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         "log"
  14         "os"
  15         "os/exec"
  16         "os/signal"
  17         "sync"
  18         "syscall"
  19         "time"
  20
  21         "git.curoverse.com/arvados.git/sdk/go/arvados"
  22         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
  23         "git.curoverse.com/arvados.git/sdk/go/dispatch"
  24 )
  25
  26 var version = "dev"
  27
  28 func main() {
  29         err := doMain()
  30         if err != nil {
  31                 log.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         flags := flag.NewFlagSet("crunch-dispatch-local", flag.ExitOnError)
  44
  45         pollInterval := flags.Int(
  46                 "poll-interval",
  47                 10,
  48                 "Interval in seconds to poll for queued containers")
  49
  50         crunchRunCommand = flags.String(
  51                 "crunch-run-command",
  52                 "/usr/bin/crunch-run",
  53                 "Crunch command to run container")
  54
  55         getVersion := flags.Bool(
  56                 "version",
  57                 false,
  58                 "Print version information and exit.")
  59
  60         // Parse args; omit the first arg which is the command name
  61         flags.Parse(os.Args[1:])
  62
  63         // Print version information if requested
  64         if *getVersion {
  65                 fmt.Printf("crunch-dispatch-local %s\n", version)
  66                 return nil
  67         }
  68
  69         log.Printf("crunch-dispatch-local %s started", version)
  70
  71         runningCmds = make(map[string]*exec.Cmd)
  72
  73         arv, err := arvadosclient.MakeArvadosClient()
  74         if err != nil {
  75                 log.Printf("Error making Arvados client: %v", err)
  76                 return err
  77         }
  78         arv.Retries = 25
  79
  80         dispatcher := dispatch.Dispatcher{
  81                 Arv:          arv,
  82                 RunContainer: run,
  83                 PollPeriod:   time.Duration(*pollInterval) * time.Second,
  84         }
  85
  86         ctx, cancel := context.WithCancel(context.Background())
  87         err = dispatcher.Run(ctx)
  88         if err != nil {
  89                 return err
  90         }
  91
  92         c := make(chan os.Signal, 1)
  93         signal.Notify(c, os.Interrupt, syscall.SIGTERM, syscall.SIGQUIT)
  94         sig := <-c
  95         log.Printf("Received %s, shutting down", sig)
  96         signal.Stop(c)
  97
  98         cancel()
  99
 100         runningCmdsMutex.Lock()
 101         // Finished dispatching; interrupt any crunch jobs that are still running
 102         for _, cmd := range runningCmds {
 103                 cmd.Process.Signal(os.Interrupt)
 104         }
 105         runningCmdsMutex.Unlock()
 106
 107         // Wait for all running crunch jobs to complete / terminate
 108         waitGroup.Wait()
 109
 110         return nil
 111 }
 112
 113 func startFunc(container arvados.Container, cmd *exec.Cmd) error {
 114         return cmd.Start()
 115 }
 116
 117 var startCmd = startFunc
 118
 119 // Run a container.
 120 //
 121 // If the container is Locked, start a new crunch-run process and wait until
 122 // crunch-run completes.  If the priority is set to zero, set an interrupt
 123 // signal to the crunch-run process.
 124 //
 125 // If the container is in any other state, or is not Complete/Cancelled after
 126 // crunch-run terminates, mark the container as Cancelled.
 127 func run(dispatcher *dispatch.Dispatcher,
 128         container arvados.Container,
 129         status <-chan arvados.Container) {
 130
 131         uuid := container.UUID
 132
 133         if container.State == dispatch.Locked {
 134                 waitGroup.Add(1)
 135
 136                 cmd := exec.Command(*crunchRunCommand, uuid)
 137                 cmd.Stdin = nil
 138                 cmd.Stderr = os.Stderr
 139                 cmd.Stdout = os.Stderr
 140
 141                 log.Printf("Starting container %v", uuid)
 142
 143                 // Add this crunch job to the list of runningCmds only if we
 144                 // succeed in starting crunch-run.
 145
 146                 runningCmdsMutex.Lock()
 147                 if err := startCmd(container, cmd); err != nil {
 148                         runningCmdsMutex.Unlock()
 149                         log.Printf("Error starting %v for %v: %q", *crunchRunCommand, uuid, err)
 150                         dispatcher.UpdateState(uuid, dispatch.Cancelled)
 151                 } else {
 152                         runningCmds[uuid] = cmd
 153                         runningCmdsMutex.Unlock()
 154
 155                         // Need to wait for crunch-run to exit
 156                         done := make(chan struct{})
 157
 158                         go func() {
 159                                 if _, err := cmd.Process.Wait(); err != nil {
 160                                         log.Printf("Error while waiting for crunch job to finish for %v: %q", uuid, err)
 161                                 }
 162                                 log.Printf("sending done")
 163                                 done <- struct{}{}
 164                         }()
 165
 166                 Loop:
 167                         for {
 168                                 select {
 169                                 case <-done:
 170                                         break Loop
 171                                 case c := <-status:
 172                                         // Interrupt the child process if priority changes to 0
 173                                         if (c.State == dispatch.Locked || c.State == dispatch.Running) && c.Priority == 0 {
 174                                                 log.Printf("Sending SIGINT to pid %d to cancel container %v", cmd.Process.Pid, uuid)
 175                                                 cmd.Process.Signal(os.Interrupt)
 176                                         }
 177                                 }
 178                         }
 179                         close(done)
 180
 181                         log.Printf("Finished container run for %v", uuid)
 182
 183                         // Remove the crunch job from runningCmds
 184                         runningCmdsMutex.Lock()
 185                         delete(runningCmds, uuid)
 186                         runningCmdsMutex.Unlock()
 187                 }
 188                 waitGroup.Done()
 189         }
 190
 191         // If the container is not finalized, then change it to "Cancelled".
 192         err := dispatcher.Arv.Get("containers", uuid, nil, &container)
 193         if err != nil {
 194                 log.Printf("Error getting final container state: %v", err)
 195         }
 196         if container.State == dispatch.Locked || container.State == dispatch.Running {
 197                 log.Printf("After %s process termination, container state for %v is %q.  Updating it to %q",
 198                         *crunchRunCommand, container.State, uuid, dispatch.Cancelled)
 199                 dispatcher.UpdateState(uuid, dispatch.Cancelled)
 200         }
 201
 202         // drain any subsequent status changes
 203         for range status {
 204         }
 205
 206         log.Printf("Finalized container %v", uuid)
 207 }