1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
30 "git.curoverse.com/arvados.git/lib/crunchstat"
31 "git.curoverse.com/arvados.git/sdk/go/arvados"
32 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
33 "git.curoverse.com/arvados.git/sdk/go/keepclient"
34 "git.curoverse.com/arvados.git/sdk/go/manifest"
35 "golang.org/x/net/context"
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 ContainerRemove(ctx context.Context, container string, options dockertypes.ContainerRemoveOptions) 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 // ContainerRunner is the main stateful struct used for a single execution of a
87 type ContainerRunner struct {
88 Docker ThinDockerClient
89 ArvClient IArvadosClient
92 ContainerConfig dockercontainer.Config
93 dockercontainer.HostConfig
99 CrunchLog *ThrottledLogger
100 Stdout io.WriteCloser
101 Stderr io.WriteCloser
102 LogCollection *CollectionWriter
110 Volumes map[string]struct{}
112 SigChan chan os.Signal
113 ArvMountExit chan error
117 statLogger io.WriteCloser
118 statReporter *crunchstat.Reporter
119 hoststatLogger io.WriteCloser
120 hoststatReporter *crunchstat.Reporter
121 statInterval time.Duration
123 // What we expect the container's cgroup parent to be.
124 expectCgroupParent string
125 // What we tell docker to use as the container's cgroup
126 // parent. Note: Ideally we would use the same field for both
127 // expectCgroupParent and setCgroupParent, and just make it
128 // default to "docker". However, when using docker < 1.10 with
129 // systemd, specifying a non-empty cgroup parent (even the
130 // default value "docker") hits a docker bug
131 // (https://github.com/docker/docker/issues/17126). Using two
132 // separate fields makes it possible to use the "expect cgroup
133 // parent to be X" feature even on sites where the "specify
134 // cgroup parent" feature breaks.
135 setCgroupParent string
137 cStateLock sync.Mutex
138 cCancelled bool // StopContainer() invoked
140 enableNetwork string // one of "default" or "always"
141 networkMode string // passed through to HostConfig.NetworkMode
142 arvMountLog *ThrottledLogger
145 // setupSignals sets up signal handling to gracefully terminate the underlying
146 // Docker container and update state when receiving a TERM, INT or QUIT signal.
147 func (runner *ContainerRunner) setupSignals() {
148 runner.SigChan = make(chan os.Signal, 1)
149 signal.Notify(runner.SigChan, syscall.SIGTERM)
150 signal.Notify(runner.SigChan, syscall.SIGINT)
151 signal.Notify(runner.SigChan, syscall.SIGQUIT)
153 go func(sig chan os.Signal) {
160 // stop the underlying Docker container.
161 func (runner *ContainerRunner) stop(sig os.Signal) {
162 runner.cStateLock.Lock()
163 defer runner.cStateLock.Unlock()
165 runner.CrunchLog.Printf("caught signal: %v", sig)
167 if runner.ContainerID == "" {
170 runner.cCancelled = true
171 runner.CrunchLog.Printf("removing container")
172 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
174 runner.CrunchLog.Printf("error removing container: %s", err)
178 var errorBlacklist = []string{
179 "(?ms).*[Cc]annot connect to the Docker daemon.*",
180 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
182 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)")
184 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
185 for _, d := range errorBlacklist {
186 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
187 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
188 if *brokenNodeHook == "" {
189 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
191 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
193 c := exec.Command(*brokenNodeHook)
194 c.Stdout = runner.CrunchLog
195 c.Stderr = runner.CrunchLog
198 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
207 // LoadImage determines the docker image id from the container record and
208 // checks if it is available in the local Docker image store. If not, it loads
209 // the image from Keep.
210 func (runner *ContainerRunner) LoadImage() (err error) {
212 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
214 var collection arvados.Collection
215 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
217 return fmt.Errorf("While getting container image collection: %v", err)
219 manifest := manifest.Manifest{Text: collection.ManifestText}
220 var img, imageID string
221 for ms := range manifest.StreamIter() {
222 img = ms.FileStreamSegments[0].Name
223 if !strings.HasSuffix(img, ".tar") {
224 return fmt.Errorf("First file in the container image collection does not end in .tar")
226 imageID = img[:len(img)-4]
229 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
231 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
233 runner.CrunchLog.Print("Loading Docker image from keep")
235 var readCloser io.ReadCloser
236 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
238 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
241 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
243 return fmt.Errorf("While loading container image into Docker: %v", err)
246 defer response.Body.Close()
247 rbody, err := ioutil.ReadAll(response.Body)
249 return fmt.Errorf("Reading response to image load: %v", err)
251 runner.CrunchLog.Printf("Docker response: %s", rbody)
253 runner.CrunchLog.Print("Docker image is available")
256 runner.ContainerConfig.Image = imageID
258 runner.Kc.ClearBlockCache()
263 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
264 c = exec.Command("arv-mount", arvMountCmd...)
266 // Copy our environment, but override ARVADOS_API_TOKEN with
267 // the container auth token.
269 for _, s := range os.Environ() {
270 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
271 c.Env = append(c.Env, s)
274 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
276 runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
277 c.Stdout = runner.arvMountLog
278 c.Stderr = runner.arvMountLog
280 runner.CrunchLog.Printf("Running %v", c.Args)
287 statReadme := make(chan bool)
288 runner.ArvMountExit = make(chan error)
293 time.Sleep(100 * time.Millisecond)
294 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
306 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
308 runner.ArvMountExit <- mnterr
309 close(runner.ArvMountExit)
315 case err := <-runner.ArvMountExit:
316 runner.ArvMount = nil
324 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
325 if runner.ArvMountPoint == "" {
326 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
331 func copyfile(src string, dst string) (err error) {
332 srcfile, err := os.Open(src)
337 os.MkdirAll(path.Dir(dst), 0777)
339 dstfile, err := os.Create(dst)
343 _, err = io.Copy(dstfile, srcfile)
348 err = srcfile.Close()
349 err2 := dstfile.Close()
362 func (runner *ContainerRunner) SetupMounts() (err error) {
363 err = runner.SetupArvMountPoint("keep")
365 return fmt.Errorf("While creating keep mount temp dir: %v", err)
368 token, err := runner.ContainerToken()
370 return fmt.Errorf("could not get container token: %s", err)
375 arvMountCmd := []string{
379 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
381 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
382 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
385 collectionPaths := []string{}
387 runner.Volumes = make(map[string]struct{})
388 needCertMount := true
389 type copyFile struct {
393 var copyFiles []copyFile
396 for bind := range runner.Container.Mounts {
397 binds = append(binds, bind)
401 for _, bind := range binds {
402 mnt := runner.Container.Mounts[bind]
403 if bind == "stdout" || bind == "stderr" {
404 // Is it a "file" mount kind?
405 if mnt.Kind != "file" {
406 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
409 // Does path start with OutputPath?
410 prefix := runner.Container.OutputPath
411 if !strings.HasSuffix(prefix, "/") {
414 if !strings.HasPrefix(mnt.Path, prefix) {
415 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
420 // Is it a "collection" mount kind?
421 if mnt.Kind != "collection" && mnt.Kind != "json" {
422 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
426 if bind == "/etc/arvados/ca-certificates.crt" {
427 needCertMount = false
430 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
431 if mnt.Kind != "collection" {
432 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
437 case mnt.Kind == "collection" && bind != "stdin":
439 if mnt.UUID != "" && mnt.PortableDataHash != "" {
440 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
444 return fmt.Errorf("Writing to existing collections currently not permitted.")
447 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
448 } else if mnt.PortableDataHash != "" {
449 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
450 return fmt.Errorf("Can never write to a collection specified by portable data hash")
452 idx := strings.Index(mnt.PortableDataHash, "/")
454 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
455 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
456 runner.Container.Mounts[bind] = mnt
458 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
459 if mnt.Path != "" && mnt.Path != "." {
460 if strings.HasPrefix(mnt.Path, "./") {
461 mnt.Path = mnt.Path[2:]
462 } else if strings.HasPrefix(mnt.Path, "/") {
463 mnt.Path = mnt.Path[1:]
465 src += "/" + mnt.Path
468 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
469 arvMountCmd = append(arvMountCmd, "--mount-tmp")
470 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
474 if bind == runner.Container.OutputPath {
475 runner.HostOutputDir = src
476 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
477 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
478 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
480 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
483 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
485 collectionPaths = append(collectionPaths, src)
487 case mnt.Kind == "tmp":
489 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
491 return fmt.Errorf("While creating mount temp dir: %v", err)
493 st, staterr := os.Stat(tmpdir)
495 return fmt.Errorf("While Stat on temp dir: %v", staterr)
497 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
499 return fmt.Errorf("While Chmod temp dir: %v", err)
501 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
502 if bind == runner.Container.OutputPath {
503 runner.HostOutputDir = tmpdir
506 case mnt.Kind == "json":
507 jsondata, err := json.Marshal(mnt.Content)
509 return fmt.Errorf("encoding json data: %v", err)
511 // Create a tempdir with a single file
512 // (instead of just a tempfile): this way we
513 // can ensure the file is world-readable
514 // inside the container, without having to
515 // make it world-readable on the docker host.
516 tmpdir, err := runner.MkTempDir(runner.parentTemp, "json")
518 return fmt.Errorf("creating temp dir: %v", err)
520 tmpfn := filepath.Join(tmpdir, "mountdata.json")
521 err = ioutil.WriteFile(tmpfn, jsondata, 0644)
523 return fmt.Errorf("writing temp file: %v", err)
525 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
527 case mnt.Kind == "git_tree":
528 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
530 return fmt.Errorf("creating temp dir: %v", err)
532 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
536 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
540 if runner.HostOutputDir == "" {
541 return fmt.Errorf("Output path does not correspond to a writable mount point")
544 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
545 for _, certfile := range arvadosclient.CertFiles {
546 _, err := os.Stat(certfile)
548 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
555 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
557 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
559 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
561 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
563 return fmt.Errorf("While trying to start arv-mount: %v", err)
566 for _, p := range collectionPaths {
569 return fmt.Errorf("While checking that input files exist: %v", err)
573 for _, cp := range copyFiles {
574 st, err := os.Stat(cp.src)
576 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
579 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
583 target := path.Join(cp.bind, walkpath[len(cp.src):])
584 if walkinfo.Mode().IsRegular() {
585 copyerr := copyfile(walkpath, target)
589 return os.Chmod(target, walkinfo.Mode()|0777)
590 } else if walkinfo.Mode().IsDir() {
591 mkerr := os.MkdirAll(target, 0777)
595 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
597 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
600 } else if st.Mode().IsRegular() {
601 err = copyfile(cp.src, cp.bind)
603 err = os.Chmod(cp.bind, st.Mode()|0777)
607 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
614 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
615 // Handle docker log protocol
616 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
617 defer close(runner.loggingDone)
619 header := make([]byte, 8)
622 _, err = io.ReadAtLeast(containerReader, header, 8)
629 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
632 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
635 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
640 runner.CrunchLog.Printf("error reading docker logs: %v", err)
643 err = runner.Stdout.Close()
645 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
648 err = runner.Stderr.Close()
650 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
653 if runner.statReporter != nil {
654 runner.statReporter.Stop()
655 err = runner.statLogger.Close()
657 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
662 func (runner *ContainerRunner) stopHoststat() error {
663 if runner.hoststatReporter == nil {
666 runner.hoststatReporter.Stop()
667 err := runner.hoststatLogger.Close()
669 return fmt.Errorf("error closing hoststat logs: %v", err)
674 func (runner *ContainerRunner) startHoststat() {
675 runner.hoststatLogger = NewThrottledLogger(runner.NewLogWriter("hoststat"))
676 runner.hoststatReporter = &crunchstat.Reporter{
677 Logger: log.New(runner.hoststatLogger, "", 0),
678 CgroupRoot: runner.cgroupRoot,
679 PollPeriod: runner.statInterval,
681 runner.hoststatReporter.Start()
684 func (runner *ContainerRunner) startCrunchstat() {
685 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
686 runner.statReporter = &crunchstat.Reporter{
687 CID: runner.ContainerID,
688 Logger: log.New(runner.statLogger, "", 0),
689 CgroupParent: runner.expectCgroupParent,
690 CgroupRoot: runner.cgroupRoot,
691 PollPeriod: runner.statInterval,
693 runner.statReporter.Start()
696 type infoCommand struct {
701 // LogHostInfo logs info about the current host, for debugging and
702 // accounting purposes. Although it's logged as "node-info", this is
703 // about the environment where crunch-run is actually running, which
704 // might differ from what's described in the node record (see
706 func (runner *ContainerRunner) LogHostInfo() (err error) {
707 w := runner.NewLogWriter("node-info")
709 commands := []infoCommand{
711 label: "Host Information",
712 cmd: []string{"uname", "-a"},
715 label: "CPU Information",
716 cmd: []string{"cat", "/proc/cpuinfo"},
719 label: "Memory Information",
720 cmd: []string{"cat", "/proc/meminfo"},
724 cmd: []string{"df", "-m", "/", os.TempDir()},
727 label: "Disk INodes",
728 cmd: []string{"df", "-i", "/", os.TempDir()},
732 // Run commands with informational output to be logged.
733 for _, command := range commands {
734 fmt.Fprintln(w, command.label)
735 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
738 if err := cmd.Run(); err != nil {
739 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
748 return fmt.Errorf("While closing node-info logs: %v", err)
753 // LogContainerRecord gets and saves the raw JSON container record from the API server
754 func (runner *ContainerRunner) LogContainerRecord() error {
755 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
756 if !logged && err == nil {
757 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
762 // LogNodeRecord logs arvados#node record corresponding to the current host.
763 func (runner *ContainerRunner) LogNodeRecord() error {
764 hostname := os.Getenv("SLURMD_NODENAME")
766 hostname, _ = os.Hostname()
768 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
769 // The "info" field has admin-only info when obtained
770 // with a privileged token, and should not be logged.
771 node, ok := resp.(map[string]interface{})
779 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
781 ArvClient: runner.ArvClient,
782 UUID: runner.Container.UUID,
783 loggingStream: label,
784 writeCloser: runner.LogCollection.Open(label + ".json"),
787 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
789 return false, fmt.Errorf("error getting %s record: %v", label, err)
793 dec := json.NewDecoder(reader)
795 var resp map[string]interface{}
796 if err = dec.Decode(&resp); err != nil {
797 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
799 items, ok := resp["items"].([]interface{})
801 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
802 } else if len(items) < 1 {
808 // Re-encode it using indentation to improve readability
809 enc := json.NewEncoder(w)
810 enc.SetIndent("", " ")
811 if err = enc.Encode(items[0]); err != nil {
812 return false, fmt.Errorf("error logging %s record: %v", label, err)
816 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
821 // AttachStreams connects the docker container stdin, stdout and stderr logs
822 // to the Arvados logger which logs to Keep and the API server logs table.
823 func (runner *ContainerRunner) AttachStreams() (err error) {
825 runner.CrunchLog.Print("Attaching container streams")
827 // If stdin mount is provided, attach it to the docker container
828 var stdinRdr arvados.File
830 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
831 if stdinMnt.Kind == "collection" {
832 var stdinColl arvados.Collection
833 collId := stdinMnt.UUID
835 collId = stdinMnt.PortableDataHash
837 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
839 return fmt.Errorf("While getting stding collection: %v", err)
842 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
843 if os.IsNotExist(err) {
844 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
845 } else if err != nil {
846 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
848 } else if stdinMnt.Kind == "json" {
849 stdinJson, err = json.Marshal(stdinMnt.Content)
851 return fmt.Errorf("While encoding stdin json data: %v", err)
856 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
857 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
858 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
860 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
863 runner.loggingDone = make(chan bool)
865 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
866 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
870 runner.Stdout = stdoutFile
872 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
875 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
876 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
880 runner.Stderr = stderrFile
882 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
887 _, err := io.Copy(response.Conn, stdinRdr)
889 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
893 response.CloseWrite()
895 } else if len(stdinJson) != 0 {
897 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
899 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
902 response.CloseWrite()
906 go runner.ProcessDockerAttach(response.Reader)
911 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
912 stdoutPath := mntPath[len(runner.Container.OutputPath):]
913 index := strings.LastIndex(stdoutPath, "/")
915 subdirs := stdoutPath[:index]
917 st, err := os.Stat(runner.HostOutputDir)
919 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
921 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
922 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
924 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
928 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
930 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
933 return stdoutFile, nil
936 // CreateContainer creates the docker container.
937 func (runner *ContainerRunner) CreateContainer() error {
938 runner.CrunchLog.Print("Creating Docker container")
940 runner.ContainerConfig.Cmd = runner.Container.Command
941 if runner.Container.Cwd != "." {
942 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
945 for k, v := range runner.Container.Environment {
946 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
949 runner.ContainerConfig.Volumes = runner.Volumes
951 runner.HostConfig = dockercontainer.HostConfig{
953 LogConfig: dockercontainer.LogConfig{
956 Resources: dockercontainer.Resources{
957 CgroupParent: runner.setCgroupParent,
961 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
962 tok, err := runner.ContainerToken()
966 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
967 "ARVADOS_API_TOKEN="+tok,
968 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
969 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
971 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
973 if runner.enableNetwork == "always" {
974 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
976 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
980 _, stdinUsed := runner.Container.Mounts["stdin"]
981 runner.ContainerConfig.OpenStdin = stdinUsed
982 runner.ContainerConfig.StdinOnce = stdinUsed
983 runner.ContainerConfig.AttachStdin = stdinUsed
984 runner.ContainerConfig.AttachStdout = true
985 runner.ContainerConfig.AttachStderr = true
987 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
989 return fmt.Errorf("While creating container: %v", err)
992 runner.ContainerID = createdBody.ID
994 return runner.AttachStreams()
997 // StartContainer starts the docker container created by CreateContainer.
998 func (runner *ContainerRunner) StartContainer() error {
999 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
1000 runner.cStateLock.Lock()
1001 defer runner.cStateLock.Unlock()
1002 if runner.cCancelled {
1005 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
1006 dockertypes.ContainerStartOptions{})
1009 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1010 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])
1012 return fmt.Errorf("could not start container: %v%s", err, advice)
1017 // WaitFinish waits for the container to terminate, capture the exit code, and
1018 // close the stdout/stderr logging.
1019 func (runner *ContainerRunner) WaitFinish() error {
1020 runner.CrunchLog.Print("Waiting for container to finish")
1022 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1023 arvMountExit := runner.ArvMountExit
1026 case waitBody := <-waitOk:
1027 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1028 code := int(waitBody.StatusCode)
1029 runner.ExitCode = &code
1031 // wait for stdout/stderr to complete
1032 <-runner.loggingDone
1035 case err := <-waitErr:
1036 return fmt.Errorf("container wait: %v", err)
1038 case <-arvMountExit:
1039 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1041 // arvMountExit will always be ready now that
1042 // it's closed, but that doesn't interest us.
1048 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
1050 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
1051 // Follow symlinks if necessary
1056 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
1057 if followed >= limitFollowSymlinks {
1058 // Got stuck in a loop or just a pathological number of links, give up.
1059 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1063 readlinktgt, err = os.Readlink(nextlink)
1069 if !strings.HasPrefix(tgt, "/") {
1070 // Relative symlink, resolve it to host path
1071 tgt = filepath.Join(filepath.Dir(path), tgt)
1073 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1074 // Absolute symlink to container output path, adjust it to host output path.
1075 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1077 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1078 // After dereferencing, symlink target must either be
1079 // within output directory, or must point to a
1080 // collection mount.
1081 err = ErrNotInOutputDir
1085 info, err = os.Lstat(tgt)
1088 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1089 path[len(runner.HostOutputDir):], readlinktgt, err)
1099 var limitFollowSymlinks = 10
1101 // UploadFile uploads files within the output directory, with special handling
1102 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1103 // from the keep mount into the output manifestText. Ensure that whether
1104 // symlinks are relative or absolute, every symlink target (even targets that
1105 // are symlinks themselves) must point to a path in either the output directory
1106 // or a collection mount.
1108 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1109 func (runner *ContainerRunner) UploadOutputFile(
1114 walkUpload *WalkUpload,
1115 relocateFrom string,
1117 followed int) (manifestText string, err error) {
1123 if info.Mode().IsDir() {
1124 // if empty, need to create a .keep file
1125 dir, direrr := os.Open(path)
1130 names, eof := dir.Readdirnames(1)
1131 if len(names) == 0 && eof == io.EOF && path != runner.HostOutputDir {
1132 containerPath := runner.OutputPath + path[len(runner.HostOutputDir):]
1133 for _, bind := range binds {
1134 mnt := runner.Container.Mounts[bind]
1135 // Check if there is a bind for this
1136 // directory, in which case assume we don't need .keep
1137 if (containerPath == bind || strings.HasPrefix(containerPath, bind+"/")) && mnt.PortableDataHash != "d41d8cd98f00b204e9800998ecf8427e+0" {
1141 outputSuffix := path[len(runner.HostOutputDir)+1:]
1142 return fmt.Sprintf("./%v d41d8cd98f00b204e9800998ecf8427e+0 0:0:.keep\n", outputSuffix), nil
1147 if followed >= limitFollowSymlinks {
1148 // Got stuck in a loop or just a pathological number of
1149 // directory links, give up.
1150 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1154 // "path" is the actual path we are visiting
1155 // "tgt" is the target of "path" (a non-symlink) after following symlinks
1156 // "relocated" is the path in the output manifest where the file should be placed,
1157 // but has HostOutputDir as a prefix.
1159 // The destination path in the output manifest may need to be
1160 // logically relocated to some other path in order to appear
1161 // in the correct location as a result of following a symlink.
1162 // Remove the relocateFrom prefix and replace it with
1164 relocated := relocateTo + path[len(relocateFrom):]
1166 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1167 if derefErr != nil && derefErr != ErrNotInOutputDir {
1171 // go through mounts and try reverse map to collection reference
1172 for _, bind := range binds {
1173 mnt := runner.Container.Mounts[bind]
1174 if (tgt == bind || strings.HasPrefix(tgt, bind+"/")) && !mnt.Writable {
1175 // get path relative to bind
1176 targetSuffix := tgt[len(bind):]
1178 // Copy mount and adjust the path to add path relative to the bind
1179 adjustedMount := mnt
1180 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1182 // Terminates in this keep mount, so add the
1183 // manifest text at appropriate location.
1184 outputSuffix := relocated[len(runner.HostOutputDir):]
1185 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1190 // If target is not a collection mount, it must be located within the
1191 // output directory, otherwise it is an error.
1192 if derefErr == ErrNotInOutputDir {
1193 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1194 path[len(runner.HostOutputDir):], readlinktgt)
1198 if info.Mode().IsRegular() {
1199 return "", walkUpload.UploadFile(relocated, tgt)
1202 if info.Mode().IsDir() {
1203 // Symlink leads to directory. Walk() doesn't follow
1204 // directory symlinks, so we walk the target directory
1205 // instead. Within the walk, file paths are relocated
1206 // so they appear under the original symlink path.
1207 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1209 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1210 binds, walkUpload, tgt, relocated, followed+1)
1212 manifestText = manifestText + m
1222 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1223 func (runner *ContainerRunner) CaptureOutput() error {
1224 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1225 // Output may have been set directly by the container, so
1226 // refresh the container record to check.
1227 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1228 nil, &runner.Container)
1232 if runner.Container.Output != "" {
1233 // Container output is already set.
1234 runner.OutputPDH = &runner.Container.Output
1239 if runner.HostOutputDir == "" {
1243 _, err := os.Stat(runner.HostOutputDir)
1245 return fmt.Errorf("While checking host output path: %v", err)
1248 // Pre-populate output from the configured mount points
1250 for bind, mnt := range runner.Container.Mounts {
1251 if mnt.Kind == "collection" {
1252 binds = append(binds, bind)
1257 var manifestText string
1259 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1260 _, err = os.Stat(collectionMetafile)
1262 // Regular directory
1264 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1265 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1268 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1269 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1271 manifestText = manifestText + m
1276 cw.EndUpload(walkUpload)
1279 return fmt.Errorf("While uploading output files: %v", err)
1282 m, err = cw.ManifestText()
1283 manifestText = manifestText + m
1285 return fmt.Errorf("While uploading output files: %v", err)
1288 // FUSE mount directory
1289 file, openerr := os.Open(collectionMetafile)
1291 return fmt.Errorf("While opening FUSE metafile: %v", err)
1295 var rec arvados.Collection
1296 err = json.NewDecoder(file).Decode(&rec)
1298 return fmt.Errorf("While reading FUSE metafile: %v", err)
1300 manifestText = rec.ManifestText
1303 for _, bind := range binds {
1304 mnt := runner.Container.Mounts[bind]
1306 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1308 if bindSuffix == bind || len(bindSuffix) <= 0 {
1309 // either does not start with OutputPath or is OutputPath itself
1313 if mnt.ExcludeFromOutput == true || mnt.Writable {
1317 // append to manifest_text
1318 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1323 manifestText = manifestText + m
1327 var response arvados.Collection
1328 manifest := manifest.Manifest{Text: manifestText}
1329 manifestText = manifest.Extract(".", ".").Text
1330 err = runner.ArvClient.Create("collections",
1332 "ensure_unique_name": true,
1333 "collection": arvadosclient.Dict{
1335 "name": "output for " + runner.Container.UUID,
1336 "manifest_text": manifestText}},
1339 return fmt.Errorf("While creating output collection: %v", err)
1341 runner.OutputPDH = &response.PortableDataHash
1345 var outputCollections = make(map[string]arvados.Collection)
1347 // Fetch the collection for the mnt.PortableDataHash
1348 // Return the manifest_text fragment corresponding to the specified mnt.Path
1349 // after making any required updates.
1351 // If mnt.Path is not specified,
1352 // return the entire manifest_text after replacing any "." with bindSuffix
1353 // If mnt.Path corresponds to one stream,
1354 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1355 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1356 // for that stream after replacing stream name with bindSuffix minus the last word
1357 // and the file name with last word of the bindSuffix
1358 // Allowed path examples:
1360 // "path":"/subdir1"
1361 // "path":"/subdir1/subdir2"
1362 // "path":"/subdir/filename" etc
1363 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1364 collection := outputCollections[mnt.PortableDataHash]
1365 if collection.PortableDataHash == "" {
1366 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1368 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1370 outputCollections[mnt.PortableDataHash] = collection
1373 if collection.ManifestText == "" {
1374 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1378 mft := manifest.Manifest{Text: collection.ManifestText}
1379 extracted := mft.Extract(mnt.Path, bindSuffix)
1380 if extracted.Err != nil {
1381 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1383 return extracted.Text, nil
1386 func (runner *ContainerRunner) CleanupDirs() {
1387 if runner.ArvMount != nil {
1389 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1390 umount.Stdout = runner.CrunchLog
1391 umount.Stderr = runner.CrunchLog
1392 runner.CrunchLog.Printf("Running %v", umount.Args)
1393 umnterr := umount.Start()
1396 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1398 // If arv-mount --unmount gets stuck for any reason, we
1399 // don't want to wait for it forever. Do Wait() in a goroutine
1400 // so it doesn't block crunch-run.
1401 umountExit := make(chan error)
1403 mnterr := umount.Wait()
1405 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1407 umountExit <- mnterr
1410 for again := true; again; {
1416 case <-runner.ArvMountExit:
1418 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1419 runner.CrunchLog.Printf("Timed out waiting for unmount")
1421 umount.Process.Kill()
1423 runner.ArvMount.Process.Kill()
1429 if runner.ArvMountPoint != "" {
1430 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1431 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1435 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1436 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1440 // CommitLogs posts the collection containing the final container logs.
1441 func (runner *ContainerRunner) CommitLogs() error {
1443 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1444 runner.cStateLock.Lock()
1445 defer runner.cStateLock.Unlock()
1447 runner.CrunchLog.Print(runner.finalState)
1449 if runner.arvMountLog != nil {
1450 runner.arvMountLog.Close()
1452 runner.CrunchLog.Close()
1454 // Closing CrunchLog above allows them to be committed to Keep at this
1455 // point, but re-open crunch log with ArvClient in case there are any
1456 // other further errors (such as failing to write the log to Keep!)
1457 // while shutting down
1458 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1459 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1460 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1463 if runner.LogsPDH != nil {
1464 // If we have already assigned something to LogsPDH,
1465 // we must be closing the re-opened log, which won't
1466 // end up getting attached to the container record and
1467 // therefore doesn't need to be saved as a collection
1468 // -- it exists only to send logs to other channels.
1472 mt, err := runner.LogCollection.ManifestText()
1474 return fmt.Errorf("While creating log manifest: %v", err)
1477 var response arvados.Collection
1478 err = runner.ArvClient.Create("collections",
1480 "ensure_unique_name": true,
1481 "collection": arvadosclient.Dict{
1483 "name": "logs for " + runner.Container.UUID,
1484 "manifest_text": mt}},
1487 return fmt.Errorf("While creating log collection: %v", err)
1489 runner.LogsPDH = &response.PortableDataHash
1493 // UpdateContainerRunning updates the container state to "Running"
1494 func (runner *ContainerRunner) UpdateContainerRunning() error {
1495 runner.cStateLock.Lock()
1496 defer runner.cStateLock.Unlock()
1497 if runner.cCancelled {
1500 return runner.ArvClient.Update("containers", runner.Container.UUID,
1501 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1504 // ContainerToken returns the api_token the container (and any
1505 // arv-mount processes) are allowed to use.
1506 func (runner *ContainerRunner) ContainerToken() (string, error) {
1507 if runner.token != "" {
1508 return runner.token, nil
1511 var auth arvados.APIClientAuthorization
1512 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1516 runner.token = auth.APIToken
1517 return runner.token, nil
1520 // UpdateContainerComplete updates the container record state on API
1521 // server to "Complete" or "Cancelled"
1522 func (runner *ContainerRunner) UpdateContainerFinal() error {
1523 update := arvadosclient.Dict{}
1524 update["state"] = runner.finalState
1525 if runner.LogsPDH != nil {
1526 update["log"] = *runner.LogsPDH
1528 if runner.finalState == "Complete" {
1529 if runner.ExitCode != nil {
1530 update["exit_code"] = *runner.ExitCode
1532 if runner.OutputPDH != nil {
1533 update["output"] = *runner.OutputPDH
1536 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1539 // IsCancelled returns the value of Cancelled, with goroutine safety.
1540 func (runner *ContainerRunner) IsCancelled() bool {
1541 runner.cStateLock.Lock()
1542 defer runner.cStateLock.Unlock()
1543 return runner.cCancelled
1546 // NewArvLogWriter creates an ArvLogWriter
1547 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1548 return &ArvLogWriter{
1549 ArvClient: runner.ArvClient,
1550 UUID: runner.Container.UUID,
1551 loggingStream: name,
1552 writeCloser: runner.LogCollection.Open(name + ".txt")}
1555 // Run the full container lifecycle.
1556 func (runner *ContainerRunner) Run() (err error) {
1557 runner.CrunchLog.Printf("crunch-run %s started", version)
1558 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1560 hostname, hosterr := os.Hostname()
1562 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1564 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1567 runner.finalState = "Queued"
1570 runner.CleanupDirs()
1572 runner.CrunchLog.Printf("crunch-run finished")
1573 runner.CrunchLog.Close()
1577 // checkErr prints e (unless it's nil) and sets err to
1578 // e (unless err is already non-nil). Thus, if err
1579 // hasn't already been assigned when Run() returns,
1580 // this cleanup func will cause Run() to return the
1581 // first non-nil error that is passed to checkErr().
1582 checkErr := func(e error) {
1586 runner.CrunchLog.Print(e)
1590 if runner.finalState == "Complete" {
1591 // There was an error in the finalization.
1592 runner.finalState = "Cancelled"
1596 // Log the error encountered in Run(), if any
1599 if runner.finalState == "Queued" {
1600 runner.UpdateContainerFinal()
1604 if runner.IsCancelled() {
1605 runner.finalState = "Cancelled"
1606 // but don't return yet -- we still want to
1607 // capture partial output and write logs
1610 checkErr(runner.CaptureOutput())
1611 checkErr(runner.stopHoststat())
1612 checkErr(runner.CommitLogs())
1613 checkErr(runner.UpdateContainerFinal())
1616 err = runner.fetchContainerRecord()
1620 runner.setupSignals()
1621 runner.startHoststat()
1623 // check for and/or load image
1624 err = runner.LoadImage()
1626 if !runner.checkBrokenNode(err) {
1627 // Failed to load image but not due to a "broken node"
1628 // condition, probably user error.
1629 runner.finalState = "Cancelled"
1631 err = fmt.Errorf("While loading container image: %v", err)
1635 // set up FUSE mount and binds
1636 err = runner.SetupMounts()
1638 runner.finalState = "Cancelled"
1639 err = fmt.Errorf("While setting up mounts: %v", err)
1643 err = runner.CreateContainer()
1647 err = runner.LogHostInfo()
1651 err = runner.LogNodeRecord()
1655 err = runner.LogContainerRecord()
1660 if runner.IsCancelled() {
1664 err = runner.UpdateContainerRunning()
1668 runner.finalState = "Cancelled"
1670 runner.startCrunchstat()
1672 err = runner.StartContainer()
1674 runner.checkBrokenNode(err)
1678 err = runner.WaitFinish()
1679 if err == nil && !runner.IsCancelled() {
1680 runner.finalState = "Complete"
1685 // Fetch the current container record (uuid = runner.Container.UUID)
1686 // into runner.Container.
1687 func (runner *ContainerRunner) fetchContainerRecord() error {
1688 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1690 return fmt.Errorf("error fetching container record: %v", err)
1692 defer reader.Close()
1694 dec := json.NewDecoder(reader)
1696 err = dec.Decode(&runner.Container)
1698 return fmt.Errorf("error decoding container record: %v", err)
1703 // NewContainerRunner creates a new container runner.
1704 func NewContainerRunner(api IArvadosClient,
1706 docker ThinDockerClient,
1707 containerUUID string) *ContainerRunner {
1709 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1710 cr.NewLogWriter = cr.NewArvLogWriter
1711 cr.RunArvMount = cr.ArvMountCmd
1712 cr.MkTempDir = ioutil.TempDir
1713 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1714 cr.Container.UUID = containerUUID
1715 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1716 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1718 loadLogThrottleParams(api)
1724 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1725 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1726 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1727 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1728 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1729 enableNetwork := flag.String("container-enable-networking", "default",
1730 `Specify if networking should be enabled for container. One of 'default', 'always':
1731 default: only enable networking if container requests it.
1732 always: containers always have networking enabled
1734 networkMode := flag.String("container-network-mode", "default",
1735 `Set networking mode for container. Corresponds to Docker network mode (--net).
1737 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1738 getVersion := flag.Bool("version", false, "Print version information and exit.")
1741 // Print version information if requested
1743 fmt.Printf("crunch-run %s\n", version)
1747 log.Printf("crunch-run %s started", version)
1749 containerId := flag.Arg(0)
1751 if *caCertsPath != "" {
1752 arvadosclient.CertFiles = []string{*caCertsPath}
1755 api, err := arvadosclient.MakeArvadosClient()
1757 log.Fatalf("%s: %v", containerId, err)
1761 kc, kcerr := keepclient.MakeKeepClient(api)
1763 log.Fatalf("%s: %v", containerId, kcerr)
1765 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1768 // API version 1.21 corresponds to Docker 1.9, which is currently the
1769 // minimum version we want to support.
1770 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1772 cr := NewContainerRunner(api, kc, docker, containerId)
1773 if dockererr != nil {
1774 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1775 cr.checkBrokenNode(dockererr)
1776 cr.CrunchLog.Close()
1780 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerId+".")
1782 log.Fatalf("%s: %v", containerId, tmperr)
1785 cr.parentTemp = parentTemp
1786 cr.statInterval = *statInterval
1787 cr.cgroupRoot = *cgroupRoot
1788 cr.expectCgroupParent = *cgroupParent
1789 cr.enableNetwork = *enableNetwork
1790 cr.networkMode = *networkMode
1791 if *cgroupParentSubsystem != "" {
1792 p := findCgroup(*cgroupParentSubsystem)
1793 cr.setCgroupParent = p
1794 cr.expectCgroupParent = p
1799 if *memprofile != "" {
1800 f, err := os.Create(*memprofile)
1802 log.Printf("could not create memory profile: ", err)
1804 runtime.GC() // get up-to-date statistics
1805 if err := pprof.WriteHeapProfile(f); err != nil {
1806 log.Printf("could not write memory profile: ", err)
1808 closeerr := f.Close()
1809 if closeerr != nil {
1810 log.Printf("closing memprofile file: ", err)
1815 log.Fatalf("%s: %v", containerId, runerr)