1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
31 "git.curoverse.com/arvados.git/lib/crunchstat"
32 "git.curoverse.com/arvados.git/sdk/go/arvados"
33 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
34 "git.curoverse.com/arvados.git/sdk/go/keepclient"
35 "git.curoverse.com/arvados.git/sdk/go/manifest"
37 dockertypes "github.com/docker/docker/api/types"
38 dockercontainer "github.com/docker/docker/api/types/container"
39 dockernetwork "github.com/docker/docker/api/types/network"
40 dockerclient "github.com/docker/docker/client"
45 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
46 type IArvadosClient interface {
47 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
48 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
49 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
50 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
51 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
52 Discovery(key string) (interface{}, error)
55 // ErrCancelled is the error returned when the container is cancelled.
56 var ErrCancelled = errors.New("Cancelled")
58 // IKeepClient is the minimal Keep API methods used by crunch-run.
59 type IKeepClient interface {
60 PutHB(hash string, buf []byte) (string, int, error)
61 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
65 // NewLogWriter is a factory function to create a new log writer.
66 type NewLogWriter func(name string) io.WriteCloser
68 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
70 type MkTempDir func(string, string) (string, error)
72 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
73 type ThinDockerClient interface {
74 ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
75 ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
76 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
77 ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
78 ContainerStop(ctx context.Context, container string, timeout *time.Duration) error
79 ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
80 ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
81 ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
82 ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
85 // ThinDockerClientProxy is a proxy implementation of ThinDockerClient
86 // that executes the docker requests on dockerclient.Client
87 type ThinDockerClientProxy struct {
88 Docker *dockerclient.Client
91 // ContainerAttach invokes dockerclient.Client.ContainerAttach
92 func (proxy ThinDockerClientProxy) ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error) {
93 return proxy.Docker.ContainerAttach(ctx, container, options)
96 // ContainerCreate invokes dockerclient.Client.ContainerCreate
97 func (proxy ThinDockerClientProxy) ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
98 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error) {
99 return proxy.Docker.ContainerCreate(ctx, config, hostConfig, networkingConfig, containerName)
102 // ContainerStart invokes dockerclient.Client.ContainerStart
103 func (proxy ThinDockerClientProxy) ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error {
104 return proxy.Docker.ContainerStart(ctx, container, options)
107 // ContainerStop invokes dockerclient.Client.ContainerStop
108 func (proxy ThinDockerClientProxy) ContainerStop(ctx context.Context, container string, timeout *time.Duration) error {
109 return proxy.Docker.ContainerStop(ctx, container, timeout)
112 // ContainerWait invokes dockerclient.Client.ContainerWait
113 func (proxy ThinDockerClientProxy) ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error) {
114 return proxy.Docker.ContainerWait(ctx, container, condition)
117 // ImageInspectWithRaw invokes dockerclient.Client.ImageInspectWithRaw
118 func (proxy ThinDockerClientProxy) ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error) {
119 return proxy.Docker.ImageInspectWithRaw(ctx, image)
122 // ImageLoad invokes dockerclient.Client.ImageLoad
123 func (proxy ThinDockerClientProxy) ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error) {
124 return proxy.Docker.ImageLoad(ctx, input, quiet)
127 // ImageRemove invokes dockerclient.Client.ImageRemove
128 func (proxy ThinDockerClientProxy) ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error) {
129 return proxy.Docker.ImageRemove(ctx, image, options)
132 // ContainerRunner is the main stateful struct used for a single execution of a
134 type ContainerRunner struct {
135 Docker ThinDockerClient
136 ArvClient IArvadosClient
139 ContainerConfig dockercontainer.Config
140 dockercontainer.HostConfig
145 loggingDone chan bool
146 CrunchLog *ThrottledLogger
147 Stdout io.WriteCloser
148 Stderr io.WriteCloser
149 LogCollection *CollectionWriter
156 CleanupTempDir []string
158 Volumes map[string]struct{}
160 SigChan chan os.Signal
161 ArvMountExit chan error
164 statLogger io.WriteCloser
165 statReporter *crunchstat.Reporter
166 statInterval time.Duration
168 // What we expect the container's cgroup parent to be.
169 expectCgroupParent string
170 // What we tell docker to use as the container's cgroup
171 // parent. Note: Ideally we would use the same field for both
172 // expectCgroupParent and setCgroupParent, and just make it
173 // default to "docker". However, when using docker < 1.10 with
174 // systemd, specifying a non-empty cgroup parent (even the
175 // default value "docker") hits a docker bug
176 // (https://github.com/docker/docker/issues/17126). Using two
177 // separate fields makes it possible to use the "expect cgroup
178 // parent to be X" feature even on sites where the "specify
179 // cgroup parent" feature breaks.
180 setCgroupParent string
182 cStateLock sync.Mutex
183 cStarted bool // StartContainer() succeeded
184 cCancelled bool // StopContainer() invoked
186 enableNetwork string // one of "default" or "always"
187 networkMode string // passed through to HostConfig.NetworkMode
188 arvMountLog *ThrottledLogger
191 // setupSignals sets up signal handling to gracefully terminate the underlying
192 // Docker container and update state when receiving a TERM, INT or QUIT signal.
193 func (runner *ContainerRunner) setupSignals() {
194 runner.SigChan = make(chan os.Signal, 1)
195 signal.Notify(runner.SigChan, syscall.SIGTERM)
196 signal.Notify(runner.SigChan, syscall.SIGINT)
197 signal.Notify(runner.SigChan, syscall.SIGQUIT)
199 go func(sig chan os.Signal) {
202 runner.CrunchLog.Printf("Caught signal %v", s)
208 // stop the underlying Docker container.
209 func (runner *ContainerRunner) stop() {
210 runner.cStateLock.Lock()
211 defer runner.cStateLock.Unlock()
212 if runner.cCancelled {
215 runner.cCancelled = true
217 timeout := time.Duration(10)
218 err := runner.Docker.ContainerStop(context.TODO(), runner.ContainerID, &(timeout))
220 runner.CrunchLog.Printf("StopContainer failed: %s", err)
222 // Suppress multiple calls to stop()
223 runner.cStarted = false
227 func (runner *ContainerRunner) stopSignals() {
228 if runner.SigChan != nil {
229 signal.Stop(runner.SigChan)
230 close(runner.SigChan)
234 var errorBlacklist = []string{
235 "(?ms).*[Cc]annot connect to the Docker daemon.*",
236 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
238 var brokenNodeHook *string = flag.String("broken-node-hook", "", "Script to run if node is detected to be broken (for example, Docker daemon is not running)")
240 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
241 for _, d := range errorBlacklist {
242 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
243 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
244 if *brokenNodeHook == "" {
245 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
247 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
249 c := exec.Command(*brokenNodeHook)
250 c.Stdout = runner.CrunchLog
251 c.Stderr = runner.CrunchLog
254 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
263 // LoadImage determines the docker image id from the container record and
264 // checks if it is available in the local Docker image store. If not, it loads
265 // the image from Keep.
266 func (runner *ContainerRunner) LoadImage() (err error) {
268 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
270 var collection arvados.Collection
271 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
273 return fmt.Errorf("While getting container image collection: %v", err)
275 manifest := manifest.Manifest{Text: collection.ManifestText}
276 var img, imageID string
277 for ms := range manifest.StreamIter() {
278 img = ms.FileStreamSegments[0].Name
279 if !strings.HasSuffix(img, ".tar") {
280 return fmt.Errorf("First file in the container image collection does not end in .tar")
282 imageID = img[:len(img)-4]
285 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
287 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
289 runner.CrunchLog.Print("Loading Docker image from keep")
291 var readCloser io.ReadCloser
292 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
294 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
297 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
299 return fmt.Errorf("While loading container image into Docker: %v", err)
302 defer response.Body.Close()
303 rbody, err := ioutil.ReadAll(response.Body)
305 return fmt.Errorf("Reading response to image load: %v", err)
307 runner.CrunchLog.Printf("Docker response: %s", rbody)
309 runner.CrunchLog.Print("Docker image is available")
312 runner.ContainerConfig.Image = imageID
314 runner.Kc.ClearBlockCache()
319 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
320 c = exec.Command("arv-mount", arvMountCmd...)
322 // Copy our environment, but override ARVADOS_API_TOKEN with
323 // the container auth token.
325 for _, s := range os.Environ() {
326 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
327 c.Env = append(c.Env, s)
330 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
332 runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
333 c.Stdout = runner.arvMountLog
334 c.Stderr = runner.arvMountLog
336 runner.CrunchLog.Printf("Running %v", c.Args)
343 statReadme := make(chan bool)
344 runner.ArvMountExit = make(chan error)
349 time.Sleep(100 * time.Millisecond)
350 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
362 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
364 runner.ArvMountExit <- mnterr
365 close(runner.ArvMountExit)
371 case err := <-runner.ArvMountExit:
372 runner.ArvMount = nil
380 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
381 if runner.ArvMountPoint == "" {
382 runner.ArvMountPoint, err = runner.MkTempDir("", prefix)
387 func (runner *ContainerRunner) SetupMounts() (err error) {
388 err = runner.SetupArvMountPoint("keep")
390 return fmt.Errorf("While creating keep mount temp dir: %v", err)
395 arvMountCmd := []string{
399 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
401 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
402 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
405 collectionPaths := []string{}
407 runner.Volumes = make(map[string]struct{})
408 needCertMount := true
411 for bind := range runner.Container.Mounts {
412 binds = append(binds, bind)
416 for _, bind := range binds {
417 mnt := runner.Container.Mounts[bind]
418 if bind == "stdout" || bind == "stderr" {
419 // Is it a "file" mount kind?
420 if mnt.Kind != "file" {
421 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
424 // Does path start with OutputPath?
425 prefix := runner.Container.OutputPath
426 if !strings.HasSuffix(prefix, "/") {
429 if !strings.HasPrefix(mnt.Path, prefix) {
430 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
435 // Is it a "collection" mount kind?
436 if mnt.Kind != "collection" && mnt.Kind != "json" {
437 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
441 if bind == "/etc/arvados/ca-certificates.crt" {
442 needCertMount = false
445 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
446 if mnt.Kind != "collection" {
447 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
452 case mnt.Kind == "collection" && bind != "stdin":
454 if mnt.UUID != "" && mnt.PortableDataHash != "" {
455 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
459 return fmt.Errorf("Writing to existing collections currently not permitted.")
462 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
463 } else if mnt.PortableDataHash != "" {
465 return fmt.Errorf("Can never write to a collection specified by portable data hash")
467 idx := strings.Index(mnt.PortableDataHash, "/")
469 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
470 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
471 runner.Container.Mounts[bind] = mnt
473 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
474 if mnt.Path != "" && mnt.Path != "." {
475 if strings.HasPrefix(mnt.Path, "./") {
476 mnt.Path = mnt.Path[2:]
477 } else if strings.HasPrefix(mnt.Path, "/") {
478 mnt.Path = mnt.Path[1:]
480 src += "/" + mnt.Path
483 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
484 arvMountCmd = append(arvMountCmd, "--mount-tmp")
485 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
489 if bind == runner.Container.OutputPath {
490 runner.HostOutputDir = src
491 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
492 return fmt.Errorf("Writable mount points are not permitted underneath the output_path: %v", bind)
494 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
496 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
498 collectionPaths = append(collectionPaths, src)
500 case mnt.Kind == "tmp":
502 tmpdir, err = runner.MkTempDir("", "")
504 return fmt.Errorf("While creating mount temp dir: %v", err)
506 st, staterr := os.Stat(tmpdir)
508 return fmt.Errorf("While Stat on temp dir: %v", staterr)
510 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
512 return fmt.Errorf("While Chmod temp dir: %v", err)
514 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
515 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
516 if bind == runner.Container.OutputPath {
517 runner.HostOutputDir = tmpdir
520 case mnt.Kind == "json":
521 jsondata, err := json.Marshal(mnt.Content)
523 return fmt.Errorf("encoding json data: %v", err)
525 // Create a tempdir with a single file
526 // (instead of just a tempfile): this way we
527 // can ensure the file is world-readable
528 // inside the container, without having to
529 // make it world-readable on the docker host.
530 tmpdir, err := runner.MkTempDir("", "")
532 return fmt.Errorf("creating temp dir: %v", err)
534 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
535 tmpfn := filepath.Join(tmpdir, "mountdata.json")
536 err = ioutil.WriteFile(tmpfn, jsondata, 0644)
538 return fmt.Errorf("writing temp file: %v", err)
540 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
542 case mnt.Kind == "git_tree":
543 tmpdir, err := runner.MkTempDir("", "")
545 return fmt.Errorf("creating temp dir: %v", err)
547 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
548 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir)
552 bind := tmpdir + ":" + bind
556 runner.Binds = append(runner.Binds, bind)
560 if runner.HostOutputDir == "" {
561 return fmt.Errorf("Output path does not correspond to a writable mount point")
564 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
565 for _, certfile := range arvadosclient.CertFiles {
566 _, err := os.Stat(certfile)
568 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
575 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
577 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
579 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
581 token, err := runner.ContainerToken()
583 return fmt.Errorf("could not get container token: %s", err)
586 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
588 return fmt.Errorf("While trying to start arv-mount: %v", err)
591 for _, p := range collectionPaths {
594 return fmt.Errorf("While checking that input files exist: %v", err)
601 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
602 // Handle docker log protocol
603 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
605 header := make([]byte, 8)
607 _, readerr := io.ReadAtLeast(containerReader, header, 8)
610 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
613 _, readerr = io.CopyN(runner.Stdout, containerReader, readsize)
616 _, readerr = io.CopyN(runner.Stderr, containerReader, readsize)
621 if readerr != io.EOF {
622 runner.CrunchLog.Printf("While reading docker logs: %v", readerr)
625 closeerr := runner.Stdout.Close()
627 runner.CrunchLog.Printf("While closing stdout logs: %v", closeerr)
630 closeerr = runner.Stderr.Close()
632 runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr)
635 if runner.statReporter != nil {
636 runner.statReporter.Stop()
637 closeerr = runner.statLogger.Close()
639 runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr)
643 runner.loggingDone <- true
644 close(runner.loggingDone)
650 func (runner *ContainerRunner) StartCrunchstat() {
651 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
652 runner.statReporter = &crunchstat.Reporter{
653 CID: runner.ContainerID,
654 Logger: log.New(runner.statLogger, "", 0),
655 CgroupParent: runner.expectCgroupParent,
656 CgroupRoot: runner.cgroupRoot,
657 PollPeriod: runner.statInterval,
659 runner.statReporter.Start()
662 type infoCommand struct {
667 // LogNodeInfo gathers node information and store it on the log for debugging
669 func (runner *ContainerRunner) LogNodeInfo() (err error) {
670 w := runner.NewLogWriter("node-info")
671 logger := log.New(w, "node-info", 0)
673 commands := []infoCommand{
675 label: "Host Information",
676 cmd: []string{"uname", "-a"},
679 label: "CPU Information",
680 cmd: []string{"cat", "/proc/cpuinfo"},
683 label: "Memory Information",
684 cmd: []string{"cat", "/proc/meminfo"},
688 cmd: []string{"df", "-m", "/", os.TempDir()},
691 label: "Disk INodes",
692 cmd: []string{"df", "-i", "/", os.TempDir()},
696 // Run commands with informational output to be logged.
698 for _, command := range commands {
699 out, err = exec.Command(command.cmd[0], command.cmd[1:]...).CombinedOutput()
701 return fmt.Errorf("While running command %q: %v",
704 logger.Println(command.label)
705 for _, line := range strings.Split(string(out), "\n") {
706 logger.Println(" ", line)
712 return fmt.Errorf("While closing node-info logs: %v", err)
717 // LogContainerRecord gets and saves the raw JSON container record from the API server
718 func (runner *ContainerRunner) LogContainerRecord() (err error) {
720 ArvClient: runner.ArvClient,
721 UUID: runner.Container.UUID,
722 loggingStream: "container",
723 writeCloser: runner.LogCollection.Open("container.json"),
726 // Get Container record JSON from the API Server
727 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
729 return fmt.Errorf("While retrieving container record from the API server: %v", err)
733 dec := json.NewDecoder(reader)
735 var cr map[string]interface{}
736 if err = dec.Decode(&cr); err != nil {
737 return fmt.Errorf("While decoding the container record JSON response: %v", err)
739 // Re-encode it using indentation to improve readability
740 enc := json.NewEncoder(w)
741 enc.SetIndent("", " ")
742 if err = enc.Encode(cr); err != nil {
743 return fmt.Errorf("While logging the JSON container record: %v", err)
747 return fmt.Errorf("While closing container.json log: %v", err)
752 // AttachStreams connects the docker container stdin, stdout and stderr logs
753 // to the Arvados logger which logs to Keep and the API server logs table.
754 func (runner *ContainerRunner) AttachStreams() (err error) {
756 runner.CrunchLog.Print("Attaching container streams")
758 // If stdin mount is provided, attach it to the docker container
759 var stdinRdr arvados.File
761 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
762 if stdinMnt.Kind == "collection" {
763 var stdinColl arvados.Collection
764 collId := stdinMnt.UUID
766 collId = stdinMnt.PortableDataHash
768 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
770 return fmt.Errorf("While getting stding collection: %v", err)
773 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
774 if os.IsNotExist(err) {
775 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
776 } else if err != nil {
777 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
779 } else if stdinMnt.Kind == "json" {
780 stdinJson, err = json.Marshal(stdinMnt.Content)
782 return fmt.Errorf("While encoding stdin json data: %v", err)
787 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
788 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
789 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
791 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
794 runner.loggingDone = make(chan bool)
796 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
797 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
801 runner.Stdout = stdoutFile
803 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
806 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
807 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
811 runner.Stderr = stderrFile
813 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
818 _, err := io.Copy(response.Conn, stdinRdr)
820 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
824 response.CloseWrite()
826 } else if len(stdinJson) != 0 {
828 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
830 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
833 response.CloseWrite()
837 go runner.ProcessDockerAttach(response.Reader)
842 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
843 stdoutPath := mntPath[len(runner.Container.OutputPath):]
844 index := strings.LastIndex(stdoutPath, "/")
846 subdirs := stdoutPath[:index]
848 st, err := os.Stat(runner.HostOutputDir)
850 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
852 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
853 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
855 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
859 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
861 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
864 return stdoutFile, nil
867 // CreateContainer creates the docker container.
868 func (runner *ContainerRunner) CreateContainer() error {
869 runner.CrunchLog.Print("Creating Docker container")
871 runner.ContainerConfig.Cmd = runner.Container.Command
872 if runner.Container.Cwd != "." {
873 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
876 for k, v := range runner.Container.Environment {
877 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
880 runner.ContainerConfig.Volumes = runner.Volumes
882 runner.HostConfig = dockercontainer.HostConfig{
884 LogConfig: dockercontainer.LogConfig{
887 Resources: dockercontainer.Resources{
888 CgroupParent: runner.setCgroupParent,
892 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
893 tok, err := runner.ContainerToken()
897 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
898 "ARVADOS_API_TOKEN="+tok,
899 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
900 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
902 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
904 if runner.enableNetwork == "always" {
905 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
907 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
911 _, stdinUsed := runner.Container.Mounts["stdin"]
912 runner.ContainerConfig.OpenStdin = stdinUsed
913 runner.ContainerConfig.StdinOnce = stdinUsed
914 runner.ContainerConfig.AttachStdin = stdinUsed
915 runner.ContainerConfig.AttachStdout = true
916 runner.ContainerConfig.AttachStderr = true
918 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
920 return fmt.Errorf("While creating container: %v", err)
923 runner.ContainerID = createdBody.ID
925 return runner.AttachStreams()
928 // StartContainer starts the docker container created by CreateContainer.
929 func (runner *ContainerRunner) StartContainer() error {
930 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
931 runner.cStateLock.Lock()
932 defer runner.cStateLock.Unlock()
933 if runner.cCancelled {
936 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
937 dockertypes.ContainerStartOptions{})
940 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
941 advice = fmt.Sprintf("\nPossible causes: command %q is missing, the interpreter given in #! is missing, or script has Windows line endings.", runner.Container.Command[0])
943 return fmt.Errorf("could not start container: %v%s", err, advice)
945 runner.cStarted = true
949 // WaitFinish waits for the container to terminate, capture the exit code, and
950 // close the stdout/stderr logging.
951 func (runner *ContainerRunner) WaitFinish() (err error) {
952 runner.CrunchLog.Print("Waiting for container to finish")
954 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, "not-running")
957 <-runner.ArvMountExit
959 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
964 var waitBody dockercontainer.ContainerWaitOKBody
966 case waitBody = <-waitOk:
967 case err = <-waitErr:
970 // Container isn't running any more
971 runner.cStarted = false
974 return fmt.Errorf("container wait: %v", err)
977 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
978 code := int(waitBody.StatusCode)
979 runner.ExitCode = &code
981 // wait for stdout/stderr to complete
987 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
989 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
990 // Follow symlinks if necessary
995 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
996 if followed >= limitFollowSymlinks {
997 // Got stuck in a loop or just a pathological number of links, give up.
998 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1002 readlinktgt, err = os.Readlink(nextlink)
1008 if !strings.HasPrefix(tgt, "/") {
1009 // Relative symlink, resolve it to host path
1010 tgt = filepath.Join(filepath.Dir(path), tgt)
1012 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1013 // Absolute symlink to container output path, adjust it to host output path.
1014 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1016 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1017 // After dereferencing, symlink target must either be
1018 // within output directory, or must point to a
1019 // collection mount.
1020 err = ErrNotInOutputDir
1024 info, err = os.Lstat(tgt)
1027 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1028 path[len(runner.HostOutputDir):], readlinktgt, err)
1038 var limitFollowSymlinks = 10
1040 // UploadFile uploads files within the output directory, with special handling
1041 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1042 // from the keep mount into the output manifestText. Ensure that whether
1043 // symlinks are relative or absolute, every symlink target (even targets that
1044 // are symlinks themselves) must point to a path in either the output directory
1045 // or a collection mount.
1047 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1048 func (runner *ContainerRunner) UploadOutputFile(
1053 walkUpload *WalkUpload,
1054 relocateFrom string,
1056 followed int) (manifestText string, err error) {
1058 if info.Mode().IsDir() {
1066 if followed >= limitFollowSymlinks {
1067 // Got stuck in a loop or just a pathological number of
1068 // directory links, give up.
1069 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1073 // When following symlinks, the source path may need to be logically
1074 // relocated to some other path within the output collection. Remove
1075 // the relocateFrom prefix and replace it with relocateTo.
1076 relocated := relocateTo + path[len(relocateFrom):]
1078 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1079 if derefErr != nil && derefErr != ErrNotInOutputDir {
1083 // go through mounts and try reverse map to collection reference
1084 for _, bind := range binds {
1085 mnt := runner.Container.Mounts[bind]
1086 if tgt == bind || strings.HasPrefix(tgt, bind+"/") {
1087 // get path relative to bind
1088 targetSuffix := tgt[len(bind):]
1090 // Copy mount and adjust the path to add path relative to the bind
1091 adjustedMount := mnt
1092 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1094 // Terminates in this keep mount, so add the
1095 // manifest text at appropriate location.
1096 outputSuffix := path[len(runner.HostOutputDir):]
1097 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1102 // If target is not a collection mount, it must be located within the
1103 // output directory, otherwise it is an error.
1104 if derefErr == ErrNotInOutputDir {
1105 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1106 path[len(runner.HostOutputDir):], readlinktgt)
1110 if info.Mode().IsRegular() {
1111 return "", walkUpload.UploadFile(relocated, tgt)
1114 if info.Mode().IsDir() {
1115 // Symlink leads to directory. Walk() doesn't follow
1116 // directory symlinks, so we walk the target directory
1117 // instead. Within the walk, file paths are relocated
1118 // so they appear under the original symlink path.
1119 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1121 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1122 binds, walkUpload, tgt, relocated, followed+1)
1124 manifestText = manifestText + m
1134 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1135 func (runner *ContainerRunner) CaptureOutput() error {
1136 if runner.finalState != "Complete" {
1140 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1141 // Output may have been set directly by the container, so
1142 // refresh the container record to check.
1143 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1144 nil, &runner.Container)
1148 if runner.Container.Output != "" {
1149 // Container output is already set.
1150 runner.OutputPDH = &runner.Container.Output
1155 if runner.HostOutputDir == "" {
1159 _, err := os.Stat(runner.HostOutputDir)
1161 return fmt.Errorf("While checking host output path: %v", err)
1164 // Pre-populate output from the configured mount points
1166 for bind, mnt := range runner.Container.Mounts {
1167 if mnt.Kind == "collection" {
1168 binds = append(binds, bind)
1173 var manifestText string
1175 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1176 _, err = os.Stat(collectionMetafile)
1178 // Regular directory
1180 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1181 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1184 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1185 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1187 manifestText = manifestText + m
1192 cw.EndUpload(walkUpload)
1195 return fmt.Errorf("While uploading output files: %v", err)
1198 m, err = cw.ManifestText()
1199 manifestText = manifestText + m
1201 return fmt.Errorf("While uploading output files: %v", err)
1204 // FUSE mount directory
1205 file, openerr := os.Open(collectionMetafile)
1207 return fmt.Errorf("While opening FUSE metafile: %v", err)
1211 var rec arvados.Collection
1212 err = json.NewDecoder(file).Decode(&rec)
1214 return fmt.Errorf("While reading FUSE metafile: %v", err)
1216 manifestText = rec.ManifestText
1219 for _, bind := range binds {
1220 mnt := runner.Container.Mounts[bind]
1222 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1224 if bindSuffix == bind || len(bindSuffix) <= 0 {
1225 // either does not start with OutputPath or is OutputPath itself
1229 if mnt.ExcludeFromOutput == true {
1233 // append to manifest_text
1234 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1239 manifestText = manifestText + m
1243 var response arvados.Collection
1244 manifest := manifest.Manifest{Text: manifestText}
1245 manifestText = manifest.Extract(".", ".").Text
1246 err = runner.ArvClient.Create("collections",
1248 "ensure_unique_name": true,
1249 "collection": arvadosclient.Dict{
1251 "name": "output for " + runner.Container.UUID,
1252 "manifest_text": manifestText}},
1255 return fmt.Errorf("While creating output collection: %v", err)
1257 runner.OutputPDH = &response.PortableDataHash
1261 var outputCollections = make(map[string]arvados.Collection)
1263 // Fetch the collection for the mnt.PortableDataHash
1264 // Return the manifest_text fragment corresponding to the specified mnt.Path
1265 // after making any required updates.
1267 // If mnt.Path is not specified,
1268 // return the entire manifest_text after replacing any "." with bindSuffix
1269 // If mnt.Path corresponds to one stream,
1270 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1271 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1272 // for that stream after replacing stream name with bindSuffix minus the last word
1273 // and the file name with last word of the bindSuffix
1274 // Allowed path examples:
1276 // "path":"/subdir1"
1277 // "path":"/subdir1/subdir2"
1278 // "path":"/subdir/filename" etc
1279 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1280 collection := outputCollections[mnt.PortableDataHash]
1281 if collection.PortableDataHash == "" {
1282 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1284 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1286 outputCollections[mnt.PortableDataHash] = collection
1289 if collection.ManifestText == "" {
1290 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1294 mft := manifest.Manifest{Text: collection.ManifestText}
1295 extracted := mft.Extract(mnt.Path, bindSuffix)
1296 if extracted.Err != nil {
1297 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1299 return extracted.Text, nil
1302 func (runner *ContainerRunner) CleanupDirs() {
1303 if runner.ArvMount != nil {
1305 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1306 umount.Stdout = runner.CrunchLog
1307 umount.Stderr = runner.CrunchLog
1308 runner.CrunchLog.Printf("Running %v", umount.Args)
1309 umnterr := umount.Start()
1312 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1314 // If arv-mount --unmount gets stuck for any reason, we
1315 // don't want to wait for it forever. Do Wait() in a goroutine
1316 // so it doesn't block crunch-run.
1317 umountExit := make(chan error)
1319 mnterr := umount.Wait()
1321 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1323 umountExit <- mnterr
1326 for again := true; again; {
1332 case <-runner.ArvMountExit:
1334 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1335 runner.CrunchLog.Printf("Timed out waiting for unmount")
1337 umount.Process.Kill()
1339 runner.ArvMount.Process.Kill()
1345 if runner.ArvMountPoint != "" {
1346 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1347 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1351 for _, tmpdir := range runner.CleanupTempDir {
1352 if rmerr := os.RemoveAll(tmpdir); rmerr != nil {
1353 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
1358 // CommitLogs posts the collection containing the final container logs.
1359 func (runner *ContainerRunner) CommitLogs() error {
1360 runner.CrunchLog.Print(runner.finalState)
1362 if runner.arvMountLog != nil {
1363 runner.arvMountLog.Close()
1365 runner.CrunchLog.Close()
1367 // Closing CrunchLog above allows them to be committed to Keep at this
1368 // point, but re-open crunch log with ArvClient in case there are any
1369 // other further errors (such as failing to write the log to Keep!)
1370 // while shutting down
1371 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1372 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1373 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1375 if runner.LogsPDH != nil {
1376 // If we have already assigned something to LogsPDH,
1377 // we must be closing the re-opened log, which won't
1378 // end up getting attached to the container record and
1379 // therefore doesn't need to be saved as a collection
1380 // -- it exists only to send logs to other channels.
1384 mt, err := runner.LogCollection.ManifestText()
1386 return fmt.Errorf("While creating log manifest: %v", err)
1389 var response arvados.Collection
1390 err = runner.ArvClient.Create("collections",
1392 "ensure_unique_name": true,
1393 "collection": arvadosclient.Dict{
1395 "name": "logs for " + runner.Container.UUID,
1396 "manifest_text": mt}},
1399 return fmt.Errorf("While creating log collection: %v", err)
1401 runner.LogsPDH = &response.PortableDataHash
1405 // UpdateContainerRunning updates the container state to "Running"
1406 func (runner *ContainerRunner) UpdateContainerRunning() error {
1407 runner.cStateLock.Lock()
1408 defer runner.cStateLock.Unlock()
1409 if runner.cCancelled {
1412 return runner.ArvClient.Update("containers", runner.Container.UUID,
1413 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1416 // ContainerToken returns the api_token the container (and any
1417 // arv-mount processes) are allowed to use.
1418 func (runner *ContainerRunner) ContainerToken() (string, error) {
1419 if runner.token != "" {
1420 return runner.token, nil
1423 var auth arvados.APIClientAuthorization
1424 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1428 runner.token = auth.APIToken
1429 return runner.token, nil
1432 // UpdateContainerComplete updates the container record state on API
1433 // server to "Complete" or "Cancelled"
1434 func (runner *ContainerRunner) UpdateContainerFinal() error {
1435 update := arvadosclient.Dict{}
1436 update["state"] = runner.finalState
1437 if runner.LogsPDH != nil {
1438 update["log"] = *runner.LogsPDH
1440 if runner.finalState == "Complete" {
1441 if runner.ExitCode != nil {
1442 update["exit_code"] = *runner.ExitCode
1444 if runner.OutputPDH != nil {
1445 update["output"] = *runner.OutputPDH
1448 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1451 // IsCancelled returns the value of Cancelled, with goroutine safety.
1452 func (runner *ContainerRunner) IsCancelled() bool {
1453 runner.cStateLock.Lock()
1454 defer runner.cStateLock.Unlock()
1455 return runner.cCancelled
1458 // NewArvLogWriter creates an ArvLogWriter
1459 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1460 return &ArvLogWriter{
1461 ArvClient: runner.ArvClient,
1462 UUID: runner.Container.UUID,
1463 loggingStream: name,
1464 writeCloser: runner.LogCollection.Open(name + ".txt")}
1467 // Run the full container lifecycle.
1468 func (runner *ContainerRunner) Run() (err error) {
1469 runner.CrunchLog.Printf("crunch-run %s started", version)
1470 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1472 hostname, hosterr := os.Hostname()
1474 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1476 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1479 runner.finalState = "Queued"
1482 runner.stopSignals()
1483 runner.CleanupDirs()
1485 runner.CrunchLog.Printf("crunch-run finished")
1486 runner.CrunchLog.Close()
1490 // checkErr prints e (unless it's nil) and sets err to
1491 // e (unless err is already non-nil). Thus, if err
1492 // hasn't already been assigned when Run() returns,
1493 // this cleanup func will cause Run() to return the
1494 // first non-nil error that is passed to checkErr().
1495 checkErr := func(e error) {
1499 runner.CrunchLog.Print(e)
1503 if runner.finalState == "Complete" {
1504 // There was an error in the finalization.
1505 runner.finalState = "Cancelled"
1509 // Log the error encountered in Run(), if any
1512 if runner.finalState == "Queued" {
1513 runner.UpdateContainerFinal()
1517 if runner.IsCancelled() {
1518 runner.finalState = "Cancelled"
1519 // but don't return yet -- we still want to
1520 // capture partial output and write logs
1523 checkErr(runner.CaptureOutput())
1524 checkErr(runner.CommitLogs())
1525 checkErr(runner.UpdateContainerFinal())
1528 err = runner.fetchContainerRecord()
1533 // setup signal handling
1534 runner.setupSignals()
1536 // check for and/or load image
1537 err = runner.LoadImage()
1539 if !runner.checkBrokenNode(err) {
1540 // Failed to load image but not due to a "broken node"
1541 // condition, probably user error.
1542 runner.finalState = "Cancelled"
1544 err = fmt.Errorf("While loading container image: %v", err)
1548 // set up FUSE mount and binds
1549 err = runner.SetupMounts()
1551 runner.finalState = "Cancelled"
1552 err = fmt.Errorf("While setting up mounts: %v", err)
1556 err = runner.CreateContainer()
1561 // Gather and record node information
1562 err = runner.LogNodeInfo()
1566 // Save container.json record on log collection
1567 err = runner.LogContainerRecord()
1572 if runner.IsCancelled() {
1576 err = runner.UpdateContainerRunning()
1580 runner.finalState = "Cancelled"
1582 runner.StartCrunchstat()
1584 err = runner.StartContainer()
1586 runner.checkBrokenNode(err)
1590 err = runner.WaitFinish()
1592 runner.finalState = "Complete"
1597 // Fetch the current container record (uuid = runner.Container.UUID)
1598 // into runner.Container.
1599 func (runner *ContainerRunner) fetchContainerRecord() error {
1600 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1602 return fmt.Errorf("error fetching container record: %v", err)
1604 defer reader.Close()
1606 dec := json.NewDecoder(reader)
1608 err = dec.Decode(&runner.Container)
1610 return fmt.Errorf("error decoding container record: %v", err)
1615 // NewContainerRunner creates a new container runner.
1616 func NewContainerRunner(api IArvadosClient,
1618 docker ThinDockerClient,
1619 containerUUID string) *ContainerRunner {
1621 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1622 cr.NewLogWriter = cr.NewArvLogWriter
1623 cr.RunArvMount = cr.ArvMountCmd
1624 cr.MkTempDir = ioutil.TempDir
1625 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1626 cr.Container.UUID = containerUUID
1627 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1628 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1630 loadLogThrottleParams(api)
1636 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1637 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1638 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1639 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1640 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1641 enableNetwork := flag.String("container-enable-networking", "default",
1642 `Specify if networking should be enabled for container. One of 'default', 'always':
1643 default: only enable networking if container requests it.
1644 always: containers always have networking enabled
1646 networkMode := flag.String("container-network-mode", "default",
1647 `Set networking mode for container. Corresponds to Docker network mode (--net).
1649 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1650 getVersion := flag.Bool("version", false, "Print version information and exit.")
1653 // Print version information if requested
1655 fmt.Printf("crunch-run %s\n", version)
1659 log.Printf("crunch-run %s started", version)
1661 containerId := flag.Arg(0)
1663 if *caCertsPath != "" {
1664 arvadosclient.CertFiles = []string{*caCertsPath}
1667 api, err := arvadosclient.MakeArvadosClient()
1669 log.Fatalf("%s: %v", containerId, err)
1673 kc, kcerr := keepclient.MakeKeepClient(api)
1675 log.Fatalf("%s: %v", containerId, kcerr)
1677 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1680 // API version 1.21 corresponds to Docker 1.9, which is currently the
1681 // minimum version we want to support.
1682 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1683 dockerClientProxy := ThinDockerClientProxy{Docker: docker}
1685 cr := NewContainerRunner(api, kc, dockerClientProxy, containerId)
1687 if dockererr != nil {
1688 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1689 cr.checkBrokenNode(dockererr)
1690 cr.CrunchLog.Close()
1694 cr.statInterval = *statInterval
1695 cr.cgroupRoot = *cgroupRoot
1696 cr.expectCgroupParent = *cgroupParent
1697 cr.enableNetwork = *enableNetwork
1698 cr.networkMode = *networkMode
1699 if *cgroupParentSubsystem != "" {
1700 p := findCgroup(*cgroupParentSubsystem)
1701 cr.setCgroupParent = p
1702 cr.expectCgroupParent = p
1707 if *memprofile != "" {
1708 f, err := os.Create(*memprofile)
1710 log.Printf("could not create memory profile: ", err)
1712 runtime.GC() // get up-to-date statistics
1713 if err := pprof.WriteHeapProfile(f); err != nil {
1714 log.Printf("could not write memory profile: ", err)
1716 closeerr := f.Close()
1717 if closeerr != nil {
1718 log.Printf("closing memprofile file: ", err)
1723 log.Fatalf("%s: %v", containerId, runerr)