// Kill forcefully stops a process
func (j *Job) Kill(sig int64) error {
var err error
switch sig {
case -9:
log.Debugln("Sending process Kill (-9) signal")
err = syscall.Kill(-j.Pgid, syscall.SIGKILL)
default:
signal := syscall.Signal(sig)
err = syscall.Kill(-j.Pgid, signal)
}
if err != nil {
log.Warnf("Error received calling kill on sub-process: %s", err)
return err
}
return nil
}
func timerTicker(timer *WallclockTimer) {
for t := range timer.ticker.C {
timer.incrementTimer()
select {
case command := <-timer.command:
log.Debugln("Command tick at: ", t)
log.Debugf("Command received: %s\n", command)
switch command {
case "reset":
log.Debugln("Received a Reset")
timer.tick = false
timer.elapsedTime = time.Duration(0)
case "start":
log.Debugln("Received a start")
timer.tick = true
timer.elapsedTime = time.Duration(0)
timer.previousTime = time.Now()
case "stop":
log.Debugln("Received a stop")
timer.tick = false
case "resume":
log.Debugln("Received a resume")
timer.tick = true
timer.previousTime = time.Now()
default:
log.Errorln("Unknown command received")
}
default:
//log.Debugln("No commands received, just keep ticking")
}
//log.Debugf("Timer elapsed time is: %v", timer.elapsedTime)
//log.Debugf("Timer timeout time is: %v", timer.timeoutTime)
if timer.elapsedTime > timer.timeoutTime {
log.Debugln("Elapsed time has exceeded timeout time.")
timer.done <- errors.New("The timer has timed out.")
timer.tick = false
// Possible bug? Reset timer to equal timeout
timer.elapsedTime = timer.timeoutTime
}
}
}
func monitor(
signals chan os.Signal,
rcChanChan chan chan agents.RemoteControlCommand,
job agents.JobControl,
psChan chan agents.ProcessStatCommand,
timer agents.Timer,
done chan error,
) {
// Catch any panics here to ensure we kill the child process before going
// to our own doom.
defer func() {
if e := recover(); e != nil {
job.Kill(-9)
panic(e)
}
}()
var logSampling <-chan time.Time
var err error
if *stdoutByteLimit > 0 {
ticker := time.NewTicker(100 * time.Millisecond)
// Replace the time channel with an actual ticker if this is in use
logSampling = ticker.C
}
var rcChan chan agents.RemoteControlCommand
for {
select {
case rcChan = <-rcChanChan:
// Catch incoming signals and operate on them as if they were remote commands
case sig := <-signals:
switch sig {
case syscall.SIGINT:
log.Debugln("Caught SIGINT, graceful shutdown")
// Initiate non-blocking send
select {
case psChan <- agents.ProcessStatCommand{}:
log.Debugln("Sending psChan a msg to sample")
default:
log.Debugln("SIGINT failed to send a sample msg on the psChan")
}
err = job.Stop()
case syscall.SIGTERM:
log.Debugln("Caught SIGTERM, end abruptly")
job.Kill(-9)
case syscall.SIGHUP:
log.Debugln("Caught SIGHUP, emit stats")
// Initiate non-blocking send
select {
case psChan <- agents.ProcessStatCommand{}:
log.Debugln("Sending psChan a msg to sample")
default:
log.Debugln("SIGHUP failed to send a sample msg on the psChan")
}
case syscall.SIGQUIT:
log.Debugln("Caught SIGQUIT, graceful shutdown")
select {
case psChan <- agents.ProcessStatCommand{}:
log.Debugln("Sending psChan a msg to sample")
default:
log.Debugln("SIGQUIT failed to send a sample msg on the psChan")
}
err = job.Stop()
}
// Process incoming remote commands, toss unknown requests
case cmd := <-rcChan:
log.Debugf("Got a command %#v\n", cmd)
switch cmd.Command {
case "suspend":
log.Debugln("RemoteCommand: Suspend")
job.Suspend()
case "resume":
log.Debugln("RemoteCommand: Resume")
job.Resume()
case "kill":
log.Debugln("RemoteCommand: Kill")
var args int64
if len(cmd.Arguments) == 0 {
args = -9
} else {
args, err = strconv.ParseInt(cmd.Arguments[0], 10, 32)
if err != nil {
log.Warnf("Unable to parse kill command argument[0] into int: %s\n", err)
args = -9
}
}
job.Kill(args)
case "stop":
log.Debugln("RemoteCommand: Stop")
select {
case psChan <- agents.ProcessStatCommand{}:
log.Debugln("Sending psChan a msg to sample")
default:
log.Debugln("RC Stop failed to send a sample msg on the psChan")
}
if err = job.Stop(); err != nil {
log.Fatalf("Error received while stopping sub-process: %s\n", err)
}
case "sample":
log.Debugln("RemoteCommand: Sample")
select {
case psChan <- agents.ProcessStatCommand{}:
log.Debugln("Sending psChan a msg to sample")
default:
log.Debugln("RC Sample failed to send a sample msg on the psChan")
}
case "change_sample_rate":
log.Debugln("RemoteCommand: Change Stats Sample Rate")
if len(cmd.Arguments) > 0 {
log.Debugf("change_sample_rate arg[0]: %s\n", cmd.Arguments[0])
d, err := time.ParseDuration(cmd.Arguments[0])
if err == nil {
select {
case psChan <- agents.ProcessStatCommand{
TimeUpdate: true,
NewTime: d}:
log.Debugln("Sending psChan a msg to update the ticker")
default:
log.Debugln("RC change_sample_rate failed to send a msg")
}
} else {
log.Warnf("Unparseable duration argument to command change_sample_rate")
}
} else {
log.Warnf("Missing argument to command change_sample_rate")
}
case "timer_reset":
log.Debugln("RemoteCommand: Timer Reset")
if err = timer.Reset(); err != nil {
log.Fatalf("Error received from timer calling Reset: %s\n", err)
}
case "timer_start":
log.Debugln("RemoteCommand: Timer Start")
if err = timer.Start(); err != nil {
log.Fatalf("Error received from timer calling Start: %s\n", err)
}
case "timer_stop":
log.Debugln("RemoteCommand: Timer Stop")
if err = timer.Stop(); err != nil {
log.Fatalf("Error received from timer calling Stop: %s\n", err)
}
case "timer_resume":
log.Debugln("RemoteCommand: Timer Resume")
if err = timer.Resume(); err != nil {
log.Fatalf("Error received from timer calling Resume: %s\n", err)
}
default:
log.Debugf("Unknown command: %s\n", cmd)
}
case timeoutMsg := <-timer.Done():
log.Debugf("Timer timeout message: %s\n", timeoutMsg)
if err = job.Stop(); err != nil {
log.Fatalf("Error received while stopping sub-process: %v\n", err)
// If there was an error stopping the process, kill the porcess.
job.Kill(-9)
}
case jobDone := <-job.Done():
log.Debugln("Command exited gracefully; shutting down.")
done <- jobDone
case _ = <-logSampling:
if job.StdoutByteCount() > 2*(*stdoutByteLimit) {
err = job.Kill(-9)
} else if job.StdoutByteCount() > *stdoutByteLimit {
err = job.Stop()
}
}
}
}
func redial(sess session) {
var err error
var stats agents.ProcessStats
var rc *agents.RemoteControl
// Initialize mini-router for incoming stats agent requests
go func() {
for s := range sess.psChan {
if stats == nil {
log.Warnln("No stats agents available (yet), dropping request")
} else if s.TimeUpdate {
stats.NewTicker(s.NewTime)
} else {
stats.Sample()
}
}
}()
rcKeys := []string{*rmtKey}
if sess.multiRmtKey {
rcKeys = deleteEmpty(strings.Split(*rmtKey, ","))
}
for {
sess.amqpConn, err = amqp.DialConfig(*uri, sess.amqpConfig)
if err != nil {
log.Warnf("Failed to connect to AMQP: %q", err)
// Rate limit reconnection attempts
time.Sleep(5 * time.Second)
} else {
rc, err = agents.NewRemoteControl(sess.amqpConn, rcKeys, *exchange)
if err != nil {
log.Warnf("Failed creating NewRemoteControl: %s", err)
} else {
sess.rcChan <- rc.Commands
}
if stats == nil {
// initial setup
stats, err = agents.NewProcessStats(
sess.amqpConn,
*procStatsKey,
*exchange,
&sess.job,
*statsInterval,
*msgTimeout,
*userJSON,
)
if err != nil {
log.Warnf("Failed creating NewProcessStats: %s", err)
}
} else {
err = stats.ReinitializeConnection(sess.amqpConn)
if err != nil {
log.Warnf("Failed to reinitialize process stats: %s", err)
}
}
closings := sess.amqpConn.NotifyClose(make(chan *amqp.Error))
// Wait for close notification and loop back around to reconnect
_ = <-closings
log.Debugln("Saw a notification for closed connection, looping")
}
}
}
// NewControlledProcess creates the child proc.
func NewControlledProcess(cmd string, arguments []string, doneChan chan error, stdoutLimit int64) (JobControl, error) {
var err error
j := &Job{
nil,
nil,
0,
0,
doneChan,
stdoutLimit,
nil,
}
// Drop command from cmdline arguments and pass the rest as arguments separately
var args []string
if len(arguments) > 0 {
args = arguments[1:]
}
j.Cmd = exec.Command(cmd, args...)
// Collect stdout from the process to redirect to real stdout
stdoutpipe, err := j.Cmd.StdoutPipe()
if err != nil {
return nil, fmt.Errorf("Failed to acquire stdout: %s", err)
}
stdout := iocontrol.NewMeasuredReader(stdoutpipe)
j.stdoutReader = stdout
var wg sync.WaitGroup
stdin, err := j.Cmd.StdinPipe()
if err != nil {
return nil, fmt.Errorf("Failed to acquire stdin: %s", err)
}
stderr, err := j.Cmd.StderrPipe()
if err != nil {
return nil, fmt.Errorf("Failed to acquire stderr: %s", err)
}
// Map all child processes under this tree so Kill really ends everything.
j.Cmd.SysProcAttr = &syscall.SysProcAttr{
Setpgid: true, // Set process group ID
}
log.Debugf("%#v\n", j.Cmd)
// Start the sub-process but don't wait for completion to pickup the Pid
// for resource monitoring.
err = j.Cmd.Start()
if err != nil {
return nil, fmt.Errorf("Failed to execute sub-process: %s\n", err)
}
j.Pid = j.Cmd.Process.Pid
j.Pgid, err = syscall.Getpgid(j.Pid)
if err != nil {
return nil, fmt.Errorf("Failed syscall.Getpgid: %s\n", err)
}
j.Proc, err = process.NewProcess(int32(j.Pgid))
if err != nil {
return nil, fmt.Errorf("Unable to create process.NewProcess: %s\n", err)
}
wg.Add(1)
go func(wg *sync.WaitGroup, r io.Reader) {
defer wg.Done()
io.Copy(os.Stdout, r)
log.Debugln("child closed stdout")
}(&wg, stdout)
go func(w io.WriteCloser) {
io.Copy(w, os.Stdin)
}(stdin)
wg.Add(1)
go func(wg *sync.WaitGroup, r io.Reader) {
defer wg.Done()
io.Copy(os.Stderr, r)
log.Debugln("child closed stderr")
}(&wg, stderr)
// Background waiting for the job to finish and emit a done channel message
// when complete.
go func(wg *sync.WaitGroup, j *Job) {
log.Debugln("Waiting on wg.Wait()")
wg.Wait()
log.Debugln("Waiting on Cmd.Wait()")
err := j.Cmd.Wait()
log.Debugf("Job finished: %q\n", err)
j.done <- err
}(&wg, j)
return j, nil
}
