//This function determines whether streaming is behind replication as it should be.
func (blp *Blp) slavePosBehindReplication() bool {
repl, err := blp.logMetadata.GetCurrentReplicationPosition()
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(fmt.Sprintf("Error in obtaining current replication position %v", err)))
}
currentCoord := blp.currentPosition.Position
if repl.MasterFilename == currentCoord.MasterFilename {
if currentCoord.MasterPosition <= repl.MasterPosition {
return true
}
} else {
replExt, err := strconv.ParseUint(strings.Split(repl.MasterFilename, ".")[1], 10, 64)
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(fmt.Sprintf("Error in obtaining current replication position %v", err)))
}
parseExt, err := strconv.ParseUint(strings.Split(currentCoord.MasterFilename, ".")[1], 10, 64)
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(fmt.Sprintf("Error in obtaining current replication position %v", err)))
}
if replExt >= parseExt {
return true
}
}
return false
}
// A non-nil return signals that event processing should stop.
func (agent *ActionAgent) dispatchAction(actionPath, data string) error {
relog.Info("action dispatch %v", actionPath)
actionNode, err := ActionNodeFromJson(data, actionPath)
if err != nil {
relog.Error("action decode failed: %v %v", actionPath, err)
return nil
}
logfile := flag.Lookup("logfile").Value.String()
if !strings.HasPrefix(logfile, "/dev") {
logfile = path.Join(path.Dir(logfile), "vtaction.log")
}
cmd := []string{
agent.vtActionBinFile,
"-action", actionNode.Action,
"-action-node", actionPath,
"-action-guid", actionNode.ActionGuid,
"-mycnf-file", agent.MycnfFile,
"-logfile", logfile,
}
cmd = append(cmd, agent.ts.GetSubprocessFlags()...)
if agent.DbConfigsFile != "" {
cmd = append(cmd, "-db-configs-file", agent.DbConfigsFile)
}
if agent.DbCredentialsFile != "" {
cmd = append(cmd, "-db-credentials-file", agent.DbCredentialsFile)
}
relog.Info("action launch %v", cmd)
vtActionCmd := exec.Command(cmd[0], cmd[1:]...)
stdOut, vtActionErr := vtActionCmd.CombinedOutput()
if vtActionErr != nil {
relog.Error("agent action failed: %v %v\n%s", actionPath, vtActionErr, stdOut)
// If the action failed, preserve single execution path semantics.
return vtActionErr
}
relog.Info("agent action completed %v %s", actionPath, stdOut)
// Save the old tablet so callbacks can have a better idea of the precise
// nature of the transition.
oldTablet := agent.Tablet().Tablet
// Actions should have side effects on the tablet, so reload the data.
if err := agent.readTablet(); err != nil {
relog.Warning("failed rereading tablet after action - services may be inconsistent: %v %v", actionPath, err)
} else {
agent.runChangeCallbacks(oldTablet, actionPath)
}
// Maybe invalidate the schema.
// This adds a dependency between tabletmanager and tabletserver,
// so it's not ideal. But I (alainjobart) think it's better
// to have up to date schema in vtocc.
if actionNode.Action == TABLET_ACTION_APPLY_SCHEMA {
tabletserver.ReloadSchema()
}
return nil
}
//This function determines whether streaming is behind replication as it should be.
func (blp *Blp) slavePosBehindReplication() bool {
repl, err := blp.globalState.getReplicationPosition()
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(REPLICATION_ERROR, fmt.Sprintf("Error in obtaining current replication position %v", err)))
}
if repl.MasterFilename == blp.currentPosition.Position.MasterFilename {
if blp.currentPosition.Position.MasterPosition <= repl.MasterPosition {
return true
}
} else {
replExt, err := strconv.ParseUint(strings.Split(repl.MasterFilename, ".")[1], 10, 64)
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(CODE_ERROR, fmt.Sprintf("Error in extracting replication position %v", err)))
}
parseExt, err := strconv.ParseUint(strings.Split(blp.currentPosition.Position.MasterFilename, ".")[1], 10, 64)
if err != nil {
relog.Error(err.Error())
panic(NewBinlogParseError(CODE_ERROR, fmt.Sprintf("Error in extracting replication position %v", err)))
}
if replExt >= parseExt {
return true
}
}
return false
}
func (zkd *Zkd) init(preserveData bool) error {
relog.Info("zkd.Init")
for _, path := range zkd.config.DirectoryList() {
if err := os.MkdirAll(path, 0775); err != nil {
relog.Error(err.Error())
return err
}
// FIXME(msolomon) validate permissions?
}
configData, err := zkd.makeCfg()
if err == nil {
err = ioutil.WriteFile(zkd.config.ConfigFile(), []byte(configData), 0664)
}
if err != nil {
relog.Error("failed creating %v: %v", zkd.config.ConfigFile(), err)
return err
}
err = zkd.config.WriteMyid()
if err != nil {
relog.Error("failed creating %v: %v", zkd.config.MyidFile(), err)
return err
}
if err = zkd.Start(); err != nil {
relog.Error("failed starting, check %v", zkd.config.LogDir())
return err
}
zkAddr := fmt.Sprintf("localhost:%v", zkd.config.ClientPort)
zk, session, err := zookeeper.Dial(zkAddr, StartWaitTime*time.Second)
if err != nil {
return err
}
event := <-session
if event.State != zookeeper.STATE_CONNECTED {
return err
}
defer zk.Close()
if !preserveData {
_, err = zk.Create("/zk", "", 0, zookeeper.WorldACL(zookeeper.PERM_ALL))
if err != nil && !zookeeper.IsError(err, zookeeper.ZNODEEXISTS) {
return err
}
if zkd.config.Global {
_, err = zk.Create("/zk/global", "", 0, zookeeper.WorldACL(zookeeper.PERM_ALL))
if err != nil && !zookeeper.IsError(err, zookeeper.ZNODEEXISTS) {
return err
}
}
}
return nil
}
func (blp *Blp) streamBinlog(sendReply mysqlctl.SendUpdateStreamResponse) {
var readErr error
defer func() {
reqIdentifier := fmt.Sprintf("%v, line: '%v'", blp.currentPosition.Position.String(), blp.currentLine)
if x := recover(); x != nil {
serr, ok := x.(*BinlogParseError)
if !ok {
relog.Error("[%v:%v] Uncaught panic for stream @ %v, err: %v ", blp.keyspaceRange.Start.Hex(), blp.keyspaceRange.End.Hex(), reqIdentifier, x)
panic(x)
}
err := *serr
if readErr != nil {
relog.Error("[%v:%v] StreamBinlog error @ %v, error: %v, readErr %v", blp.keyspaceRange.Start.Hex(), blp.keyspaceRange.End.Hex(), reqIdentifier, err, readErr)
err = BinlogParseError{Msg: fmt.Sprintf("%v, readErr: %v", err, readErr)}
} else {
relog.Error("[%v:%v] StreamBinlog error @ %v, error: %v", blp.keyspaceRange.Start.Hex(), blp.keyspaceRange.End.Hex(), reqIdentifier, err)
}
sendError(sendReply, reqIdentifier, err, blp.currentPosition)
}
}()
blr := mysqlctl.NewBinlogReader(blp.binlogPrefix)
var binlogReader io.Reader
var blrReader, blrWriter *os.File
var err, pipeErr error
blrReader, blrWriter, pipeErr = os.Pipe()
if pipeErr != nil {
panic(NewBinlogParseError(pipeErr.Error()))
}
defer blrWriter.Close()
defer blrReader.Close()
readErrChan := make(chan error, 1)
//This reads the binlogs - read end of data pipeline.
go blp.getBinlogStream(blrWriter, blr, readErrChan)
//Decode end of the data pipeline.
binlogDecoder := new(mysqlctl.BinlogDecoder)
binlogReader, err = binlogDecoder.DecodeMysqlBinlog(blrReader)
if err != nil {
panic(NewBinlogParseError(err.Error()))
}
//This function monitors the exit of read data pipeline.
go func(readErr *error, readErrChan chan error, binlogDecoder *mysqlctl.BinlogDecoder) {
*readErr = <-readErrChan
//relog.Info("Read data-pipeline returned readErr: '%v'", *readErr)
if *readErr != nil {
binlogDecoder.Kill()
}
}(&readErr, readErrChan, binlogDecoder)
blp.parseBinlogEvents(sendReply, binlogReader)
}
func (pd *pdns) Serve(r io.Reader, w io.Writer) {
relog.Info("starting zkns resolver")
bufr := bufio.NewReader(r)
needHandshake := true
for {
line, isPrefix, err := bufr.ReadLine()
if err == nil && isPrefix {
err = errLongLine
}
if err == io.EOF {
return
}
if err != nil {
relog.Error("failed reading request: %v", err)
continue
}
if needHandshake {
if !bytes.Equal(line, GREETING_ABI_V2) {
relog.Error("handshake failed: %v != %v", line, GREETING_ABI_V2)
write(w, FAIL_REPLY)
} else {
needHandshake = false
write(w, GREETING_REPLY)
}
continue
}
requestCount.Add(1)
req, err := parseReq(line)
if err != nil {
errorCount.Add(1)
relog.Error("failed parsing request: %v", err)
write(w, FAIL_REPLY)
continue
}
switch req.kind {
case KIND_Q:
respLines, err := pd.handleQReq(req)
if err != nil {
errorCount.Add(1)
relog.Error("failed query: %v %v", req.qname, err)
write(w, FAIL_REPLY)
continue
}
for _, line := range respLines {
write(w, line)
}
case KIND_AXFR:
// FIXME(mike) unimplemented
}
write(w, END_REPLY)
}
}
// Operate on restore tablet.
// Check that the SnapshotManifest is valid and the master has not changed.
// Put Mysql in read-only mode.
// Load the snapshot from source tablet.
// FIXME(alainjobart) which state should the tablet be in? it is a slave,
// but with a much smaller keyspace. For now, do the same as snapshot,
// but this is very dangerous, it cannot be used as a real slave
// or promoted to master in the same shard!
// Put tablet into the replication graph as a spare.
func (ta *TabletActor) partialRestore(actionNode *ActionNode) error {
args := actionNode.args.(*RestoreArgs)
BackfillAlias(args.ZkSrcTabletPath, &args.SrcTabletAlias)
BackfillAlias(args.ZkParentPath, &args.ParentAlias)
// read our current tablet, verify its state
tablet, err := ta.ts.GetTablet(ta.tabletAlias)
if err != nil {
return err
}
if tablet.Type != topo.TYPE_IDLE {
return fmt.Errorf("expected idle type, not %v: %v", tablet.Type, ta.tabletAlias)
}
// read the source tablet
sourceTablet, err := ta.ts.GetTablet(args.SrcTabletAlias)
if err != nil {
return err
}
// read the parent tablet, verify its state
parentTablet, err := ta.ts.GetTablet(args.ParentAlias)
if err != nil {
return err
}
if parentTablet.Type != topo.TYPE_MASTER {
return fmt.Errorf("restore expected master parent: %v %v", parentTablet.Type, args.ParentAlias)
}
// read & unpack the manifest
ssm := new(mysqlctl.SplitSnapshotManifest)
if err := fetchAndParseJsonFile(sourceTablet.Addr, args.SrcFilePath, ssm); err != nil {
return err
}
// change our type to RESTORE and set all the other arguments.
if err := ta.changeTypeToRestore(tablet, sourceTablet, parentTablet.Alias(), ssm.KeyRange); err != nil {
return err
}
// do the work
if err := ta.mysqld.RestoreFromPartialSnapshot(ssm, args.FetchConcurrency, args.FetchRetryCount); err != nil {
relog.Error("RestoreFromPartialSnapshot failed: %v", err)
if err := Scrap(ta.ts, ta.tabletAlias, false); err != nil {
relog.Error("Failed to Scrap after failed RestoreFromPartialSnapshot: %v", err)
}
return err
}
// change to TYPE_MASTER, we're done!
return ChangeType(ta.ts, ta.tabletAlias, topo.TYPE_MASTER, true)
}
func handleSnapshot(rw http.ResponseWriter, req *http.Request, snapshotDir string, allowedPaths []string) {
// if we get any error, we'll try to write a server error
// (it will fail if the header has already been written, but at least
// we won't crash vttablet)
defer func() {
if x := recover(); x != nil {
relog.Error("vttablet http server panic: %v", x)
http.Error(rw, fmt.Sprintf("500 internal server error: %v", x), http.StatusInternalServerError)
}
}()
// /snapshot must be rewritten to the actual location of the snapshot.
relative, err := filepath.Rel(mysqlctl.SnapshotURLPath, req.URL.Path)
if err != nil {
relog.Error("bad snapshot relative path %v %v", req.URL.Path, err)
http.Error(rw, "400 bad request", http.StatusBadRequest)
return
}
// Make sure that realPath is absolute and resolve any escaping from
// snapshotDir through a symlink.
realPath, err := filepath.Abs(path.Join(snapshotDir, relative))
if err != nil {
relog.Error("bad snapshot absolute path %v %v", req.URL.Path, err)
http.Error(rw, "400 bad request", http.StatusBadRequest)
return
}
realPath, err = filepath.EvalSymlinks(realPath)
if err != nil {
relog.Error("bad snapshot symlink eval %v %v", req.URL.Path, err)
http.Error(rw, "400 bad request", http.StatusBadRequest)
return
}
// Resolve all the possible roots and make sure we're serving
// from one of them
for _, allowedPath := range allowedPaths {
// eval the symlinks of the allowed path
allowedPath, err := filepath.EvalSymlinks(allowedPath)
if err != nil {
continue
}
if strings.HasPrefix(realPath, allowedPath) {
sendFile(rw, req, realPath)
return
}
}
relog.Error("bad snapshot real path %v %v", req.URL.Path, realPath)
http.Error(rw, "400 bad request", http.StatusBadRequest)
}
func handleError(err *error, blp *BinlogPlayer) {
lastTxnPosition := blp.recoveryState.Position
if x := recover(); x != nil {
serr, ok := x.(error)
if ok {
*err = serr
relog.Error("Last Txn Position '%v', error %v", lastTxnPosition, serr)
return
}
relog.Error("uncaught panic %v", x)
panic(x)
}
}
func (wr *Wrangler) SnapshotSourceEnd(tabletAlias topo.TabletAlias, slaveStartRequired, readWrite bool, originalType topo.TabletType) (err error) {
var ti *topo.TabletInfo
ti, err = wr.ts.GetTablet(tabletAlias)
if err != nil {
return
}
var actionPath string
actionPath, err = wr.ai.SnapshotSourceEnd(tabletAlias, &tm.SnapshotSourceEndArgs{slaveStartRequired, !readWrite})
if err != nil {
return
}
// wait for completion, and save the error
err = wr.ai.WaitForCompletion(actionPath, wr.actionTimeout())
if err != nil {
relog.Error("SnapshotSourceEnd failed (%v), leaving tablet type alone", err)
return
}
if ti.Tablet.Parent.Uid == topo.NO_TABLET {
ti.Tablet.Type = topo.TYPE_MASTER
err = topo.UpdateTablet(wr.ts, ti)
} else {
err = wr.ChangeType(ti.Alias(), originalType, false)
}
return err
}
func (hh *httpHandler) ServeHTTP(c http.ResponseWriter, req *http.Request) {
conn := &httpConnectionBroker{c, req.Body}
codec := hh.cFactory(conn)
if err := rpc.ServeRequestWithContext(codec, &proto.Context{RemoteAddr: req.RemoteAddr}); err != nil {
relog.Error("rpcwrap: %v", err)
}
}
// ConcurrentMap applies fun in a concurrent manner on integers from 0
// to n-1 (they are assumed to be indexes of some slice containing
// items to be processed). The first error returned by a fun
// application will returned (subsequent errors will only be
// logged). It will use concurrency goroutines.
func ConcurrentMap(concurrency, n int, fun MapFunc) error {
errors := make(chan error)
work := make(chan int, n)
for i := 0; i < n; i++ {
work <- i
}
close(work)
for j := 0; j < concurrency; j++ {
go func() {
for i := range work {
errors <- fun(i)
}
}()
}
var err error
for i := 0; i < n; i++ {
if e := <-errors; e != nil {
if err != nil {
relog.Error("multiple errors, this one happened but it won't be returned: %v", err)
}
err = e
}
}
return err
}
// 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.
func (wr *Wrangler) PartialSnapshot(tabletAlias topo.TabletAlias, keyName string, startKey, endKey key.HexKeyspaceId, forceMasterSnapshot bool, concurrency int) (manifest string, parent topo.TabletAlias, err error) {
restoreAfterSnapshot, err := wr.prepareToSnapshot(tabletAlias, forceMasterSnapshot)
if err != nil {
return
}
defer func() {
err = replaceError(err, restoreAfterSnapshot())
}()
actionPath, err := wr.ai.PartialSnapshot(tabletAlias, &tm.PartialSnapshotArgs{keyName, startKey, endKey, concurrency})
if err != nil {
return
}
results, actionErr := wr.ai.WaitForCompletionReply(actionPath, wr.actionTimeout())
var reply *tm.SnapshotReply
if actionErr != nil {
relog.Error("PartialSnapshot failed, still restoring tablet type: %v", actionErr)
reply = &tm.SnapshotReply{}
} else {
reply = results.(*tm.SnapshotReply)
tm.BackfillAlias(reply.ZkParentPath, &reply.ParentAlias)
}
return reply.ManifestPath, reply.ParentAlias, actionErr
}
func main() {
dbConfigsFile, dbCredentialsFile := dbconfigs.RegisterCommonFlags()
flag.Parse()
relog.Info("started vtaction %v", os.Args)
rpc.HandleHTTP()
jsonrpc.ServeHTTP()
jsonrpc.ServeRPC()
bsonrpc.ServeHTTP()
bsonrpc.ServeRPC()
logFile, err := os.OpenFile(*logFilename,
os.O_APPEND|os.O_WRONLY|os.O_CREATE, 0666)
if err != nil {
relog.Fatal("Can't open log file: %v", err)
}
relog.SetOutput(logFile)
relog.SetPrefix(fmt.Sprintf("vtaction [%v] ", os.Getpid()))
if err := relog.SetLevelByName(*logLevel); err != nil {
relog.Fatal("%v", err)
}
relog.HijackLog(nil)
relog.HijackStdio(logFile, logFile)
mycnf, mycnfErr := mysqlctl.ReadMycnf(*mycnfFile)
if mycnfErr != nil {
relog.Fatal("mycnf read failed: %v", mycnfErr)
}
relog.Debug("mycnf: %v", jscfg.ToJson(mycnf))
dbcfgs, cfErr := dbconfigs.Init(mycnf.SocketFile, *dbConfigsFile, *dbCredentialsFile)
if err != nil {
relog.Fatal("%s", cfErr)
}
mysqld := mysqlctl.NewMysqld(mycnf, dbcfgs.Dba, dbcfgs.Repl)
topoServer := topo.GetServer()
defer topo.CloseServers()
actor := tabletmanager.NewTabletActor(mysqld, topoServer)
// we delegate out startup to the micromanagement server so these actions
// will occur after we have obtained our socket.
bindAddr := fmt.Sprintf(":%v", *port)
httpServer := &http.Server{Addr: bindAddr}
go func() {
if err := httpServer.ListenAndServe(); err != nil {
relog.Error("httpServer.ListenAndServe err: %v", err)
}
}()
actionErr := actor.HandleAction(*actionNode, *action, *actionGuid, *force)
if actionErr != nil {
relog.Fatal("action error: %v", actionErr)
}
relog.Info("finished vtaction %v", os.Args)
}
func (rowCache *InvalidationProcessor) processEvent(event *mysqlctl.UpdateResponse) error {
position := ""
if event.BinlogPosition.Valid() {
position = event.BinlogPosition.String()
}
if event.Error != "" {
relog.Error("Update stream returned error '%v'", event.Error)
// Check if update stream error is fatal, else record it and move on.
if strings.HasPrefix(event.Error, mysqlctl.FATAL) {
relog.Info("Returning Service Error")
return NewInvalidationError(FATAL_ERROR, event.Error, position)
}
rowCache.updateErrCounters(NewInvalidationError(INVALID_EVENT, event.Error, position))
return nil
}
if !event.BinlogPosition.Valid() {
rowCache.updateErrCounters(NewInvalidationError(INVALID_EVENT, "no error, position is not set", ""))
return nil
}
var err error
switch event.EventData.SqlType {
case mysqlctl.DDL:
err = rowCache.handleDdlEvent(event)
if err != nil {
return err
}
case mysqlctl.BEGIN:
rowCache.dmlBuffer = rowCache.dmlBuffer[:0]
if rowCache.inTxn {
rowCache.updateErrCounters(NewInvalidationError(INVALID_EVENT, "Invalid 'BEGIN' event, transaction already in progress", position))
return nil
}
rowCache.inTxn = true
case mysqlctl.COMMIT:
if !rowCache.inTxn {
rowCache.updateErrCounters(NewInvalidationError(INVALID_EVENT, "Invalid 'COMMIT' event for a non-transaction", position))
return nil
}
err = rowCache.handleTxn(event)
if err != nil {
return err
}
rowCache.inTxn = false
rowCache.dmlBuffer = rowCache.dmlBuffer[:0]
case "insert", "update", "delete":
dml, err := rowCache.buildDmlData(event)
if err != nil {
return err
}
if dml != nil {
rowCache.dmlBuffer = append(rowCache.dmlBuffer, dml)
}
default:
rowCache.updateErrCounters(NewInvalidationError(INVALID_EVENT, fmt.Sprintf("Unknown SqlType, %v %v", event.EventData.SqlType, event.EventData.Sql), position))
//return NewInvalidationError(INVALID_EVENT, fmt.Sprintf("Unknown SqlType, %v %v", event.EventData.SqlType, event.EventData.Sql))
}
return nil
}
//Main entry function for reading and parsing the binlog.
func (blp *Blp) StreamBinlog(sendReply SendUpdateStreamResponse, binlogPrefix string) {
var err error
for {
err = blp.streamBinlog(sendReply, binlogPrefix)
sErr, ok := err.(*BinlogParseError)
if ok && sErr.IsEOF() {
// Double check the current parse position
// with replication position, if not so, it is an error
// otherwise it is a true EOF so retry.
relog.Info("EOF, retrying")
ok, replErr := blp.isBehindReplication()
if replErr != nil {
err = replErr
} else if !ok {
err = NewBinlogParseError(REPLICATION_ERROR, "EOF, but parse position behind replication")
} else {
time.Sleep(5.0 * time.Second)
blp.startPosition = &ReplicationCoordinates{MasterFilename: blp.currentPosition.Position.MasterFilename,
MasterPosition: blp.currentPosition.Position.MasterPosition,
RelayFilename: blp.currentPosition.Position.RelayFilename,
RelayPosition: blp.currentPosition.Position.RelayPosition}
continue
}
}
relog.Error("StreamBinlog error @ %v, error: %v", blp.currentPosition.String(), err.Error())
SendError(sendReply, err, blp.currentPosition)
break
}
}
func main() {
flag.Parse()
servenv.Init("vt_binlog_player")
if *startPosFile == "" {
relog.Fatal("start-pos-file was not supplied.")
}
if *dbConfigFile == "" {
relog.Fatal("Cannot start without db-config-file")
}
blp, err := initBinlogPlayer(*startPosFile, *dbConfigFile, *lookupConfigFile, *dbCredFile, *useCheckpoint, *debug, *port)
if err != nil {
relog.Fatal("Error in initializing binlog player - '%v'", err)
}
blp.txnBatch = *txnBatch
blp.maxTxnInterval = time.Duration(*maxTxnInterval) * time.Second
blp.execDdl = *execDdl
if *tables != "" {
tables := strings.Split(*tables, ",")
blp.tables = make([]string, len(tables))
for i, table := range tables {
blp.tables[i] = strings.TrimSpace(table)
}
relog.Info("len tables %v tables %v", len(blp.tables), blp.tables)
}
relog.Info("BinlogPlayer client for keyrange '%v:%v' starting @ '%v'",
blp.startPosition.KeyrangeStart,
blp.startPosition.KeyrangeEnd,
blp.startPosition.Position)
if *port != 0 {
umgmt.AddStartupCallback(func() {
umgmt.StartHttpServer(fmt.Sprintf(":%v", *port))
})
}
umgmt.AddStartupCallback(func() {
c := make(chan os.Signal, 1)
signal.Notify(c, syscall.SIGTERM)
go func() {
for sig := range c {
umgmt.SigTermHandler(sig)
}
}()
})
umgmt.AddCloseCallback(func() {
close(interrupted)
})
//Make a request to the server and start processing the events.
stdout = bufio.NewWriterSize(os.Stdout, 16*1024)
err = blp.applyBinlogEvents()
if err != nil {
relog.Error("Error in applying binlog events, err %v", err)
}
relog.Info("vt_binlog_player done")
}
// zkocc: a proxy for zk
func main() {
flag.Parse()
if err := servenv.Init("zkocc"); err != nil {
relog.Fatal("Error in servenv.Init: %v", err)
}
rpc.HandleHTTP()
jsonrpc.ServeHTTP()
jsonrpc.ServeRPC()
bsonrpc.ServeHTTP()
bsonrpc.ServeRPC()
zk.RegisterZkReader(zkocc.NewZkReader(*resolveLocal, flag.Args()))
// we delegate out startup to the micromanagement server so these actions
// will occur after we have obtained our socket.
umgmt.SetLameDuckPeriod(float32(*lameDuckPeriod))
umgmt.SetRebindDelay(float32(*rebindDelay))
umgmt.AddStartupCallback(func() {
umgmt.StartHttpServer(fmt.Sprintf(":%v", *port))
})
relog.Info("started zkocc %v", *port)
umgmtSocket := fmt.Sprintf("/tmp/zkocc-%08x-umgmt.sock", *port)
if umgmtErr := umgmt.ListenAndServe(umgmtSocket); umgmtErr != nil {
relog.Error("umgmt.ListenAndServe err: %v", umgmtErr)
}
relog.Info("done")
}
func (rowCache *InvalidationProcessor) updateErrCounters(err *InvalidationError) {
relog.Error(err.Error())
if errorStats == nil {
relog.Warning("errorStats is not initialized")
return
}
errorStats.Add(err.errType, 1)
}
func commandStaleActions(wr *wrangler.Wrangler, subFlags *flag.FlagSet, args []string) (string, error) {
maxStaleness := subFlags.Duration("max-staleness", 5*time.Minute, "how long since the last modification before an action considered stale")
purge := subFlags.Bool("purge", false, "purge stale actions")
subFlags.Parse(args)
if subFlags.NArg() == 0 {
relog.Fatal("action StaleActions requires <zk action path>")
}
zkts, ok := wr.TopoServer().(*zktopo.Server)
if !ok {
return "", fmt.Errorf("StaleActions requires a zktopo.Server")
}
zkPaths, err := resolveWildcards(wr, subFlags.Args())
if err != nil {
return "", err
}
var errCount sync2.AtomicInt32
wg := sync.WaitGroup{}
for _, apath := range zkPaths {
wg.Add(1)
go func(zkActionPath string) {
defer wg.Done()
staleActions, err := staleActions(zkts, zkActionPath, *maxStaleness)
if err != nil {
errCount.Add(1)
relog.Error("can't check stale actions: %v %v", zkActionPath, err)
return
}
for _, action := range staleActions {
fmt.Println(fmtAction(action))
}
if *purge && len(staleActions) > 0 {
err := zkts.PurgeActions(zkActionPath, tm.ActionNodeCanBePurged)
if err != nil {
errCount.Add(1)
relog.Error("can't purge stale actions: %v %v", zkActionPath, err)
return
}
}
}(apath)
}
wg.Wait()
if errCount.Get() > 0 {
return "", fmt.Errorf("some errors occurred, check the log")
}
return "", nil
}
func (dc DBClient) Rollback() error {
_, err := dc.dbConn.ExecuteFetch(ROLLBACK, 1, false)
if err != nil {
relog.Error("ROLLBACK failed w/ error %v", err)
dc.dbConn.Close()
}
return err
}
func (dc DBClient) Commit() error {
_, err := dc.dbConn.ExecuteFetch(COMMIT, 1, false)
if err != nil {
relog.Error("COMMIT failed w/ error %v", err)
dc.dbConn.Close()
}
return err
}
func (dc DBClient) Begin() error {
_, err := dc.dbConn.ExecuteFetch(BEGIN, 1, false)
if err != nil {
relog.Error("BEGIN failed w/ error %v", err)
dc.handleError(err)
}
return err
}
// replaceError replaces original with recent if recent is not nil,
// logging original if it wasn't nil. This should be used in deferred
// cleanup functions if they change the returned error.
func replaceError(original, recent error) error {
if recent == nil {
return original
}
if original != nil {
relog.Error("One of multiple error: %v", original)
}
return recent
}
func (client *Client) CloseListeners() error {
request := new(Request)
reply := new(Reply)
err := client.Call("UmgmtService.CloseListeners", request, reply)
if err != nil {
relog.Error("CloseListeners err: %v", err)
}
return err
}
func (client *Client) GracefulShutdown() error {
request := new(Request)
reply := new(Reply)
err := client.Call("UmgmtService.GracefulShutdown", request, reply)
if err != nil && err != io.ErrUnexpectedEOF {
relog.Error("GracefulShutdown err: %v", err)
}
return err
}
// return a Reader from which the decoded binlog can be read
func (decoder *BinlogDecoder) DecodeMysqlBinlog(binlog *os.File) (io.Reader, error) {
dir, err := findVtMysqlbinlogDir()
if err != nil {
return nil, err
}
dir = path.Join(dir, "bin")
name := "vt_mysqlbinlog"
arg := []string{"vt_mysqlbinlog", "-"}
dataRdFile, dataWrFile, pipeErr := os.Pipe()
if pipeErr != nil {
relog.Error("DecodeMysqlBinlog: error in creating pipe %v", pipeErr)
return nil, pipeErr
}
// let the caller close the read file
defer dataWrFile.Close()
fds := []*os.File{
binlog,
dataWrFile,
os.Stderr,
}
attrs := &os.ProcAttr{Dir: dir, Files: fds}
process, err := os.StartProcess(name, arg, attrs)
if err != nil {
relog.Error("DecodeMysqlBinlog: error in decoding binlog %v", err)
return nil, err
}
decoder.process = process
go func() {
// just make sure we don't spawn zombies
waitMsg, err := decoder.process.Wait()
if err != nil {
relog.Error("vt_mysqlbinlog exited: %v err: %v", waitMsg, err)
} else {
relog.Info("vt_mysqlbinlog exited: %v err: %v", waitMsg, err)
}
}()
return dataRdFile, nil
}
func (dc DBClient) ExecuteFetch(query string, maxrows int, wantfields bool) (*proto.QueryResult, error) {
mqr, err := dc.dbConn.ExecuteFetch(query, maxrows, wantfields)
if err != nil {
relog.Error("ExecuteFetch failed w/ error %v", err)
dc.handleError(err)
return nil, err
}
qr := proto.QueryResult(*mqr)
return &qr, nil
}
func (wr *Wrangler) UnreserveForRestoreMulti(dstTabletAliases []topo.TabletAlias) {
for _, dstTabletAlias := range dstTabletAliases {
ufrErr := wr.UnreserveForRestore(dstTabletAlias)
if ufrErr != nil {
relog.Error("Failed to UnreserveForRestore destination tablet after failed source snapshot: %v", ufrErr)
} else {
relog.Info("Un-reserved %v", dstTabletAlias)
}
}
}
func (client *Client) Ping() (string, error) {
request := new(Request)
reply := new(Reply)
err := client.Call("UmgmtService.Ping", request, reply)
if err != nil {
relog.Error("rpc err: %v", err)
return "ERROR", err
}
return reply.Message, nil
}