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
109 CleanupTempDir []string
111 Volumes map[string]struct{}
113 SigChan chan os.Signal
114 ArvMountExit chan error
118 statLogger io.WriteCloser
119 statReporter *crunchstat.Reporter
120 hoststatLogger io.WriteCloser
121 hoststatReporter *crunchstat.Reporter
122 statInterval time.Duration
124 // What we expect the container's cgroup parent to be.
125 expectCgroupParent string
126 // What we tell docker to use as the container's cgroup
127 // parent. Note: Ideally we would use the same field for both
128 // expectCgroupParent and setCgroupParent, and just make it
129 // default to "docker". However, when using docker < 1.10 with
130 // systemd, specifying a non-empty cgroup parent (even the
131 // default value "docker") hits a docker bug
132 // (https://github.com/docker/docker/issues/17126). Using two
133 // separate fields makes it possible to use the "expect cgroup
134 // parent to be X" feature even on sites where the "specify
135 // cgroup parent" feature breaks.
136 setCgroupParent string
138 cStateLock sync.Mutex
139 cCancelled bool // StopContainer() invoked
141 enableNetwork string // one of "default" or "always"
142 networkMode string // passed through to HostConfig.NetworkMode
143 arvMountLog *ThrottledLogger
146 // setupSignals sets up signal handling to gracefully terminate the underlying
147 // Docker container and update state when receiving a TERM, INT or QUIT signal.
148 func (runner *ContainerRunner) setupSignals() {
149 runner.SigChan = make(chan os.Signal, 1)
150 signal.Notify(runner.SigChan, syscall.SIGTERM)
151 signal.Notify(runner.SigChan, syscall.SIGINT)
152 signal.Notify(runner.SigChan, syscall.SIGQUIT)
154 go func(sig chan os.Signal) {
156 runner.CrunchLog.Printf("caught signal: %v", s)
162 // stop the underlying Docker container.
163 func (runner *ContainerRunner) stop() {
164 runner.cStateLock.Lock()
165 defer runner.cStateLock.Unlock()
166 if runner.ContainerID == "" {
169 runner.cCancelled = true
170 runner.CrunchLog.Printf("removing container")
171 err := runner.Docker.ContainerRemove(context.TODO(), runner.ContainerID, dockertypes.ContainerRemoveOptions{Force: true})
173 runner.CrunchLog.Printf("error removing container: %s", err)
177 func (runner *ContainerRunner) stopSignals() {
178 if runner.SigChan != nil {
179 signal.Stop(runner.SigChan)
183 var errorBlacklist = []string{
184 "(?ms).*[Cc]annot connect to the Docker daemon.*",
185 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
187 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)")
189 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
190 for _, d := range errorBlacklist {
191 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
192 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
193 if *brokenNodeHook == "" {
194 runner.CrunchLog.Printf("No broken node hook provided, cannot mark node as broken.")
196 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
198 c := exec.Command(*brokenNodeHook)
199 c.Stdout = runner.CrunchLog
200 c.Stderr = runner.CrunchLog
203 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
212 // LoadImage determines the docker image id from the container record and
213 // checks if it is available in the local Docker image store. If not, it loads
214 // the image from Keep.
215 func (runner *ContainerRunner) LoadImage() (err error) {
217 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
219 var collection arvados.Collection
220 err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
222 return fmt.Errorf("While getting container image collection: %v", err)
224 manifest := manifest.Manifest{Text: collection.ManifestText}
225 var img, imageID string
226 for ms := range manifest.StreamIter() {
227 img = ms.FileStreamSegments[0].Name
228 if !strings.HasSuffix(img, ".tar") {
229 return fmt.Errorf("First file in the container image collection does not end in .tar")
231 imageID = img[:len(img)-4]
234 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
236 _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
238 runner.CrunchLog.Print("Loading Docker image from keep")
240 var readCloser io.ReadCloser
241 readCloser, err = runner.Kc.ManifestFileReader(manifest, img)
243 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
246 response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, true)
248 return fmt.Errorf("While loading container image into Docker: %v", err)
251 defer response.Body.Close()
252 rbody, err := ioutil.ReadAll(response.Body)
254 return fmt.Errorf("Reading response to image load: %v", err)
256 runner.CrunchLog.Printf("Docker response: %s", rbody)
258 runner.CrunchLog.Print("Docker image is available")
261 runner.ContainerConfig.Image = imageID
263 runner.Kc.ClearBlockCache()
268 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
269 c = exec.Command("arv-mount", arvMountCmd...)
271 // Copy our environment, but override ARVADOS_API_TOKEN with
272 // the container auth token.
274 for _, s := range os.Environ() {
275 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
276 c.Env = append(c.Env, s)
279 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
281 runner.arvMountLog = NewThrottledLogger(runner.NewLogWriter("arv-mount"))
282 c.Stdout = runner.arvMountLog
283 c.Stderr = runner.arvMountLog
285 runner.CrunchLog.Printf("Running %v", c.Args)
292 statReadme := make(chan bool)
293 runner.ArvMountExit = make(chan error)
298 time.Sleep(100 * time.Millisecond)
299 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
311 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
313 runner.ArvMountExit <- mnterr
314 close(runner.ArvMountExit)
320 case err := <-runner.ArvMountExit:
321 runner.ArvMount = nil
329 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
330 if runner.ArvMountPoint == "" {
331 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
336 func copyfile(src string, dst string) (err error) {
337 srcfile, err := os.Open(src)
342 os.MkdirAll(path.Dir(dst), 0777)
344 dstfile, err := os.Create(dst)
348 _, err = io.Copy(dstfile, srcfile)
353 err = srcfile.Close()
354 err2 := dstfile.Close()
367 func (runner *ContainerRunner) SetupMounts() (err error) {
368 err = runner.SetupArvMountPoint("keep")
370 return fmt.Errorf("While creating keep mount temp dir: %v", err)
373 token, err := runner.ContainerToken()
375 return fmt.Errorf("could not get container token: %s", err)
380 arvMountCmd := []string{
384 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
386 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
387 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
390 collectionPaths := []string{}
392 runner.Volumes = make(map[string]struct{})
393 needCertMount := true
394 type copyFile struct {
398 var copyFiles []copyFile
401 for bind := range runner.Container.Mounts {
402 binds = append(binds, bind)
406 for _, bind := range binds {
407 mnt := runner.Container.Mounts[bind]
408 if bind == "stdout" || bind == "stderr" {
409 // Is it a "file" mount kind?
410 if mnt.Kind != "file" {
411 return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
414 // Does path start with OutputPath?
415 prefix := runner.Container.OutputPath
416 if !strings.HasSuffix(prefix, "/") {
419 if !strings.HasPrefix(mnt.Path, prefix) {
420 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
425 // Is it a "collection" mount kind?
426 if mnt.Kind != "collection" && mnt.Kind != "json" {
427 return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
431 if bind == "/etc/arvados/ca-certificates.crt" {
432 needCertMount = false
435 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
436 if mnt.Kind != "collection" {
437 return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind)
442 case mnt.Kind == "collection" && bind != "stdin":
444 if mnt.UUID != "" && mnt.PortableDataHash != "" {
445 return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
449 return fmt.Errorf("Writing to existing collections currently not permitted.")
452 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
453 } else if mnt.PortableDataHash != "" {
454 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
455 return fmt.Errorf("Can never write to a collection specified by portable data hash")
457 idx := strings.Index(mnt.PortableDataHash, "/")
459 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
460 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
461 runner.Container.Mounts[bind] = mnt
463 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
464 if mnt.Path != "" && mnt.Path != "." {
465 if strings.HasPrefix(mnt.Path, "./") {
466 mnt.Path = mnt.Path[2:]
467 } else if strings.HasPrefix(mnt.Path, "/") {
468 mnt.Path = mnt.Path[1:]
470 src += "/" + mnt.Path
473 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
474 arvMountCmd = append(arvMountCmd, "--mount-tmp")
475 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
479 if bind == runner.Container.OutputPath {
480 runner.HostOutputDir = src
481 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
482 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
483 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
485 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
488 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
490 collectionPaths = append(collectionPaths, src)
492 case mnt.Kind == "tmp":
494 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
496 return fmt.Errorf("While creating mount temp dir: %v", err)
498 st, staterr := os.Stat(tmpdir)
500 return fmt.Errorf("While Stat on temp dir: %v", staterr)
502 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
504 return fmt.Errorf("While Chmod temp dir: %v", err)
506 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
507 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
508 if bind == runner.Container.OutputPath {
509 runner.HostOutputDir = tmpdir
512 case mnt.Kind == "json":
513 jsondata, err := json.Marshal(mnt.Content)
515 return fmt.Errorf("encoding json data: %v", err)
517 // Create a tempdir with a single file
518 // (instead of just a tempfile): this way we
519 // can ensure the file is world-readable
520 // inside the container, without having to
521 // make it world-readable on the docker host.
522 tmpdir, err := runner.MkTempDir(runner.parentTemp, "json")
524 return fmt.Errorf("creating temp dir: %v", err)
526 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
527 tmpfn := filepath.Join(tmpdir, "mountdata.json")
528 err = ioutil.WriteFile(tmpfn, jsondata, 0644)
530 return fmt.Errorf("writing temp file: %v", err)
532 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
534 case mnt.Kind == "git_tree":
535 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
537 return fmt.Errorf("creating temp dir: %v", err)
539 runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir)
540 err = gitMount(mnt).extractTree(runner.ArvClient, tmpdir, token)
544 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
548 if runner.HostOutputDir == "" {
549 return fmt.Errorf("Output path does not correspond to a writable mount point")
552 if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI {
553 for _, certfile := range arvadosclient.CertFiles {
554 _, err := os.Stat(certfile)
556 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
563 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
565 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
567 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
569 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
571 return fmt.Errorf("While trying to start arv-mount: %v", err)
574 for _, p := range collectionPaths {
577 return fmt.Errorf("While checking that input files exist: %v", err)
581 for _, cp := range copyFiles {
582 dir, err := os.Stat(cp.src)
584 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
587 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
591 if walkinfo.Mode().IsRegular() {
592 return copyfile(walkpath, path.Join(cp.bind, walkpath[len(cp.src):]))
593 } else if walkinfo.Mode().IsDir() {
594 return os.MkdirAll(path.Join(cp.bind, walkpath[len(cp.src):]), 0777)
596 return fmt.Errorf("Source %q is not a regular file or directory", cp.src)
600 err = copyfile(cp.src, cp.bind)
603 return fmt.Errorf("While staging writable file from %q to %q: %v", cp.src, cp.bind, err)
610 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
611 // Handle docker log protocol
612 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
613 defer close(runner.loggingDone)
615 header := make([]byte, 8)
618 _, err = io.ReadAtLeast(containerReader, header, 8)
625 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
628 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
631 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
636 runner.CrunchLog.Printf("error reading docker logs: %v", err)
639 err = runner.Stdout.Close()
641 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
644 err = runner.Stderr.Close()
646 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
649 if runner.statReporter != nil {
650 runner.statReporter.Stop()
651 err = runner.statLogger.Close()
653 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
658 func (runner *ContainerRunner) stopHoststat() error {
659 if runner.hoststatReporter == nil {
662 runner.hoststatReporter.Stop()
663 err := runner.hoststatLogger.Close()
665 return fmt.Errorf("error closing hoststat logs: %v", err)
670 func (runner *ContainerRunner) startHoststat() {
671 runner.hoststatLogger = NewThrottledLogger(runner.NewLogWriter("hoststat"))
672 runner.hoststatReporter = &crunchstat.Reporter{
673 Logger: log.New(runner.hoststatLogger, "", 0),
674 CgroupRoot: runner.cgroupRoot,
675 PollPeriod: runner.statInterval,
677 runner.hoststatReporter.Start()
680 func (runner *ContainerRunner) startCrunchstat() {
681 runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat"))
682 runner.statReporter = &crunchstat.Reporter{
683 CID: runner.ContainerID,
684 Logger: log.New(runner.statLogger, "", 0),
685 CgroupParent: runner.expectCgroupParent,
686 CgroupRoot: runner.cgroupRoot,
687 PollPeriod: runner.statInterval,
689 runner.statReporter.Start()
692 type infoCommand struct {
697 // LogHostInfo logs info about the current host, for debugging and
698 // accounting purposes. Although it's logged as "node-info", this is
699 // about the environment where crunch-run is actually running, which
700 // might differ from what's described in the node record (see
702 func (runner *ContainerRunner) LogHostInfo() (err error) {
703 w := runner.NewLogWriter("node-info")
705 commands := []infoCommand{
707 label: "Host Information",
708 cmd: []string{"uname", "-a"},
711 label: "CPU Information",
712 cmd: []string{"cat", "/proc/cpuinfo"},
715 label: "Memory Information",
716 cmd: []string{"cat", "/proc/meminfo"},
720 cmd: []string{"df", "-m", "/", os.TempDir()},
723 label: "Disk INodes",
724 cmd: []string{"df", "-i", "/", os.TempDir()},
728 // Run commands with informational output to be logged.
729 for _, command := range commands {
730 fmt.Fprintln(w, command.label)
731 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
734 if err := cmd.Run(); err != nil {
735 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
744 return fmt.Errorf("While closing node-info logs: %v", err)
749 // LogContainerRecord gets and saves the raw JSON container record from the API server
750 func (runner *ContainerRunner) LogContainerRecord() error {
751 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
752 if !logged && err == nil {
753 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
758 // LogNodeRecord logs arvados#node record corresponding to the current host.
759 func (runner *ContainerRunner) LogNodeRecord() error {
760 hostname := os.Getenv("SLURMD_NODENAME")
762 hostname, _ = os.Hostname()
764 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
765 // The "info" field has admin-only info when obtained
766 // with a privileged token, and should not be logged.
767 node, ok := resp.(map[string]interface{})
775 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
777 ArvClient: runner.ArvClient,
778 UUID: runner.Container.UUID,
779 loggingStream: label,
780 writeCloser: runner.LogCollection.Open(label + ".json"),
783 reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
785 return false, fmt.Errorf("error getting %s record: %v", label, err)
789 dec := json.NewDecoder(reader)
791 var resp map[string]interface{}
792 if err = dec.Decode(&resp); err != nil {
793 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
795 items, ok := resp["items"].([]interface{})
797 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
798 } else if len(items) < 1 {
804 // Re-encode it using indentation to improve readability
805 enc := json.NewEncoder(w)
806 enc.SetIndent("", " ")
807 if err = enc.Encode(items[0]); err != nil {
808 return false, fmt.Errorf("error logging %s record: %v", label, err)
812 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
817 // AttachStreams connects the docker container stdin, stdout and stderr logs
818 // to the Arvados logger which logs to Keep and the API server logs table.
819 func (runner *ContainerRunner) AttachStreams() (err error) {
821 runner.CrunchLog.Print("Attaching container streams")
823 // If stdin mount is provided, attach it to the docker container
824 var stdinRdr arvados.File
826 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
827 if stdinMnt.Kind == "collection" {
828 var stdinColl arvados.Collection
829 collId := stdinMnt.UUID
831 collId = stdinMnt.PortableDataHash
833 err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
835 return fmt.Errorf("While getting stding collection: %v", err)
838 stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
839 if os.IsNotExist(err) {
840 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
841 } else if err != nil {
842 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
844 } else if stdinMnt.Kind == "json" {
845 stdinJson, err = json.Marshal(stdinMnt.Content)
847 return fmt.Errorf("While encoding stdin json data: %v", err)
852 stdinUsed := stdinRdr != nil || len(stdinJson) != 0
853 response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
854 dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
856 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
859 runner.loggingDone = make(chan bool)
861 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
862 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
866 runner.Stdout = stdoutFile
868 runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
871 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
872 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
876 runner.Stderr = stderrFile
878 runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
883 _, err := io.Copy(response.Conn, stdinRdr)
885 runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
889 response.CloseWrite()
891 } else if len(stdinJson) != 0 {
893 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
895 runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
898 response.CloseWrite()
902 go runner.ProcessDockerAttach(response.Reader)
907 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
908 stdoutPath := mntPath[len(runner.Container.OutputPath):]
909 index := strings.LastIndex(stdoutPath, "/")
911 subdirs := stdoutPath[:index]
913 st, err := os.Stat(runner.HostOutputDir)
915 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
917 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
918 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
920 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
924 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
926 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
929 return stdoutFile, nil
932 // CreateContainer creates the docker container.
933 func (runner *ContainerRunner) CreateContainer() error {
934 runner.CrunchLog.Print("Creating Docker container")
936 runner.ContainerConfig.Cmd = runner.Container.Command
937 if runner.Container.Cwd != "." {
938 runner.ContainerConfig.WorkingDir = runner.Container.Cwd
941 for k, v := range runner.Container.Environment {
942 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
945 runner.ContainerConfig.Volumes = runner.Volumes
947 runner.HostConfig = dockercontainer.HostConfig{
949 LogConfig: dockercontainer.LogConfig{
952 Resources: dockercontainer.Resources{
953 CgroupParent: runner.setCgroupParent,
957 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
958 tok, err := runner.ContainerToken()
962 runner.ContainerConfig.Env = append(runner.ContainerConfig.Env,
963 "ARVADOS_API_TOKEN="+tok,
964 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
965 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
967 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
969 if runner.enableNetwork == "always" {
970 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
972 runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
976 _, stdinUsed := runner.Container.Mounts["stdin"]
977 runner.ContainerConfig.OpenStdin = stdinUsed
978 runner.ContainerConfig.StdinOnce = stdinUsed
979 runner.ContainerConfig.AttachStdin = stdinUsed
980 runner.ContainerConfig.AttachStdout = true
981 runner.ContainerConfig.AttachStderr = true
983 createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
985 return fmt.Errorf("While creating container: %v", err)
988 runner.ContainerID = createdBody.ID
990 return runner.AttachStreams()
993 // StartContainer starts the docker container created by CreateContainer.
994 func (runner *ContainerRunner) StartContainer() error {
995 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
996 runner.cStateLock.Lock()
997 defer runner.cStateLock.Unlock()
998 if runner.cCancelled {
1001 err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
1002 dockertypes.ContainerStartOptions{})
1005 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1006 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])
1008 return fmt.Errorf("could not start container: %v%s", err, advice)
1013 // WaitFinish waits for the container to terminate, capture the exit code, and
1014 // close the stdout/stderr logging.
1015 func (runner *ContainerRunner) WaitFinish() error {
1016 runner.CrunchLog.Print("Waiting for container to finish")
1018 waitOk, waitErr := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID, dockercontainer.WaitConditionNotRunning)
1019 arvMountExit := runner.ArvMountExit
1022 case waitBody := <-waitOk:
1023 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1024 code := int(waitBody.StatusCode)
1025 runner.ExitCode = &code
1027 // wait for stdout/stderr to complete
1028 <-runner.loggingDone
1031 case err := <-waitErr:
1032 return fmt.Errorf("container wait: %v", err)
1034 case <-arvMountExit:
1035 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1037 // arvMountExit will always be ready now that
1038 // it's closed, but that doesn't interest us.
1044 var ErrNotInOutputDir = fmt.Errorf("Must point to path within the output directory")
1046 func (runner *ContainerRunner) derefOutputSymlink(path string, startinfo os.FileInfo) (tgt string, readlinktgt string, info os.FileInfo, err error) {
1047 // Follow symlinks if necessary
1052 for followed := 0; info.Mode()&os.ModeSymlink != 0; followed++ {
1053 if followed >= limitFollowSymlinks {
1054 // Got stuck in a loop or just a pathological number of links, give up.
1055 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1059 readlinktgt, err = os.Readlink(nextlink)
1065 if !strings.HasPrefix(tgt, "/") {
1066 // Relative symlink, resolve it to host path
1067 tgt = filepath.Join(filepath.Dir(path), tgt)
1069 if strings.HasPrefix(tgt, runner.Container.OutputPath+"/") && !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1070 // Absolute symlink to container output path, adjust it to host output path.
1071 tgt = filepath.Join(runner.HostOutputDir, tgt[len(runner.Container.OutputPath):])
1073 if !strings.HasPrefix(tgt, runner.HostOutputDir+"/") {
1074 // After dereferencing, symlink target must either be
1075 // within output directory, or must point to a
1076 // collection mount.
1077 err = ErrNotInOutputDir
1081 info, err = os.Lstat(tgt)
1084 err = fmt.Errorf("Symlink in output %q points to invalid location %q: %v",
1085 path[len(runner.HostOutputDir):], readlinktgt, err)
1095 var limitFollowSymlinks = 10
1097 // UploadFile uploads files within the output directory, with special handling
1098 // for symlinks. If the symlink leads to a keep mount, copy the manifest text
1099 // from the keep mount into the output manifestText. Ensure that whether
1100 // symlinks are relative or absolute, every symlink target (even targets that
1101 // are symlinks themselves) must point to a path in either the output directory
1102 // or a collection mount.
1104 // Assumes initial value of "path" is absolute, and located within runner.HostOutputDir.
1105 func (runner *ContainerRunner) UploadOutputFile(
1110 walkUpload *WalkUpload,
1111 relocateFrom string,
1113 followed int) (manifestText string, err error) {
1119 if info.Mode().IsDir() {
1120 // if empty, need to create a .keep file
1121 dir, direrr := os.Open(path)
1126 names, eof := dir.Readdirnames(1)
1127 if len(names) == 0 && eof == io.EOF && path != runner.HostOutputDir {
1128 containerPath := runner.OutputPath + path[len(runner.HostOutputDir):]
1129 for _, bind := range binds {
1130 mnt := runner.Container.Mounts[bind]
1131 // Check if there is a bind for this
1132 // directory, in which case assume we don't need .keep
1133 if (containerPath == bind || strings.HasPrefix(containerPath, bind+"/")) && mnt.PortableDataHash != "d41d8cd98f00b204e9800998ecf8427e+0" {
1137 outputSuffix := path[len(runner.HostOutputDir)+1:]
1138 return fmt.Sprintf("./%v d41d8cd98f00b204e9800998ecf8427e+0 0:0:.keep\n", outputSuffix), nil
1143 if followed >= limitFollowSymlinks {
1144 // Got stuck in a loop or just a pathological number of
1145 // directory links, give up.
1146 err = fmt.Errorf("Followed more than %v symlinks from path %q", limitFollowSymlinks, path)
1150 // "path" is the actual path we are visiting
1151 // "tgt" is the target of "path" (a non-symlink) after following symlinks
1152 // "relocated" is the path in the output manifest where the file should be placed,
1153 // but has HostOutputDir as a prefix.
1155 // The destination path in the output manifest may need to be
1156 // logically relocated to some other path in order to appear
1157 // in the correct location as a result of following a symlink.
1158 // Remove the relocateFrom prefix and replace it with
1160 relocated := relocateTo + path[len(relocateFrom):]
1162 tgt, readlinktgt, info, derefErr := runner.derefOutputSymlink(path, info)
1163 if derefErr != nil && derefErr != ErrNotInOutputDir {
1167 // go through mounts and try reverse map to collection reference
1168 for _, bind := range binds {
1169 mnt := runner.Container.Mounts[bind]
1170 if (tgt == bind || strings.HasPrefix(tgt, bind+"/")) && !mnt.Writable {
1171 // get path relative to bind
1172 targetSuffix := tgt[len(bind):]
1174 // Copy mount and adjust the path to add path relative to the bind
1175 adjustedMount := mnt
1176 adjustedMount.Path = filepath.Join(adjustedMount.Path, targetSuffix)
1178 // Terminates in this keep mount, so add the
1179 // manifest text at appropriate location.
1180 outputSuffix := relocated[len(runner.HostOutputDir):]
1181 manifestText, err = runner.getCollectionManifestForPath(adjustedMount, outputSuffix)
1186 // If target is not a collection mount, it must be located within the
1187 // output directory, otherwise it is an error.
1188 if derefErr == ErrNotInOutputDir {
1189 err = fmt.Errorf("Symlink in output %q points to invalid location %q, must point to path within the output directory.",
1190 path[len(runner.HostOutputDir):], readlinktgt)
1194 if info.Mode().IsRegular() {
1195 return "", walkUpload.UploadFile(relocated, tgt)
1198 if info.Mode().IsDir() {
1199 // Symlink leads to directory. Walk() doesn't follow
1200 // directory symlinks, so we walk the target directory
1201 // instead. Within the walk, file paths are relocated
1202 // so they appear under the original symlink path.
1203 err = filepath.Walk(tgt, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
1205 m, walkerr = runner.UploadOutputFile(walkpath, walkinfo, walkerr,
1206 binds, walkUpload, tgt, relocated, followed+1)
1208 manifestText = manifestText + m
1218 // HandleOutput sets the output, unmounts the FUSE mount, and deletes temporary directories
1219 func (runner *ContainerRunner) CaptureOutput() error {
1220 if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
1221 // Output may have been set directly by the container, so
1222 // refresh the container record to check.
1223 err := runner.ArvClient.Get("containers", runner.Container.UUID,
1224 nil, &runner.Container)
1228 if runner.Container.Output != "" {
1229 // Container output is already set.
1230 runner.OutputPDH = &runner.Container.Output
1235 if runner.HostOutputDir == "" {
1239 _, err := os.Stat(runner.HostOutputDir)
1241 return fmt.Errorf("While checking host output path: %v", err)
1244 // Pre-populate output from the configured mount points
1246 for bind, mnt := range runner.Container.Mounts {
1247 if mnt.Kind == "collection" {
1248 binds = append(binds, bind)
1253 var manifestText string
1255 collectionMetafile := fmt.Sprintf("%s/.arvados#collection", runner.HostOutputDir)
1256 _, err = os.Stat(collectionMetafile)
1258 // Regular directory
1260 cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
1261 walkUpload := cw.BeginUpload(runner.HostOutputDir, runner.CrunchLog.Logger)
1264 err = filepath.Walk(runner.HostOutputDir, func(path string, info os.FileInfo, err error) error {
1265 m, err = runner.UploadOutputFile(path, info, err, binds, walkUpload, "", "", 0)
1267 manifestText = manifestText + m
1272 cw.EndUpload(walkUpload)
1275 return fmt.Errorf("While uploading output files: %v", err)
1278 m, err = cw.ManifestText()
1279 manifestText = manifestText + m
1281 return fmt.Errorf("While uploading output files: %v", err)
1284 // FUSE mount directory
1285 file, openerr := os.Open(collectionMetafile)
1287 return fmt.Errorf("While opening FUSE metafile: %v", err)
1291 var rec arvados.Collection
1292 err = json.NewDecoder(file).Decode(&rec)
1294 return fmt.Errorf("While reading FUSE metafile: %v", err)
1296 manifestText = rec.ManifestText
1299 for _, bind := range binds {
1300 mnt := runner.Container.Mounts[bind]
1302 bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath)
1304 if bindSuffix == bind || len(bindSuffix) <= 0 {
1305 // either does not start with OutputPath or is OutputPath itself
1309 if mnt.ExcludeFromOutput == true || mnt.Writable {
1313 // append to manifest_text
1314 m, err := runner.getCollectionManifestForPath(mnt, bindSuffix)
1319 manifestText = manifestText + m
1323 var response arvados.Collection
1324 manifest := manifest.Manifest{Text: manifestText}
1325 manifestText = manifest.Extract(".", ".").Text
1326 err = runner.ArvClient.Create("collections",
1328 "ensure_unique_name": true,
1329 "collection": arvadosclient.Dict{
1331 "name": "output for " + runner.Container.UUID,
1332 "manifest_text": manifestText}},
1335 return fmt.Errorf("While creating output collection: %v", err)
1337 runner.OutputPDH = &response.PortableDataHash
1341 var outputCollections = make(map[string]arvados.Collection)
1343 // Fetch the collection for the mnt.PortableDataHash
1344 // Return the manifest_text fragment corresponding to the specified mnt.Path
1345 // after making any required updates.
1347 // If mnt.Path is not specified,
1348 // return the entire manifest_text after replacing any "." with bindSuffix
1349 // If mnt.Path corresponds to one stream,
1350 // return the manifest_text for that stream after replacing that stream name with bindSuffix
1351 // Otherwise, check if a filename in any one stream is being sought. Return the manifest_text
1352 // for that stream after replacing stream name with bindSuffix minus the last word
1353 // and the file name with last word of the bindSuffix
1354 // Allowed path examples:
1356 // "path":"/subdir1"
1357 // "path":"/subdir1/subdir2"
1358 // "path":"/subdir/filename" etc
1359 func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) {
1360 collection := outputCollections[mnt.PortableDataHash]
1361 if collection.PortableDataHash == "" {
1362 err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection)
1364 return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err)
1366 outputCollections[mnt.PortableDataHash] = collection
1369 if collection.ManifestText == "" {
1370 runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash)
1374 mft := manifest.Manifest{Text: collection.ManifestText}
1375 extracted := mft.Extract(mnt.Path, bindSuffix)
1376 if extracted.Err != nil {
1377 return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
1379 return extracted.Text, nil
1382 func (runner *ContainerRunner) CleanupDirs() {
1383 if runner.ArvMount != nil {
1385 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1386 umount.Stdout = runner.CrunchLog
1387 umount.Stderr = runner.CrunchLog
1388 runner.CrunchLog.Printf("Running %v", umount.Args)
1389 umnterr := umount.Start()
1392 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1394 // If arv-mount --unmount gets stuck for any reason, we
1395 // don't want to wait for it forever. Do Wait() in a goroutine
1396 // so it doesn't block crunch-run.
1397 umountExit := make(chan error)
1399 mnterr := umount.Wait()
1401 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1403 umountExit <- mnterr
1406 for again := true; again; {
1412 case <-runner.ArvMountExit:
1414 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1415 runner.CrunchLog.Printf("Timed out waiting for unmount")
1417 umount.Process.Kill()
1419 runner.ArvMount.Process.Kill()
1425 if runner.ArvMountPoint != "" {
1426 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1427 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1431 for _, tmpdir := range runner.CleanupTempDir {
1432 if rmerr := os.RemoveAll(tmpdir); rmerr != nil {
1433 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", tmpdir, rmerr)
1437 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1438 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1442 // CommitLogs posts the collection containing the final container logs.
1443 func (runner *ContainerRunner) CommitLogs() error {
1444 runner.CrunchLog.Print(runner.finalState)
1446 if runner.arvMountLog != nil {
1447 runner.arvMountLog.Close()
1449 runner.CrunchLog.Close()
1451 // Closing CrunchLog above allows them to be committed to Keep at this
1452 // point, but re-open crunch log with ArvClient in case there are any
1453 // other further errors (such as failing to write the log to Keep!)
1454 // while shutting down
1455 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
1456 UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
1457 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1459 if runner.LogsPDH != nil {
1460 // If we have already assigned something to LogsPDH,
1461 // we must be closing the re-opened log, which won't
1462 // end up getting attached to the container record and
1463 // therefore doesn't need to be saved as a collection
1464 // -- it exists only to send logs to other channels.
1468 mt, err := runner.LogCollection.ManifestText()
1470 return fmt.Errorf("While creating log manifest: %v", err)
1473 var response arvados.Collection
1474 err = runner.ArvClient.Create("collections",
1476 "ensure_unique_name": true,
1477 "collection": arvadosclient.Dict{
1479 "name": "logs for " + runner.Container.UUID,
1480 "manifest_text": mt}},
1483 return fmt.Errorf("While creating log collection: %v", err)
1485 runner.LogsPDH = &response.PortableDataHash
1489 // UpdateContainerRunning updates the container state to "Running"
1490 func (runner *ContainerRunner) UpdateContainerRunning() error {
1491 runner.cStateLock.Lock()
1492 defer runner.cStateLock.Unlock()
1493 if runner.cCancelled {
1496 return runner.ArvClient.Update("containers", runner.Container.UUID,
1497 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil)
1500 // ContainerToken returns the api_token the container (and any
1501 // arv-mount processes) are allowed to use.
1502 func (runner *ContainerRunner) ContainerToken() (string, error) {
1503 if runner.token != "" {
1504 return runner.token, nil
1507 var auth arvados.APIClientAuthorization
1508 err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1512 runner.token = auth.APIToken
1513 return runner.token, nil
1516 // UpdateContainerComplete updates the container record state on API
1517 // server to "Complete" or "Cancelled"
1518 func (runner *ContainerRunner) UpdateContainerFinal() error {
1519 update := arvadosclient.Dict{}
1520 update["state"] = runner.finalState
1521 if runner.LogsPDH != nil {
1522 update["log"] = *runner.LogsPDH
1524 if runner.finalState == "Complete" {
1525 if runner.ExitCode != nil {
1526 update["exit_code"] = *runner.ExitCode
1528 if runner.OutputPDH != nil {
1529 update["output"] = *runner.OutputPDH
1532 return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1535 // IsCancelled returns the value of Cancelled, with goroutine safety.
1536 func (runner *ContainerRunner) IsCancelled() bool {
1537 runner.cStateLock.Lock()
1538 defer runner.cStateLock.Unlock()
1539 return runner.cCancelled
1542 // NewArvLogWriter creates an ArvLogWriter
1543 func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
1544 return &ArvLogWriter{
1545 ArvClient: runner.ArvClient,
1546 UUID: runner.Container.UUID,
1547 loggingStream: name,
1548 writeCloser: runner.LogCollection.Open(name + ".txt")}
1551 // Run the full container lifecycle.
1552 func (runner *ContainerRunner) Run() (err error) {
1553 runner.CrunchLog.Printf("crunch-run %s started", version)
1554 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1556 hostname, hosterr := os.Hostname()
1558 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1560 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1563 runner.finalState = "Queued"
1566 runner.stopSignals()
1567 runner.CleanupDirs()
1569 runner.CrunchLog.Printf("crunch-run finished")
1570 runner.CrunchLog.Close()
1574 // checkErr prints e (unless it's nil) and sets err to
1575 // e (unless err is already non-nil). Thus, if err
1576 // hasn't already been assigned when Run() returns,
1577 // this cleanup func will cause Run() to return the
1578 // first non-nil error that is passed to checkErr().
1579 checkErr := func(e error) {
1583 runner.CrunchLog.Print(e)
1587 if runner.finalState == "Complete" {
1588 // There was an error in the finalization.
1589 runner.finalState = "Cancelled"
1593 // Log the error encountered in Run(), if any
1596 if runner.finalState == "Queued" {
1597 runner.UpdateContainerFinal()
1601 if runner.IsCancelled() {
1602 runner.finalState = "Cancelled"
1603 // but don't return yet -- we still want to
1604 // capture partial output and write logs
1607 checkErr(runner.CaptureOutput())
1608 checkErr(runner.stopHoststat())
1609 checkErr(runner.CommitLogs())
1610 checkErr(runner.UpdateContainerFinal())
1613 err = runner.fetchContainerRecord()
1617 runner.setupSignals()
1618 runner.startHoststat()
1620 // check for and/or load image
1621 err = runner.LoadImage()
1623 if !runner.checkBrokenNode(err) {
1624 // Failed to load image but not due to a "broken node"
1625 // condition, probably user error.
1626 runner.finalState = "Cancelled"
1628 err = fmt.Errorf("While loading container image: %v", err)
1632 // set up FUSE mount and binds
1633 err = runner.SetupMounts()
1635 runner.finalState = "Cancelled"
1636 err = fmt.Errorf("While setting up mounts: %v", err)
1640 err = runner.CreateContainer()
1644 err = runner.LogHostInfo()
1648 err = runner.LogNodeRecord()
1652 err = runner.LogContainerRecord()
1657 if runner.IsCancelled() {
1661 err = runner.UpdateContainerRunning()
1665 runner.finalState = "Cancelled"
1667 runner.startCrunchstat()
1669 err = runner.StartContainer()
1671 runner.checkBrokenNode(err)
1675 err = runner.WaitFinish()
1676 if err == nil && !runner.IsCancelled() {
1677 runner.finalState = "Complete"
1682 // Fetch the current container record (uuid = runner.Container.UUID)
1683 // into runner.Container.
1684 func (runner *ContainerRunner) fetchContainerRecord() error {
1685 reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1687 return fmt.Errorf("error fetching container record: %v", err)
1689 defer reader.Close()
1691 dec := json.NewDecoder(reader)
1693 err = dec.Decode(&runner.Container)
1695 return fmt.Errorf("error decoding container record: %v", err)
1700 // NewContainerRunner creates a new container runner.
1701 func NewContainerRunner(api IArvadosClient,
1703 docker ThinDockerClient,
1704 containerUUID string) *ContainerRunner {
1706 cr := &ContainerRunner{ArvClient: api, Kc: kc, Docker: docker}
1707 cr.NewLogWriter = cr.NewArvLogWriter
1708 cr.RunArvMount = cr.ArvMountCmd
1709 cr.MkTempDir = ioutil.TempDir
1710 cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
1711 cr.Container.UUID = containerUUID
1712 cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
1713 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1715 loadLogThrottleParams(api)
1721 statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1722 cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1723 cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1724 cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1725 caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates")
1726 enableNetwork := flag.String("container-enable-networking", "default",
1727 `Specify if networking should be enabled for container. One of 'default', 'always':
1728 default: only enable networking if container requests it.
1729 always: containers always have networking enabled
1731 networkMode := flag.String("container-network-mode", "default",
1732 `Set networking mode for container. Corresponds to Docker network mode (--net).
1734 memprofile := flag.String("memprofile", "", "write memory profile to `file` after running container")
1735 getVersion := flag.Bool("version", false, "Print version information and exit.")
1738 // Print version information if requested
1740 fmt.Printf("crunch-run %s\n", version)
1744 log.Printf("crunch-run %s started", version)
1746 containerId := flag.Arg(0)
1748 if *caCertsPath != "" {
1749 arvadosclient.CertFiles = []string{*caCertsPath}
1752 api, err := arvadosclient.MakeArvadosClient()
1754 log.Fatalf("%s: %v", containerId, err)
1758 kc, kcerr := keepclient.MakeKeepClient(api)
1760 log.Fatalf("%s: %v", containerId, kcerr)
1762 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1765 // API version 1.21 corresponds to Docker 1.9, which is currently the
1766 // minimum version we want to support.
1767 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1769 cr := NewContainerRunner(api, kc, docker, containerId)
1770 if dockererr != nil {
1771 cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
1772 cr.checkBrokenNode(dockererr)
1773 cr.CrunchLog.Close()
1777 parentTemp, tmperr := cr.MkTempDir("", "crunch-run")
1779 log.Fatalf("%s: %v", containerId, tmperr)
1782 cr.parentTemp = parentTemp
1783 cr.statInterval = *statInterval
1784 cr.cgroupRoot = *cgroupRoot
1785 cr.expectCgroupParent = *cgroupParent
1786 cr.enableNetwork = *enableNetwork
1787 cr.networkMode = *networkMode
1788 if *cgroupParentSubsystem != "" {
1789 p := findCgroup(*cgroupParentSubsystem)
1790 cr.setCgroupParent = p
1791 cr.expectCgroupParent = p
1796 if *memprofile != "" {
1797 f, err := os.Create(*memprofile)
1799 log.Printf("could not create memory profile: ", err)
1801 runtime.GC() // get up-to-date statistics
1802 if err := pprof.WriteHeapProfile(f); err != nil {
1803 log.Printf("could not write memory profile: ", err)
1805 closeerr := f.Close()
1806 if closeerr != nil {
1807 log.Printf("closing memprofile file: ", err)
1812 log.Fatalf("%s: %v", containerId, runerr)