, swrd *actionnode.SlaveWasRestartedArgs) error {
return tmc.SlaveWasRestarted(ctx, ti, swrd)
})
// Compute the list of Cells we need to rebuild: old master and
// all other cells if reparenting to another cell.
cells := []string{shardInfo.MasterAlias.Cell}
if shardInfo.MasterAlias.Cell != tablet.Alias.Cell {
cells = nil
}
// now update the master record in the shard object
event.DispatchUpdate(ev, "updating shard record")
log.Infof("Updating Shard's MasterAlias record")
shardInfo.MasterAlias = tablet.Alias
if err = topo.UpdateShard(ctx, agent.TopoServer, shardInfo); err != nil {
return true, err
}
// and rebuild the shard serving graph
event.DispatchUpdate(ev, "rebuilding shard serving graph")
log.Infof("Rebuilding shard serving graph data")
if _, err = topotools.RebuildShard(ctx, logger, agent.TopoServer, tablet.Keyspace, tablet.Shard, cells, agent.LockTimeout, interrupted); err != nil {
return true, err
}
event.DispatchUpdate(ev, "finished")
return true, nil
}
