// This function streams the reader to the Gearman job (through job.SendData())
func streamToGearman(reader io.Reader, job baseworker.Job) error {
buffer := make([]byte, 1024)
for {
n, err := reader.Read(buffer)
// Process the data before processing the error (as per the io.Reader docs)
if n > 0 {
job.SendData(buffer[:n])
}
if err == io.EOF {
return nil
} else if err != nil {
return err
}
}
}
// sendStderrWarnings sends the last X lines in the stderr output and to the job's warnings
// field
func sendStderrWarnings(buffer io.Reader, job baseworker.Job, warningLines int) error {
scanner := bufio.NewScanner(buffer)
// Create a circular buffer for the last X lines
lastStderrLines := ring.New(warningLines)
for scanner.Scan() {
lastStderrLines = lastStderrLines.Next()
lastStderrLines.Value = scanner.Bytes()
}
// Walk forward through the buffer to get all the last X entries. Note that we call next first
// so that we start at the oldest entry.
for i := 0; i < lastStderrLines.Len(); i++ {
if lastStderrLines = lastStderrLines.Next(); lastStderrLines.Value != nil {
job.SendWarning(append(lastStderrLines.Value.([]byte), byte('\n')))
}
}
return scanner.Err()
}
// Process runs the Gearman job by running the configured task.
// We need to implement the Task interface so we return (byte[], error)
// though the byte[] is always nil.
func (conf TaskConfig) Process(job baseworker.Job) (b []byte, returnErr error) {
jobID := getJobID(job)
if jobID == "" {
jobID = strconv.Itoa(rand.Int())
lg.InfoD("rand-job-id", logger.M{"msg": "no job id parsed, random assigned."})
}
jobData := string(job.Data())
data := logger.M{
"function": conf.FunctionName,
"job_id": jobID,
"job_data": jobData,
}
// This wraps the actual processing to do some logging
lg.InfoD("START", data)
start := time.Now()
for try := 0; try < conf.RetryCount+1; try++ {
// We create a temporary directory to be used as the work directory of the process.
// A new work directory is created for every retry of the process.
// We try to use MEOS_SANDBOX, the default will be the system temp directory.
tempDirPath, err := ioutil.TempDir(os.Getenv("MESOS_SANDBOX"),
fmt.Sprintf("%s-%s-%d-", conf.FunctionName, jobID, try))
if err != nil {
lg.CriticalD("tempdir-failure", logger.M{"error": err.Error()})
return nil, err
}
defer os.RemoveAll(tempDirPath)
// insert the job id and the work directory path into the environment
extraEnvVars := []string{
fmt.Sprintf("JOB_ID=%s", jobID),
fmt.Sprintf("WORK_DIR=%s", tempDirPath)}
err = conf.doProcess(job, extraEnvVars, try)
end := time.Now()
data["type"] = "gauge"
// Return if the job was successful.
if err == nil {
lg.InfoD("SUCCESS", logger.M{
"type": "counter",
"function": conf.FunctionName})
legacyLg.InfoD("success", logger.M{
"type": "counter",
"function": conf.FunctionName})
data["value"] = 1
data["success"] = true
lg.InfoD("END", data)
// Hopefully none of our jobs last long enough for a uint64...
// Note that we cannot use lg.GaugeIntD because duration is uint64
lg.InfoD("duration", logger.M{
"value": uint64(end.Sub(start).Seconds() * 1000),
"type": "gauge",
"function": conf.FunctionName,
"job_id": jobID,
"job_data": jobData})
return nil, nil
}
data["value"] = 0
data["success"] = false
data["error_message"] = err.Error()
returnErr = err
if try != conf.RetryCount {
lg.ErrorD("RETRY", data)
}
}
lg.InfoD("FAILURE", logger.M{"type": "counter", "function": conf.FunctionName})
legacyLg.InfoD("failure", logger.M{"type": "counter", "function": conf.FunctionName})
lg.ErrorD("END", data)
return nil, returnErr
}
func (conf TaskConfig) doProcess(job baseworker.Job, envVars []string, tryCount int) error {
defer func() {
// If we panicked then set the panic message as a warning. Gearman-go will
// handle marking this job as failed.
if r := recover(); r != nil {
err := r.(error)
job.SendWarning([]byte(err.Error()))
}
}()
// shutdownTicker will effectively control the executution of the ticker.
shutdownTicker := make(chan interface{})
defer func() {
shutdownTicker <- 1
}()
// every minute we will output a heartbeat kayvee log for the job.
tickUnit := time.Minute
ticker := time.NewTicker(tickUnit)
go func() {
defer ticker.Stop()
units := 0
for {
select {
case <-shutdownTicker:
close(shutdownTicker)
return
case <-ticker.C:
units++
lg.GaugeIntD("heartbeat", units, logger.M{
"try_number": tryCount,
"function": job.Fn(),
"job_id": getJobID(job),
"unit": tickUnit.String(),
})
}
}
}()
var args []string
var err error
if conf.ParseArgs {
args, err = argsparser.ParseArgs(string(job.Data()))
if err != nil {
return fmt.Errorf("Failed to parse args: %s", err.Error())
}
} else {
args = []string{string(job.Data())}
}
cmd := exec.Command(conf.FunctionCmd, args...)
// insert provided env vars into the job
cmd.Env = append(os.Environ(), envVars...)
// create new pgid for this process so we can later kill all subprocess launched by it
cmd.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
// Write the stdout and stderr of the process to both this process' stdout and stderr
// and also write it to a byte buffer so that we can return it with the Gearman job
// data as necessary.
var stderrbuf bytes.Buffer
cmd.Stderr = io.MultiWriter(os.Stderr, &stderrbuf)
defer sendStderrWarnings(&stderrbuf, job, conf.WarningLines)
stdoutReader, stdoutWriter := io.Pipe()
cmd.Stdout = io.MultiWriter(os.Stdout, stdoutWriter)
done := make(chan error)
// Track when the job has started so that we don't try and sigterm a nil process
started := make(chan struct{})
go func() {
defer close(done)
finishedProcessingStdout := make(chan error)
go func() {
finishedProcessingStdout <- streamToGearman(stdoutReader, job)
}()
if err := cmd.Start(); err != nil {
done <- err
return
}
close(started)
// Save the cmdErr. We want to process stdout and stderr before we return it
cmdErr := cmd.Wait()
stdoutWriter.Close()
stdoutErr := <-finishedProcessingStdout
if cmdErr != nil {
done <- cmdErr
} else if stdoutErr != nil {
done <- stdoutErr
}
}()
<-started
// No timeout
if conf.CmdTimeout == 0 {
select {
case err := <-done:
// Will be nil if the channel was closed without any errors
return err
case <-conf.Halt:
if err := stopProcess(cmd.Process, conf.CmdTimeout); err != nil {
return fmt.Errorf("error stopping process: %s", err)
}
return fmt.Errorf("killed process due to sigterm")
}
}
select {
case err := <-done:
// Will be nil if the channel was closed without any errors
return err
case <-conf.Halt:
if err := stopProcess(cmd.Process, conf.CmdTimeout); err != nil {
return fmt.Errorf("error stopping process: %s", err)
}
return nil
case <-time.After(conf.CmdTimeout):
if err := stopProcess(cmd.Process, 0); err != nil {
return fmt.Errorf("error timing out process after %s: %s", conf.CmdTimeout.String(), err)
}
return fmt.Errorf("process timed out after %s", conf.CmdTimeout.String())
}
}
// getJobID returns the jobId from the job handle
func getJobID(job baseworker.Job) string {
splits := strings.Split(job.Handle(), ":")
return splits[len(splits)-1]
}
