Merge branch '18947-githttpd'

[arvados.git] / services / crunch-dispatch-local / crunch-dispatch-local.go

diff --git a/services/crunch-dispatch-local/crunch-dispatch-local.go b/services/crunch-dispatch-local/crunch-dispatch-local.go
index 2922817b557ccef70fa25f32c66b8447575abb5d..c33c2358ca3b7612ef92498edbef9a5562d6b9f5 100644 (file)
--- a/services/crunch-dispatch-local/crunch-dispatch-local.go
+++ b/services/crunch-dispatch-local/crunch-dispatch-local.go
@@ -17,6 +17,8 @@ import (
        "syscall"
        "time"
 
+       "git.arvados.org/arvados.git/lib/cmd"
+       "git.arvados.org/arvados.git/lib/config"
        "git.arvados.org/arvados.git/sdk/go/arvados"
        "git.arvados.org/arvados.git/sdk/go/arvadosclient"
        "git.arvados.org/arvados.git/sdk/go/dispatch"
@@ -25,26 +27,19 @@ import (
 
 var version = "dev"
 
-func main() {
-       err := doMain()
-       if err != nil {
-               logrus.Fatalf("%q", err)
-       }
-}
-
 var (
        runningCmds      map[string]*exec.Cmd
        runningCmdsMutex sync.Mutex
        waitGroup        sync.WaitGroup
-       crunchRunCommand *string
+       crunchRunCommand string
 )
 
-func doMain() error {
-       logger := logrus.StandardLogger()
+func main() {
+       baseLogger := logrus.StandardLogger()
        if os.Getenv("DEBUG") != "" {
-               logger.SetLevel(logrus.DebugLevel)
+               baseLogger.SetLevel(logrus.DebugLevel)
        }
-       logger.Formatter = &logrus.JSONFormatter{
+       baseLogger.Formatter = &logrus.JSONFormatter{
                TimestampFormat: "2006-01-02T15:04:05.000000000Z07:00",
        }
 
@@ -55,7 +50,7 @@ func doMain() error {
                10,
                "Interval in seconds to poll for queued containers")
 
-       crunchRunCommand = flags.String(
+       flags.StringVar(&crunchRunCommand,
                "crunch-run-command",
                "/usr/bin/crunch-run",
                "Crunch command to run container")
@@ -65,23 +60,57 @@ func doMain() error {
                false,
                "Print version information and exit.")
 
-       // Parse args; omit the first arg which is the command name
-       flags.Parse(os.Args[1:])
+       if ok, code := cmd.ParseFlags(flags, os.Args[0], os.Args[1:], "", os.Stderr); !ok {
+               os.Exit(code)
+       }
 
        // Print version information if requested
        if *getVersion {
                fmt.Printf("crunch-dispatch-local %s\n", version)
-               return nil
+               return
+       }
+
+       loader := config.NewLoader(nil, baseLogger)
+       cfg, err := loader.Load()
+       if err != nil {
+               fmt.Fprintf(os.Stderr, "error loading config: %s\n", err)
+               os.Exit(1)
+       }
+       cluster, err := cfg.GetCluster("")
+       if err != nil {
+               fmt.Fprintf(os.Stderr, "config error: %s\n", err)
+               os.Exit(1)
        }
 
+       logger := baseLogger.WithField("ClusterID", cluster.ClusterID)
        logger.Printf("crunch-dispatch-local %s started", version)
 
        runningCmds = make(map[string]*exec.Cmd)
 
+       var client arvados.Client
+       client.APIHost = cluster.Services.Controller.ExternalURL.Host
+       client.AuthToken = cluster.SystemRootToken
+       client.Insecure = cluster.TLS.Insecure
+
+       if client.APIHost != "" || client.AuthToken != "" {
+               // Copy real configs into env vars so [a]
+               // MakeArvadosClient() uses them, and [b] they get
+               // propagated to crunch-run via SLURM.
+               os.Setenv("ARVADOS_API_HOST", client.APIHost)
+               os.Setenv("ARVADOS_API_TOKEN", client.AuthToken)
+               os.Setenv("ARVADOS_API_HOST_INSECURE", "")
+               if client.Insecure {
+                       os.Setenv("ARVADOS_API_HOST_INSECURE", "1")
+               }
+               os.Setenv("ARVADOS_EXTERNAL_CLIENT", "")
+       } else {
+               logger.Warnf("Client credentials missing from config, so falling back on environment variables (deprecated).")
+       }
+
        arv, err := arvadosclient.MakeArvadosClient()
        if err != nil {
                logger.Errorf("error making Arvados client: %v", err)
-               return err
+               os.Exit(1)
        }
        arv.Retries = 25
 
@@ -90,13 +119,14 @@ func doMain() error {
        dispatcher := dispatch.Dispatcher{
                Logger:       logger,
                Arv:          arv,
-               RunContainer: (&LocalRun{startFunc, make(chan bool, 8), ctx}).run,
+               RunContainer: (&LocalRun{startFunc, make(chan bool, 8), ctx, cluster}).run,
                PollPeriod:   time.Duration(*pollInterval) * time.Second,
        }
 
        err = dispatcher.Run(ctx)
        if err != nil {
-               return err
+               logger.Error(err)
+               return
        }
 
        c := make(chan os.Signal, 1)
@@ -116,8 +146,6 @@ func doMain() error {
 
        // Wait for all running crunch jobs to complete / terminate
        waitGroup.Wait()
-
-       return nil
 }
 
 func startFunc(container arvados.Container, cmd *exec.Cmd) error {
@@ -128,6 +156,7 @@ type LocalRun struct {
        startCmd         func(container arvados.Container, cmd *exec.Cmd) error
        concurrencyLimit chan bool
        ctx              context.Context
+       cluster          *arvados.Cluster
 }
 
 // Run a container.
@@ -140,7 +169,7 @@ type LocalRun struct {
 // crunch-run terminates, mark the container as Cancelled.
 func (lr *LocalRun) run(dispatcher *dispatch.Dispatcher,
        container arvados.Container,
-       status <-chan arvados.Container) {
+       status <-chan arvados.Container) error {
 
        uuid := container.UUID
 
@@ -150,7 +179,7 @@ func (lr *LocalRun) run(dispatcher *dispatch.Dispatcher,
                case lr.concurrencyLimit <- true:
                        break
                case <-lr.ctx.Done():
-                       return
+                       return lr.ctx.Err()
                }
 
                defer func() { <-lr.concurrencyLimit }()
@@ -169,7 +198,7 @@ func (lr *LocalRun) run(dispatcher *dispatch.Dispatcher,
                waitGroup.Add(1)
                defer waitGroup.Done()
 
-               cmd := exec.Command(*crunchRunCommand, uuid)
+               cmd := exec.Command(crunchRunCommand, "--runtime-engine="+lr.cluster.Containers.RuntimeEngine, uuid)
                cmd.Stdin = nil
                cmd.Stderr = os.Stderr
                cmd.Stdout = os.Stderr
@@ -182,7 +211,7 @@ func (lr *LocalRun) run(dispatcher *dispatch.Dispatcher,
                runningCmdsMutex.Lock()
                if err := lr.startCmd(container, cmd); err != nil {
                        runningCmdsMutex.Unlock()
-                       dispatcher.Logger.Warnf("error starting %q for %s: %s", *crunchRunCommand, uuid, err)
+                       dispatcher.Logger.Warnf("error starting %q for %s: %s", crunchRunCommand, uuid, err)
                        dispatcher.UpdateState(uuid, dispatch.Cancelled)
                } else {
                        runningCmds[uuid] = cmd
@@ -232,7 +261,7 @@ Finish:
        }
        if container.State == dispatch.Locked || container.State == dispatch.Running {
                dispatcher.Logger.Warnf("after %q process termination, container state for %v is %q; updating it to %q",
-                       *crunchRunCommand, uuid, container.State, dispatch.Cancelled)
+                       crunchRunCommand, uuid, container.State, dispatch.Cancelled)
                dispatcher.UpdateState(uuid, dispatch.Cancelled)
        }
 
@@ -241,4 +270,5 @@ Finish:
        }
 
        dispatcher.Logger.Printf("finalized container %v", uuid)
+       return nil
 }