services/crunch-dispatch-local/crunch-dispatch-local.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package main
   6
   7 // Dispatcher service for Crunch that runs containers locally.
   8
   9 import (
  10         "context"
  11         "flag"
  12         "fmt"
  13         "os"
  14         "os/exec"
  15         "os/signal"
  16         "sync"
  17         "syscall"
  18         "time"
  19
  20         "git.curoverse.com/arvados.git/sdk/go/arvados"
  21         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  22         "git.curoverse.com/arvados.git/sdk/go/dispatch"
  23         "github.com/sirupsen/logrus"
  24 )
  25
  26 var version = "dev"
  27
  28 func main() {
  29         err := doMain()
  30         if err != nil {
  31                 logrus.Fatalf("%q", err)
  32         }
  33 }
  34
  35 var (
  36         runningCmds      map[string]*exec.Cmd
  37         runningCmdsMutex sync.Mutex
  38         waitGroup        sync.WaitGroup
  39         crunchRunCommand *string
  40 )
  41
  42 func doMain() error {
  43         logger := logrus.StandardLogger()
  44         if os.Getenv("DEBUG") != "" {
  45                 logger.SetLevel(logrus.DebugLevel)
  46         }
  47         logger.Formatter = &logrus.JSONFormatter{
  48                 TimestampFormat: "2006-01-02T15:04:05.000000000Z07:00",
  49         }
  50
  51         flags := flag.NewFlagSet("crunch-dispatch-local", flag.ExitOnError)
  52
  53         pollInterval := flags.Int(
  54                 "poll-interval",
  55                 10,
  56                 "Interval in seconds to poll for queued containers")
  57
  58         crunchRunCommand = flags.String(
  59                 "crunch-run-command",
  60                 "/usr/bin/crunch-run",
  61                 "Crunch command to run container")
  62
  63         getVersion := flags.Bool(
  64                 "version",
  65                 false,
  66                 "Print version information and exit.")
  67
  68         // Parse args; omit the first arg which is the command name
  69         flags.Parse(os.Args[1:])
  70
  71         // Print version information if requested
  72         if *getVersion {
  73                 fmt.Printf("crunch-dispatch-local %s\n", version)
  74                 return nil
  75         }
  76
  77         logger.Printf("crunch-dispatch-local %s started", version)
  78
  79         runningCmds = make(map[string]*exec.Cmd)
  80
  81         arv, err := arvadosclient.MakeArvadosClient()
  82         if err != nil {
  83                 logger.Errorf("error making Arvados client: %v", err)
  84                 return err
  85         }
  86         arv.Retries = 25
  87
  88         ctx, cancel := context.WithCancel(context.Background())
  89
  90         dispatcher := dispatch.Dispatcher{
  91                 Logger:       logger,
  92                 Arv:          arv,
  93                 RunContainer: (&LocalRun{startFunc, make(chan bool, 8), ctx}).run,
  94                 PollPeriod:   time.Duration(*pollInterval) * time.Second,
  95         }
  96
  97         err = dispatcher.Run(ctx)
  98         if err != nil {
  99                 return err
 100         }
 101
 102         c := make(chan os.Signal, 1)
 103         signal.Notify(c, os.Interrupt, syscall.SIGTERM, syscall.SIGQUIT)
 104         sig := <-c
 105         logger.Printf("Received %s, shutting down", sig)
 106         signal.Stop(c)
 107
 108         cancel()
 109
 110         runningCmdsMutex.Lock()
 111         // Finished dispatching; interrupt any crunch jobs that are still running
 112         for _, cmd := range runningCmds {
 113                 cmd.Process.Signal(os.Interrupt)
 114         }
 115         runningCmdsMutex.Unlock()
 116
 117         // Wait for all running crunch jobs to complete / terminate
 118         waitGroup.Wait()
 119
 120         return nil
 121 }
 122
 123 func startFunc(container arvados.Container, cmd *exec.Cmd) error {
 124         return cmd.Start()
 125 }
 126
 127 type LocalRun struct {
 128         startCmd         func(container arvados.Container, cmd *exec.Cmd) error
 129         concurrencyLimit chan bool
 130         ctx              context.Context
 131 }
 132
 133 // Run a container.
 134 //
 135 // If the container is Locked, start a new crunch-run process and wait until
 136 // crunch-run completes.  If the priority is set to zero, set an interrupt
 137 // signal to the crunch-run process.
 138 //
 139 // If the container is in any other state, or is not Complete/Cancelled after
 140 // crunch-run terminates, mark the container as Cancelled.
 141 func (lr *LocalRun) run(dispatcher *dispatch.Dispatcher,
 142         container arvados.Container,
 143         status <-chan arvados.Container) {
 144
 145         uuid := container.UUID
 146
 147         if container.State == dispatch.Locked {
 148
 149                 select {
 150                 case lr.concurrencyLimit <- true:
 151                         break
 152                 case <-lr.ctx.Done():
 153                         return
 154                 }
 155
 156                 defer func() { <-lr.concurrencyLimit }()
 157
 158                 select {
 159                 case c := <-status:
 160                         // Check for state updates after possibly
 161                         // waiting to be ready-to-run
 162                         if c.Priority == 0 {
 163                                 goto Finish
 164                         }
 165                 default:
 166                         break
 167                 }
 168
 169                 waitGroup.Add(1)
 170                 defer waitGroup.Done()
 171
 172                 cmd := exec.Command(*crunchRunCommand, uuid)
 173                 cmd.Stdin = nil
 174                 cmd.Stderr = os.Stderr
 175                 cmd.Stdout = os.Stderr
 176
 177                 dispatcher.Logger.Printf("starting container %v", uuid)
 178
 179                 // Add this crunch job to the list of runningCmds only if we
 180                 // succeed in starting crunch-run.
 181
 182                 runningCmdsMutex.Lock()
 183                 if err := lr.startCmd(container, cmd); err != nil {
 184                         runningCmdsMutex.Unlock()
 185                         dispatcher.Logger.Warnf("error starting %q for %s: %s", *crunchRunCommand, uuid, err)
 186                         dispatcher.UpdateState(uuid, dispatch.Cancelled)
 187                 } else {
 188                         runningCmds[uuid] = cmd
 189                         runningCmdsMutex.Unlock()
 190
 191                         // Need to wait for crunch-run to exit
 192                         done := make(chan struct{})
 193
 194                         go func() {
 195                                 if _, err := cmd.Process.Wait(); err != nil {
 196                                         dispatcher.Logger.Warnf("error while waiting for crunch job to finish for %v: %q", uuid, err)
 197                                 }
 198                                 dispatcher.Logger.Debugf("sending done")
 199                                 done <- struct{}{}
 200                         }()
 201
 202                 Loop:
 203                         for {
 204                                 select {
 205                                 case <-done:
 206                                         break Loop
 207                                 case c := <-status:
 208                                         // Interrupt the child process if priority changes to 0
 209                                         if (c.State == dispatch.Locked || c.State == dispatch.Running) && c.Priority == 0 {
 210                                                 dispatcher.Logger.Printf("sending SIGINT to pid %d to cancel container %v", cmd.Process.Pid, uuid)
 211                                                 cmd.Process.Signal(os.Interrupt)
 212                                         }
 213                                 }
 214                         }
 215                         close(done)
 216
 217                         dispatcher.Logger.Printf("finished container run for %v", uuid)
 218
 219                         // Remove the crunch job from runningCmds
 220                         runningCmdsMutex.Lock()
 221                         delete(runningCmds, uuid)
 222                         runningCmdsMutex.Unlock()
 223                 }
 224         }
 225
 226 Finish:
 227
 228         // If the container is not finalized, then change it to "Cancelled".
 229         err := dispatcher.Arv.Get("containers", uuid, nil, &container)
 230         if err != nil {
 231                 dispatcher.Logger.Warnf("error getting final container state: %v", err)
 232         }
 233         if container.State == dispatch.Locked || container.State == dispatch.Running {
 234                 dispatcher.Logger.Warnf("after %q process termination, container state for %v is %q; updating it to %q",
 235                         *crunchRunCommand, uuid, container.State, dispatch.Cancelled)
 236                 dispatcher.UpdateState(uuid, dispatch.Cancelled)
 237         }
 238
 239         // drain any subsequent status changes
 240         for range status {
 241         }
 242
 243         dispatcher.Logger.Printf("finalized container %v", uuid)
 244 }