// execPtOsc shells out and uses pt-osc to actually run a migration.
func (runner *runner) execPtOsc(currentMigration *migration.Migration,
ptOscOptionGenerator commandOptionGenerator, copyPercentChan chan int, unstageDone bool) (bool, error) {
// unstageDone = whether or not unstagedMigrationsWaitGroup has already had .Done() called on it
defer func() {
if !unstageDone {
unstagedMigrationsWaitGroup.Done()
}
}()
canceled := false
var fileRoutineWaitGroup sync.WaitGroup
err := runner.createAndRetrieveFiles(currentMigration)
if err != nil {
return canceled, err
}
ptOscLogFile, ptOscLogWriter, err := setupLogWriter(currentMigration.LogFile)
if err != nil {
glog.Errorf("mig_id=%d: error creating pt-osc log file '%s' (error: %s)", currentMigration.Id, currentMigration.LogFile, err)
return canceled, ErrPtOscExec
}
defer ptOscLogFile.Close()
// generate the pt-osc command to run
commandOptions := ptOscOptionGenerator(currentMigration)
glog.Infof("mig_id=%d: Running %s %v", currentMigration.Id, runner.PtOscPath, strings.Join(commandOptions, " "))
cmd := exec.Command(runner.PtOscPath, commandOptions...)
// capture stdout and stderr of the command
stdout, err := cmd.StdoutPipe()
if err != nil {
glog.Errorf("mig_id=%d: error getting stdout pipe for pt-osc exec (error: %s)", currentMigration.Id, err)
return canceled, ErrPtOscExec
}
stderr, err := cmd.StderrPipe()
if err != nil {
glog.Errorf("mig_id=%d: error getting stderr pipe for pt-osc exec (error: %s)", currentMigration.Id, err)
return canceled, ErrPtOscExec
}
// start the pt-osc comand
if err := cmd.Start(); err != nil {
glog.Errorf("mig_id=%d: error starting pt-osc exec (error: %s)", currentMigration.Id, err)
return canceled, ErrPtOscExec
}
// setup a channel and goroutine for logging output of stdout/stderr
ptOscLogChan := make(chan string)
fileRoutineWaitGroup.Add(1)
fileSyncWaitGroup.Add(1)
go runner.writeToPtOscLog(ptOscLogWriter, ptOscLogChan, writeLineToPtOscLog, currentMigration.Id, &fileRoutineWaitGroup)
// setup a goroutine for sending statefiles to the api
ptOscStateFileChan := make(chan string)
fileRoutineWaitGroup.Add(1)
fileSyncWaitGroup.Add(1)
go runner.syncStateFile(currentMigration, ptOscStateFileChan, &fileRoutineWaitGroup)
// setup goroutines for watching stdout/stderr of the command
stdoutErrChan := make(chan error)
stderrErrChan := make(chan error)
go currentMigration.WatchMigrationStdout(stdout, stdoutErrChan, ptOscLogChan)
if currentMigration.Status == migration.RunMigrationStatus {
// setup a goroutine to continually update the % copied of the migration
fileRoutineWaitGroup.Add(1)
fileSyncWaitGroup.Add(1)
go runner.updateMigrationCopyPercentage(currentMigration, copyPercentChan, &fileRoutineWaitGroup)
go currentMigration.WatchMigrationCopyStderr(stderr, copyPercentChan, stderrErrChan, ptOscLogChan)
} else {
go currentMigration.WatchMigrationStderr(stderr, stderrErrChan, ptOscLogChan)
}
// save the pid of the pt-osc process
currentMigration.Pid = cmd.Process.Pid
glog.Infof("mig_id=%d: pt-osc pid for status %d is %d.", currentMigration.Id, currentMigration.Status, currentMigration.Pid)
// add the migration id and pid to the running migration map
runningMigMutex.Lock()
runningMigrations[currentMigration.Id] = currentMigration.Pid
runningMigMutex.Unlock()
if !unstageDone {
unstageDone = true
unstagedMigrationsWaitGroup.Done()
}
// wait for both stdout and stderr error channels to receive a signal
stdoutErr := <-stdoutErrChan
stderrErr := <-stderrErrChan
close(ptOscLogChan)
close(ptOscStateFileChan)
// remove the migration id from the running migration map
runningMigMutex.Lock()
delete(runningMigrations, currentMigration.Id)
runningMigMutex.Unlock()
// get the exit status of the command. if it was sent a SIGKILL (most likely
// by another goroutine) we want to know because we will treat it differently
failed := false
if err := cmd.Wait(); err != nil {
if exiterr, ok := err.(*exec.ExitError); ok {
if status, ok := exiterr.Sys().(syscall.WaitStatus); ok {
exitSignal := status.Signal().String()
glog.Infof("mig_id=%d: exit signal was %s.", currentMigration.Id, exitSignal)
if (exitSignal == SIGKILLSignal) && (currentMigration.Status == migration.RunMigrationStatus) {
// was killed
glog.Infof("mig_id=%d: migration must have been canceled", currentMigration.Id)
canceled = true
} else if currentMigration.Status == migration.RunMigrationStatus {
// died for an unexpected reason
glog.Infof("mig_id=%d: migration died for an unexpected reason", currentMigration.Id)
failed = true
}
}
}
} else {
if (stderrErr == nil) && (currentMigration.Status == migration.RunMigrationStatus) {
// wasn't killed. copy completed 100%
glog.Infof("mig_id=%d: updating migration with copy percentage of 100", currentMigration.Id)
copyPercentChan <- 100
}
}
if copyPercentChan != nil {
glog.Infof("Closing copy percent channel")
close(copyPercentChan)
}
fileRoutineWaitGroup.Wait() // wait for go routines to finish
// favor returning error from unexpected failure, then error from stderr,
// and lastly error from stdout
if failed {
return canceled, ErrUnexpectedExit
} else if stderrErr != nil {
return canceled, stderrErr
} else if stdoutErr != nil {
return canceled, stdoutErr
}
return canceled, nil
}
