1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
30 "git.arvados.org/arvados.git/lib/cmd"
31 "git.arvados.org/arvados.git/lib/crunchstat"
32 "git.arvados.org/arvados.git/sdk/go/arvados"
33 "git.arvados.org/arvados.git/sdk/go/arvadosclient"
34 "git.arvados.org/arvados.git/sdk/go/keepclient"
35 "git.arvados.org/arvados.git/sdk/go/manifest"
36 "golang.org/x/net/context"
38 dockertypes "github.com/docker/docker/api/types"
39 dockercontainer "github.com/docker/docker/api/types/container"
40 dockernetwork "github.com/docker/docker/api/types/network"
41 dockerclient "github.com/docker/docker/client"
46 var Command = command{}
48 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
49 type IArvadosClient interface {
50 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
51 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
52 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
53 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
54 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
55 Discovery(key string) (interface{}, error)
58 // ErrCancelled is the error returned when the container is cancelled.
59 var ErrCancelled = errors.New("Cancelled")
61 // IKeepClient is the minimal Keep API methods used by crunch-run.
62 type IKeepClient interface {
63 PutB(buf []byte) (string, int, error)
64 ReadAt(locator string, p []byte, off int) (int, error)
65 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
66 LocalLocator(locator string) (string, error)
70 // NewLogWriter is a factory function to create a new log writer.
71 type NewLogWriter func(name string) (io.WriteCloser, error)
73 type RunArvMount func(args []string, tok string) (*exec.Cmd, error)
75 type MkTempDir func(string, string) (string, error)
77 // ThinDockerClient is the minimal Docker client interface used by crunch-run.
78 type ThinDockerClient interface {
79 ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
80 ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
81 networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
82 ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
83 ContainerRemove(ctx context.Context, container string, options dockertypes.ContainerRemoveOptions) error
84 ContainerWait(ctx context.Context, container string, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error)
85 ContainerInspect(ctx context.Context, id string) (dockertypes.ContainerJSON, error)
86 ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
87 ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
88 ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
91 type PsProcess interface {
92 CmdlineSlice() ([]string, error)
95 // ContainerRunner is the main stateful struct used for a single execution of a
97 type ContainerRunner struct {
98 ContainerExecRunner ThinContainerExecRunner
100 // Dispatcher client is initialized with the Dispatcher token.
101 // This is a privileged token used to manage container status
104 // We have both dispatcherClient and DispatcherArvClient
105 // because there are two different incompatible Arvados Go
106 // SDKs and we have to use both (hopefully this gets fixed in
108 dispatcherClient *arvados.Client
109 DispatcherArvClient IArvadosClient
110 DispatcherKeepClient IKeepClient
112 // Container client is initialized with the Container token
113 // This token controls the permissions of the container, and
114 // must be used for operations such as reading collections.
116 // Same comment as above applies to
117 // containerClient/ContainerArvClient.
118 containerClient *arvados.Client
119 ContainerArvClient IArvadosClient
120 ContainerKeepClient IKeepClient
122 Container arvados.Container
126 NewLogWriter NewLogWriter
127 loggingDone chan bool
128 CrunchLog *ThrottledLogger
129 Stdout io.WriteCloser
130 Stderr io.WriteCloser
133 LogCollection arvados.CollectionFileSystem
135 RunArvMount RunArvMount
141 Volumes map[string]struct{}
143 SigChan chan os.Signal
144 ArvMountExit chan error
145 SecretMounts map[string]arvados.Mount
146 MkArvClient func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error)
150 statLogger io.WriteCloser
151 statReporter *crunchstat.Reporter
152 hoststatLogger io.WriteCloser
153 hoststatReporter *crunchstat.Reporter
154 statInterval time.Duration
156 // What we expect the container's cgroup parent to be.
157 expectCgroupParent string
158 // What we tell docker to use as the container's cgroup
159 // parent. Note: Ideally we would use the same field for both
160 // expectCgroupParent and setCgroupParent, and just make it
161 // default to "docker". However, when using docker < 1.10 with
162 // systemd, specifying a non-empty cgroup parent (even the
163 // default value "docker") hits a docker bug
164 // (https://github.com/docker/docker/issues/17126). Using two
165 // separate fields makes it possible to use the "expect cgroup
166 // parent to be X" feature even on sites where the "specify
167 // cgroup parent" feature breaks.
168 setCgroupParent string
170 cStateLock sync.Mutex
171 cCancelled bool // StopContainer() invoked
172 cRemoved bool // docker confirmed the container no longer exists
174 enableNetwork string // one of "default" or "always"
175 networkMode string // passed through to HostConfig.NetworkMode
176 arvMountLog *ThrottledLogger
178 containerWatchdogInterval time.Duration
183 // setupSignals sets up signal handling to gracefully terminate the underlying
184 // Docker container and update state when receiving a TERM, INT or QUIT signal.
185 func (runner *ContainerRunner) setupSignals() {
186 runner.SigChan = make(chan os.Signal, 1)
187 signal.Notify(runner.SigChan, syscall.SIGTERM)
188 signal.Notify(runner.SigChan, syscall.SIGINT)
189 signal.Notify(runner.SigChan, syscall.SIGQUIT)
191 go func(sig chan os.Signal) {
198 // stop the underlying Docker container.
199 func (runner *ContainerRunner) stop(sig os.Signal) {
200 runner.cStateLock.Lock()
201 defer runner.cStateLock.Unlock()
203 runner.CrunchLog.Printf("caught signal: %v", sig)
205 if runner.ContainerID == "" {
208 runner.cCancelled = true
209 runner.CrunchLog.Printf("removing container")
210 err := runner.ContainerExecRunner.ContainerRemove(context.TODO(), runner.ContainerID, ContainerRemoveOptions{Force: true})
212 runner.CrunchLog.Printf("error removing container: %s", err)
214 if err == nil || strings.Contains(err.Error(), "No such container: "+runner.ContainerID) {
215 runner.cRemoved = true
219 var errorBlacklist = []string{
220 "(?ms).*[Cc]annot connect to the Docker daemon.*",
221 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
222 "(?ms).*grpc: the connection is unavailable.*",
224 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)")
226 func (runner *ContainerRunner) runBrokenNodeHook() {
227 if *brokenNodeHook == "" {
228 path := filepath.Join(lockdir, brokenfile)
229 runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
230 f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
232 runner.CrunchLog.Printf("Error writing %s: %s", path, err)
237 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
239 c := exec.Command(*brokenNodeHook)
240 c.Stdout = runner.CrunchLog
241 c.Stderr = runner.CrunchLog
244 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
249 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
250 for _, d := range errorBlacklist {
251 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
252 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
253 runner.runBrokenNodeHook()
260 // LoadImage determines the docker image id from the container record and
261 // checks if it is available in the local Docker image store. If not, it loads
262 // the image from Keep.
263 func (runner *ContainerRunner) LoadImage() (err error) {
265 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
267 var collection arvados.Collection
268 err = runner.ContainerArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection)
270 return fmt.Errorf("While getting container image collection: %v", err)
272 manifest := manifest.Manifest{Text: collection.ManifestText}
273 var img, imageID string
274 for ms := range manifest.StreamIter() {
275 img = ms.FileStreamSegments[0].Name
276 if !strings.HasSuffix(img, ".tar") {
277 return fmt.Errorf("First file in the container image collection does not end in .tar")
279 imageID = img[:len(img)-4]
282 runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
284 _, _, err = runner.ContainerExecRunner.ImageInspectWithRaw(context.TODO(), imageID)
286 runner.CrunchLog.Print("Loading Docker image from keep")
288 var readCloser io.ReadCloser
289 readCloser, err = runner.ContainerKeepClient.ManifestFileReader(manifest, img)
291 return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
294 response, err := runner.ContainerExecRunner.ImageLoad(context.TODO(), readCloser, true)
296 return fmt.Errorf("While loading container image into Docker: %v", err)
299 defer response.Body.Close()
300 rbody, err := ioutil.ReadAll(response.Body)
302 return fmt.Errorf("Reading response to image load: %v", err)
304 runner.CrunchLog.Printf("Docker response: %s", rbody)
306 runner.CrunchLog.Print("Docker image is available")
309 runner.ContainerExecRunner.SetImage(imageID)
311 runner.ContainerKeepClient.ClearBlockCache()
316 func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) (c *exec.Cmd, err error) {
317 c = exec.Command("arv-mount", arvMountCmd...)
319 // Copy our environment, but override ARVADOS_API_TOKEN with
320 // the container auth token.
322 for _, s := range os.Environ() {
323 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
324 c.Env = append(c.Env, s)
327 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
329 w, err := runner.NewLogWriter("arv-mount")
333 runner.arvMountLog = NewThrottledLogger(w)
334 c.Stdout = runner.arvMountLog
335 c.Stderr = runner.arvMountLog
337 runner.CrunchLog.Printf("Running %v", c.Args)
344 statReadme := make(chan bool)
345 runner.ArvMountExit = make(chan error)
350 time.Sleep(100 * time.Millisecond)
351 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
363 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
365 runner.ArvMountExit <- mnterr
366 close(runner.ArvMountExit)
372 case err := <-runner.ArvMountExit:
373 runner.ArvMount = nil
381 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
382 if runner.ArvMountPoint == "" {
383 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
388 func copyfile(src string, dst string) (err error) {
389 srcfile, err := os.Open(src)
394 os.MkdirAll(path.Dir(dst), 0777)
396 dstfile, err := os.Create(dst)
400 _, err = io.Copy(dstfile, srcfile)
405 err = srcfile.Close()
406 err2 := dstfile.Close()
419 func (runner *ContainerRunner) SetupMounts() (err error) {
420 err = runner.SetupArvMountPoint("keep")
422 return fmt.Errorf("While creating keep mount temp dir: %v", err)
425 token, err := runner.ContainerToken()
427 return fmt.Errorf("could not get container token: %s", err)
432 arvMountCmd := []string{
436 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
438 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
439 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
442 collectionPaths := []string{}
444 runner.Volumes = make(map[string]struct{})
445 needCertMount := true
446 type copyFile struct {
450 var copyFiles []copyFile
453 for bind := range runner.Container.Mounts {
454 binds = append(binds, bind)
456 for bind := range runner.SecretMounts {
457 if _, ok := runner.Container.Mounts[bind]; ok {
458 return fmt.Errorf("secret mount %q conflicts with regular mount", bind)
460 if runner.SecretMounts[bind].Kind != "json" &&
461 runner.SecretMounts[bind].Kind != "text" {
462 return fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
463 bind, runner.SecretMounts[bind].Kind)
465 binds = append(binds, bind)
469 for _, bind := range binds {
470 mnt, ok := runner.Container.Mounts[bind]
472 mnt = runner.SecretMounts[bind]
474 if bind == "stdout" || bind == "stderr" {
475 // Is it a "file" mount kind?
476 if mnt.Kind != "file" {
477 return fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
480 // Does path start with OutputPath?
481 prefix := runner.Container.OutputPath
482 if !strings.HasSuffix(prefix, "/") {
485 if !strings.HasPrefix(mnt.Path, prefix) {
486 return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
491 // Is it a "collection" mount kind?
492 if mnt.Kind != "collection" && mnt.Kind != "json" {
493 return fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
497 if bind == "/etc/arvados/ca-certificates.crt" {
498 needCertMount = false
501 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
502 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
503 return fmt.Errorf("only mount points of kind 'collection', 'text' or 'json' are supported underneath the output_path for %q, was %q", bind, mnt.Kind)
508 case mnt.Kind == "collection" && bind != "stdin":
510 if mnt.UUID != "" && mnt.PortableDataHash != "" {
511 return fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
515 return fmt.Errorf("writing to existing collections currently not permitted")
518 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
519 } else if mnt.PortableDataHash != "" {
520 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
521 return fmt.Errorf("can never write to a collection specified by portable data hash")
523 idx := strings.Index(mnt.PortableDataHash, "/")
525 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
526 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
527 runner.Container.Mounts[bind] = mnt
529 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
530 if mnt.Path != "" && mnt.Path != "." {
531 if strings.HasPrefix(mnt.Path, "./") {
532 mnt.Path = mnt.Path[2:]
533 } else if strings.HasPrefix(mnt.Path, "/") {
534 mnt.Path = mnt.Path[1:]
536 src += "/" + mnt.Path
539 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
540 arvMountCmd = append(arvMountCmd, "--mount-tmp")
541 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
545 if bind == runner.Container.OutputPath {
546 runner.HostOutputDir = src
547 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
548 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
549 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
551 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind))
554 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind))
556 collectionPaths = append(collectionPaths, src)
558 case mnt.Kind == "tmp":
560 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
562 return fmt.Errorf("while creating mount temp dir: %v", err)
564 st, staterr := os.Stat(tmpdir)
566 return fmt.Errorf("while Stat on temp dir: %v", staterr)
568 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
570 return fmt.Errorf("while Chmod temp dir: %v", err)
572 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", tmpdir, bind))
573 if bind == runner.Container.OutputPath {
574 runner.HostOutputDir = tmpdir
577 case mnt.Kind == "json" || mnt.Kind == "text":
579 if mnt.Kind == "json" {
580 filedata, err = json.Marshal(mnt.Content)
582 return fmt.Errorf("encoding json data: %v", err)
585 text, ok := mnt.Content.(string)
587 return fmt.Errorf("content for mount %q must be a string", bind)
589 filedata = []byte(text)
592 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
594 return fmt.Errorf("creating temp dir: %v", err)
596 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
597 err = ioutil.WriteFile(tmpfn, filedata, 0444)
599 return fmt.Errorf("writing temp file: %v", err)
601 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
602 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
604 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind))
607 case mnt.Kind == "git_tree":
608 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
610 return fmt.Errorf("creating temp dir: %v", err)
612 err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
616 runner.Binds = append(runner.Binds, tmpdir+":"+bind+":ro")
620 if runner.HostOutputDir == "" {
621 return fmt.Errorf("output path does not correspond to a writable mount point")
624 if needCertMount && runner.Container.RuntimeConstraints.API {
625 for _, certfile := range arvadosclient.CertFiles {
626 _, err := os.Stat(certfile)
628 runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile))
635 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id")
637 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
639 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
641 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
643 return fmt.Errorf("while trying to start arv-mount: %v", err)
646 for _, p := range collectionPaths {
649 return fmt.Errorf("while checking that input files exist: %v", err)
653 for _, cp := range copyFiles {
654 st, err := os.Stat(cp.src)
656 return fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
659 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
663 target := path.Join(cp.bind, walkpath[len(cp.src):])
664 if walkinfo.Mode().IsRegular() {
665 copyerr := copyfile(walkpath, target)
669 return os.Chmod(target, walkinfo.Mode()|0777)
670 } else if walkinfo.Mode().IsDir() {
671 mkerr := os.MkdirAll(target, 0777)
675 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
677 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
680 } else if st.Mode().IsRegular() {
681 err = copyfile(cp.src, cp.bind)
683 err = os.Chmod(cp.bind, st.Mode()|0777)
687 return fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
694 func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) {
695 // Handle docker log protocol
696 // https://docs.docker.com/engine/reference/api/docker_remote_api_v1.15/#attach-to-a-container
697 defer close(runner.loggingDone)
699 header := make([]byte, 8)
702 _, err = io.ReadAtLeast(containerReader, header, 8)
709 readsize := int64(header[7]) | (int64(header[6]) << 8) | (int64(header[5]) << 16) | (int64(header[4]) << 24)
712 _, err = io.CopyN(runner.Stdout, containerReader, readsize)
715 _, err = io.CopyN(runner.Stderr, containerReader, readsize)
720 runner.CrunchLog.Printf("error reading docker logs: %v", err)
723 err = runner.Stdout.Close()
725 runner.CrunchLog.Printf("error closing stdout logs: %v", err)
728 err = runner.Stderr.Close()
730 runner.CrunchLog.Printf("error closing stderr logs: %v", err)
733 if runner.statReporter != nil {
734 runner.statReporter.Stop()
735 err = runner.statLogger.Close()
737 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
742 func (runner *ContainerRunner) stopHoststat() error {
743 if runner.hoststatReporter == nil {
746 runner.hoststatReporter.Stop()
747 err := runner.hoststatLogger.Close()
749 return fmt.Errorf("error closing hoststat logs: %v", err)
754 func (runner *ContainerRunner) startHoststat() error {
755 w, err := runner.NewLogWriter("hoststat")
759 runner.hoststatLogger = NewThrottledLogger(w)
760 runner.hoststatReporter = &crunchstat.Reporter{
761 Logger: log.New(runner.hoststatLogger, "", 0),
762 CgroupRoot: runner.cgroupRoot,
763 PollPeriod: runner.statInterval,
765 runner.hoststatReporter.Start()
769 func (runner *ContainerRunner) startCrunchstat() error {
770 w, err := runner.NewLogWriter("crunchstat")
774 runner.statLogger = NewThrottledLogger(w)
775 runner.statReporter = &crunchstat.Reporter{
776 CID: runner.ContainerID,
777 Logger: log.New(runner.statLogger, "", 0),
778 CgroupParent: runner.expectCgroupParent,
779 CgroupRoot: runner.cgroupRoot,
780 PollPeriod: runner.statInterval,
781 TempDir: runner.parentTemp,
783 runner.statReporter.Start()
787 type infoCommand struct {
792 // LogHostInfo logs info about the current host, for debugging and
793 // accounting purposes. Although it's logged as "node-info", this is
794 // about the environment where crunch-run is actually running, which
795 // might differ from what's described in the node record (see
797 func (runner *ContainerRunner) LogHostInfo() (err error) {
798 w, err := runner.NewLogWriter("node-info")
803 commands := []infoCommand{
805 label: "Host Information",
806 cmd: []string{"uname", "-a"},
809 label: "CPU Information",
810 cmd: []string{"cat", "/proc/cpuinfo"},
813 label: "Memory Information",
814 cmd: []string{"cat", "/proc/meminfo"},
818 cmd: []string{"df", "-m", "/", os.TempDir()},
821 label: "Disk INodes",
822 cmd: []string{"df", "-i", "/", os.TempDir()},
826 // Run commands with informational output to be logged.
827 for _, command := range commands {
828 fmt.Fprintln(w, command.label)
829 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
832 if err := cmd.Run(); err != nil {
833 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
842 return fmt.Errorf("While closing node-info logs: %v", err)
847 // LogContainerRecord gets and saves the raw JSON container record from the API server
848 func (runner *ContainerRunner) LogContainerRecord() error {
849 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
850 if !logged && err == nil {
851 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
856 // LogNodeRecord logs the current host's InstanceType config entry (or
857 // the arvados#node record, if running via crunch-dispatch-slurm).
858 func (runner *ContainerRunner) LogNodeRecord() error {
859 if it := os.Getenv("InstanceType"); it != "" {
860 // Dispatched via arvados-dispatch-cloud. Save
861 // InstanceType config fragment received from
862 // dispatcher on stdin.
863 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
868 _, err = io.WriteString(w, it)
874 // Dispatched via crunch-dispatch-slurm. Look up
875 // apiserver's node record corresponding to
877 hostname := os.Getenv("SLURMD_NODENAME")
879 hostname, _ = os.Hostname()
881 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
882 // The "info" field has admin-only info when
883 // obtained with a privileged token, and
884 // should not be logged.
885 node, ok := resp.(map[string]interface{})
893 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
894 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
899 ArvClient: runner.DispatcherArvClient,
900 UUID: runner.Container.UUID,
901 loggingStream: label,
905 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
907 return false, fmt.Errorf("error getting %s record: %v", label, err)
911 dec := json.NewDecoder(reader)
913 var resp map[string]interface{}
914 if err = dec.Decode(&resp); err != nil {
915 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
917 items, ok := resp["items"].([]interface{})
919 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
920 } else if len(items) < 1 {
926 // Re-encode it using indentation to improve readability
927 enc := json.NewEncoder(w)
928 enc.SetIndent("", " ")
929 if err = enc.Encode(items[0]); err != nil {
930 return false, fmt.Errorf("error logging %s record: %v", label, err)
934 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
939 // AttachStreams connects the docker container stdin, stdout and stderr logs
940 // to the Arvados logger which logs to Keep and the API server logs table.
941 func (runner *ContainerRunner) AttachStreams() (err error) {
943 runner.CrunchLog.Print("Attaching container streams")
945 // If stdin mount is provided, attach it to the docker container
946 var stdinRdr arvados.File
948 if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
949 if stdinMnt.Kind == "collection" {
950 var stdinColl arvados.Collection
951 collID := stdinMnt.UUID
953 collID = stdinMnt.PortableDataHash
955 err = runner.ContainerArvClient.Get("collections", collID, nil, &stdinColl)
957 return fmt.Errorf("While getting stdin collection: %v", err)
960 stdinRdr, err = runner.ContainerKeepClient.ManifestFileReader(
961 manifest.Manifest{Text: stdinColl.ManifestText},
963 if os.IsNotExist(err) {
964 return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
965 } else if err != nil {
966 return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
968 } else if stdinMnt.Kind == "json" {
969 stdinJSON, err = json.Marshal(stdinMnt.Content)
971 return fmt.Errorf("While encoding stdin json data: %v", err)
976 stdinUsed := stdinRdr != nil || len(stdinJSON) != 0
977 response, err := runner.ContainerExecRunner.ContainerAttach(context.TODO(), runner.ContainerID,
978 ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
980 return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
983 runner.loggingDone = make(chan bool)
985 if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok {
986 stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
990 runner.Stdout = stdoutFile
991 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
994 runner.Stdout = NewThrottledLogger(w)
997 if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
998 stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
1002 runner.Stderr = stderrFile
1003 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
1006 runner.Stderr = NewThrottledLogger(w)
1009 if stdinRdr != nil {
1011 _, err := io.Copy(response.Conn, stdinRdr)
1013 runner.CrunchLog.Printf("While writing stdin collection to docker container: %v", err)
1017 response.CloseWrite()
1019 } else if len(stdinJSON) != 0 {
1021 _, err := io.Copy(response.Conn, bytes.NewReader(stdinJSON))
1023 runner.CrunchLog.Printf("While writing stdin json to docker container: %v", err)
1026 response.CloseWrite()
1030 go runner.ProcessDockerAttach(response.Reader)
1035 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
1036 stdoutPath := mntPath[len(runner.Container.OutputPath):]
1037 index := strings.LastIndex(stdoutPath, "/")
1039 subdirs := stdoutPath[:index]
1041 st, err := os.Stat(runner.HostOutputDir)
1043 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
1045 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
1046 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
1048 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
1052 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
1054 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
1057 return stdoutFile, nil
1060 // CreateContainer creates the docker container.
1061 func (runner *ContainerRunner) CreateContainer() error {
1062 runner.CrunchLog.Print("Creating Docker container")
1064 containerConfig, err := runner.ContainerExecRunner.GetContainerConfig()
1065 hostConfig, err := runner.ContainerExecRunner.GetHostConfig()
1067 containerConfig.Cmd = runner.Container.Command
1068 if runner.Container.Cwd != "." {
1069 containerConfig.WorkingDir = runner.Container.Cwd
1072 for k, v := range runner.Container.Environment {
1073 containerConfig.Env = append(containerConfig.Env, k+"="+v)
1076 containerConfig.Volumes = runner.Volumes
1078 maxRAM := int64(runner.Container.RuntimeConstraints.RAM)
1079 minDockerRAM := int64(16)
1080 if maxRAM < minDockerRAM*1024*1024 {
1081 // Docker daemon won't let you set a limit less than ~10 MiB
1082 maxRAM = minDockerRAM * 1024 * 1024
1084 hostConfig = HostConfig{
1085 Binds: runner.Binds,
1086 LogConfig: LogConfig{
1089 Resources: Resources{
1090 CgroupParent: runner.setCgroupParent,
1091 NanoCPUs: int64(runner.Container.RuntimeConstraints.VCPUs) * 1000000000,
1092 Memory: maxRAM, // RAM
1093 MemorySwap: maxRAM, // RAM+swap
1094 KernelMemory: maxRAM, // kernel portion
1097 runner.ContainerExecRunner.SetHostConfig(hostConfig)
1098 if runner.Container.RuntimeConstraints.API {
1099 tok, err := runner.ContainerToken()
1103 containerConfig.Env = append(containerConfig.Env,
1104 "ARVADOS_API_TOKEN="+tok,
1105 "ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
1106 "ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
1108 runner.ContainerExecRunner.SetNetworkMode(NetworkMode(runner.networkMode))
1110 if runner.enableNetwork == "always" {
1111 runner.ContainerExecRunner.SetNetworkMode(NetworkMode(runner.networkMode))
1113 runner.ContainerExecRunner.SetNetworkMode("none")
1117 _, stdinUsed := runner.Container.Mounts["stdin"]
1118 containerConfig.OpenStdin = stdinUsed
1119 containerConfig.StdinOnce = stdinUsed
1120 containerConfig.AttachStdin = stdinUsed
1121 containerConfig.AttachStdout = true
1122 containerConfig.AttachStderr = true
1124 createdBody, err := runner.ContainerExecRunner.ContainerCreate(context.TODO(), containerConfig, hostConfig, nil, runner.Container.UUID)
1126 return fmt.Errorf("While creating container: %v", err)
1129 runner.ContainerID = createdBody.ID
1131 return runner.AttachStreams()
1134 // StartContainer starts the docker container created by CreateContainer.
1135 func (runner *ContainerRunner) StartContainer() error {
1136 runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
1137 runner.cStateLock.Lock()
1138 defer runner.cStateLock.Unlock()
1139 if runner.cCancelled {
1142 err := runner.ContainerExecRunner.ContainerStart(context.TODO(), runner.ContainerID,
1143 ContainerStartOptions{})
1146 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1147 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])
1149 return fmt.Errorf("could not start container: %v%s", err, advice)
1154 // WaitFinish waits for the container to terminate, capture the exit code, and
1155 // close the stdout/stderr logging.
1156 func (runner *ContainerRunner) WaitFinish() error {
1157 var runTimeExceeded <-chan time.Time
1158 runner.CrunchLog.Print("Waiting for container to finish")
1160 waitOk, waitErr := runner.ContainerExecRunner.ContainerWait(context.TODO(), runner.ContainerID, WaitConditionNotRunning)
1161 arvMountExit := runner.ArvMountExit
1162 if timeout := runner.Container.SchedulingParameters.MaxRunTime; timeout > 0 {
1163 runTimeExceeded = time.After(time.Duration(timeout) * time.Second)
1166 containerGone := make(chan struct{})
1168 defer close(containerGone)
1169 if runner.containerWatchdogInterval < 1 {
1170 runner.containerWatchdogInterval = time.Minute
1172 for range time.NewTicker(runner.containerWatchdogInterval).C {
1173 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(runner.containerWatchdogInterval))
1174 ctr, err := runner.ContainerExecRunner.ContainerInspect(ctx, runner.ContainerID)
1176 runner.cStateLock.Lock()
1177 done := runner.cRemoved || runner.ExitCode != nil
1178 runner.cStateLock.Unlock()
1181 } else if err != nil {
1182 runner.CrunchLog.Printf("Error inspecting container: %s", err)
1183 runner.checkBrokenNode(err)
1185 } else if ctr.State == nil || !(ctr.State.Running || ctr.State.Status == "created") {
1186 runner.CrunchLog.Printf("Container is not running: State=%v", ctr.State)
1194 case waitBody := <-waitOk:
1195 runner.CrunchLog.Printf("Container exited with code: %v", waitBody.StatusCode)
1196 code := int(waitBody.StatusCode)
1197 runner.ExitCode = &code
1199 // wait for stdout/stderr to complete
1200 <-runner.loggingDone
1203 case err := <-waitErr:
1204 return fmt.Errorf("container wait: %v", err)
1206 case <-arvMountExit:
1207 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1209 // arvMountExit will always be ready now that
1210 // it's closed, but that doesn't interest us.
1213 case <-runTimeExceeded:
1214 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1216 runTimeExceeded = nil
1218 case <-containerGone:
1219 return errors.New("docker client never returned status")
1224 func (runner *ContainerRunner) updateLogs() {
1225 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1228 sigusr1 := make(chan os.Signal, 1)
1229 signal.Notify(sigusr1, syscall.SIGUSR1)
1230 defer signal.Stop(sigusr1)
1232 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1233 saveAtSize := crunchLogUpdateSize
1239 saveAtTime = time.Now()
1241 runner.logMtx.Lock()
1242 done := runner.LogsPDH != nil
1243 runner.logMtx.Unlock()
1247 size := runner.LogCollection.Size()
1248 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1251 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1252 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1253 saved, err := runner.saveLogCollection(false)
1255 runner.CrunchLog.Printf("error updating log collection: %s", err)
1259 var updated arvados.Container
1260 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1261 "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1264 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1272 // CaptureOutput saves data from the container's output directory if
1273 // needed, and updates the container output accordingly.
1274 func (runner *ContainerRunner) CaptureOutput() error {
1275 if runner.Container.RuntimeConstraints.API {
1276 // Output may have been set directly by the container, so
1277 // refresh the container record to check.
1278 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1279 nil, &runner.Container)
1283 if runner.Container.Output != "" {
1284 // Container output is already set.
1285 runner.OutputPDH = &runner.Container.Output
1290 txt, err := (&copier{
1291 client: runner.containerClient,
1292 arvClient: runner.ContainerArvClient,
1293 keepClient: runner.ContainerKeepClient,
1294 hostOutputDir: runner.HostOutputDir,
1295 ctrOutputDir: runner.Container.OutputPath,
1296 binds: runner.Binds,
1297 mounts: runner.Container.Mounts,
1298 secretMounts: runner.SecretMounts,
1299 logger: runner.CrunchLog,
1304 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1305 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1306 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1310 txt, err = fs.MarshalManifest(".")
1315 var resp arvados.Collection
1316 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1317 "ensure_unique_name": true,
1318 "collection": arvadosclient.Dict{
1320 "name": "output for " + runner.Container.UUID,
1321 "manifest_text": txt,
1325 return fmt.Errorf("error creating output collection: %v", err)
1327 runner.OutputPDH = &resp.PortableDataHash
1331 func (runner *ContainerRunner) CleanupDirs() {
1332 if runner.ArvMount != nil {
1334 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1335 umount.Stdout = runner.CrunchLog
1336 umount.Stderr = runner.CrunchLog
1337 runner.CrunchLog.Printf("Running %v", umount.Args)
1338 umnterr := umount.Start()
1341 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1343 // If arv-mount --unmount gets stuck for any reason, we
1344 // don't want to wait for it forever. Do Wait() in a goroutine
1345 // so it doesn't block crunch-run.
1346 umountExit := make(chan error)
1348 mnterr := umount.Wait()
1350 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1352 umountExit <- mnterr
1355 for again := true; again; {
1361 case <-runner.ArvMountExit:
1363 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1364 runner.CrunchLog.Printf("Timed out waiting for unmount")
1366 umount.Process.Kill()
1368 runner.ArvMount.Process.Kill()
1374 if runner.ArvMountPoint != "" {
1375 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1376 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1380 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1381 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1385 // CommitLogs posts the collection containing the final container logs.
1386 func (runner *ContainerRunner) CommitLogs() error {
1388 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1389 runner.cStateLock.Lock()
1390 defer runner.cStateLock.Unlock()
1392 runner.CrunchLog.Print(runner.finalState)
1394 if runner.arvMountLog != nil {
1395 runner.arvMountLog.Close()
1397 runner.CrunchLog.Close()
1399 // Closing CrunchLog above allows them to be committed to Keep at this
1400 // point, but re-open crunch log with ArvClient in case there are any
1401 // other further errors (such as failing to write the log to Keep!)
1402 // while shutting down
1403 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1404 ArvClient: runner.DispatcherArvClient,
1405 UUID: runner.Container.UUID,
1406 loggingStream: "crunch-run",
1409 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1412 if runner.LogsPDH != nil {
1413 // If we have already assigned something to LogsPDH,
1414 // we must be closing the re-opened log, which won't
1415 // end up getting attached to the container record and
1416 // therefore doesn't need to be saved as a collection
1417 // -- it exists only to send logs to other channels.
1420 saved, err := runner.saveLogCollection(true)
1422 return fmt.Errorf("error saving log collection: %s", err)
1424 runner.logMtx.Lock()
1425 defer runner.logMtx.Unlock()
1426 runner.LogsPDH = &saved.PortableDataHash
1430 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1431 runner.logMtx.Lock()
1432 defer runner.logMtx.Unlock()
1433 if runner.LogsPDH != nil {
1434 // Already finalized.
1437 updates := arvadosclient.Dict{
1438 "name": "logs for " + runner.Container.UUID,
1440 mt, err1 := runner.LogCollection.MarshalManifest(".")
1442 // Only send updated manifest text if there was no
1444 updates["manifest_text"] = mt
1447 // Even if flushing the manifest had an error, we still want
1448 // to update the log record, if possible, to push the trash_at
1449 // and delete_at times into the future. Details on bug
1452 updates["is_trashed"] = true
1454 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1455 updates["trash_at"] = exp
1456 updates["delete_at"] = exp
1458 reqBody := arvadosclient.Dict{"collection": updates}
1460 if runner.logUUID == "" {
1461 reqBody["ensure_unique_name"] = true
1462 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1464 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1467 runner.logUUID = response.UUID
1470 if err1 != nil || err2 != nil {
1471 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1476 // UpdateContainerRunning updates the container state to "Running"
1477 func (runner *ContainerRunner) UpdateContainerRunning() error {
1478 runner.cStateLock.Lock()
1479 defer runner.cStateLock.Unlock()
1480 if runner.cCancelled {
1483 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1484 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1487 // ContainerToken returns the api_token the container (and any
1488 // arv-mount processes) are allowed to use.
1489 func (runner *ContainerRunner) ContainerToken() (string, error) {
1490 if runner.token != "" {
1491 return runner.token, nil
1494 var auth arvados.APIClientAuthorization
1495 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1499 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1500 return runner.token, nil
1503 // UpdateContainerFinal updates the container record state on API
1504 // server to "Complete" or "Cancelled"
1505 func (runner *ContainerRunner) UpdateContainerFinal() error {
1506 update := arvadosclient.Dict{}
1507 update["state"] = runner.finalState
1508 if runner.LogsPDH != nil {
1509 update["log"] = *runner.LogsPDH
1511 if runner.finalState == "Complete" {
1512 if runner.ExitCode != nil {
1513 update["exit_code"] = *runner.ExitCode
1515 if runner.OutputPDH != nil {
1516 update["output"] = *runner.OutputPDH
1519 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1522 // IsCancelled returns the value of Cancelled, with goroutine safety.
1523 func (runner *ContainerRunner) IsCancelled() bool {
1524 runner.cStateLock.Lock()
1525 defer runner.cStateLock.Unlock()
1526 return runner.cCancelled
1529 // NewArvLogWriter creates an ArvLogWriter
1530 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1531 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1535 return &ArvLogWriter{
1536 ArvClient: runner.DispatcherArvClient,
1537 UUID: runner.Container.UUID,
1538 loggingStream: name,
1539 writeCloser: writer,
1543 // Run the full container lifecycle.
1544 func (runner *ContainerRunner) Run() (err error) {
1545 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1546 runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID)
1548 hostname, hosterr := os.Hostname()
1550 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1552 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1555 runner.finalState = "Queued"
1558 runner.CleanupDirs()
1560 runner.CrunchLog.Printf("crunch-run finished")
1561 runner.CrunchLog.Close()
1564 err = runner.fetchContainerRecord()
1568 if runner.Container.State != "Locked" {
1569 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1573 // checkErr prints e (unless it's nil) and sets err to
1574 // e (unless err is already non-nil). Thus, if err
1575 // hasn't already been assigned when Run() returns,
1576 // this cleanup func will cause Run() to return the
1577 // first non-nil error that is passed to checkErr().
1578 checkErr := func(errorIn string, e error) {
1582 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1586 if runner.finalState == "Complete" {
1587 // There was an error in the finalization.
1588 runner.finalState = "Cancelled"
1592 // Log the error encountered in Run(), if any
1593 checkErr("Run", err)
1595 if runner.finalState == "Queued" {
1596 runner.UpdateContainerFinal()
1600 if runner.IsCancelled() {
1601 runner.finalState = "Cancelled"
1602 // but don't return yet -- we still want to
1603 // capture partial output and write logs
1606 checkErr("CaptureOutput", runner.CaptureOutput())
1607 checkErr("stopHoststat", runner.stopHoststat())
1608 checkErr("CommitLogs", runner.CommitLogs())
1609 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1612 runner.setupSignals()
1613 err = runner.startHoststat()
1618 // check for and/or load image
1619 err = runner.LoadImage()
1621 if !runner.checkBrokenNode(err) {
1622 // Failed to load image but not due to a "broken node"
1623 // condition, probably user error.
1624 runner.finalState = "Cancelled"
1626 err = fmt.Errorf("While loading container image: %v", err)
1630 // set up FUSE mount and binds
1631 err = runner.SetupMounts()
1633 runner.finalState = "Cancelled"
1634 err = fmt.Errorf("While setting up mounts: %v", err)
1638 err = runner.CreateContainer()
1642 err = runner.LogHostInfo()
1646 err = runner.LogNodeRecord()
1650 err = runner.LogContainerRecord()
1655 if runner.IsCancelled() {
1659 err = runner.UpdateContainerRunning()
1663 runner.finalState = "Cancelled"
1665 err = runner.startCrunchstat()
1670 err = runner.StartContainer()
1672 runner.checkBrokenNode(err)
1676 err = runner.WaitFinish()
1677 if err == nil && !runner.IsCancelled() {
1678 runner.finalState = "Complete"
1683 // Fetch the current container record (uuid = runner.Container.UUID)
1684 // into runner.Container.
1685 func (runner *ContainerRunner) fetchContainerRecord() error {
1686 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1688 return fmt.Errorf("error fetching container record: %v", err)
1690 defer reader.Close()
1692 dec := json.NewDecoder(reader)
1694 err = dec.Decode(&runner.Container)
1696 return fmt.Errorf("error decoding container record: %v", err)
1700 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1703 containerToken, err := runner.ContainerToken()
1705 return fmt.Errorf("error getting container token: %v", err)
1708 runner.ContainerArvClient, runner.ContainerKeepClient,
1709 runner.containerClient, err = runner.MkArvClient(containerToken)
1711 return fmt.Errorf("error creating container API client: %v", err)
1714 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1716 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1717 return fmt.Errorf("error fetching secret_mounts: %v", err)
1719 // ok && apierr.HttpStatusCode == 404, which means
1720 // secret_mounts isn't supported by this API server.
1722 runner.SecretMounts = sm.SecretMounts
1727 // NewContainerRunner creates a new container runner.
1728 func NewContainerRunner(dispatcherClient *arvados.Client,
1729 dispatcherArvClient IArvadosClient,
1730 dispatcherKeepClient IKeepClient,
1731 containerRunner ThinContainerExecRunner,
1732 containerUUID string) (*ContainerRunner, error) {
1734 cr := &ContainerRunner{
1735 dispatcherClient: dispatcherClient,
1736 DispatcherArvClient: dispatcherArvClient,
1737 DispatcherKeepClient: dispatcherKeepClient,
1738 ContainerExecRunner: containerRunner,
1740 cr.NewLogWriter = cr.NewArvLogWriter
1741 cr.RunArvMount = cr.ArvMountCmd
1742 cr.MkTempDir = ioutil.TempDir
1743 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1744 cl, err := arvadosclient.MakeArvadosClient()
1746 return nil, nil, nil, err
1749 kc, err := keepclient.MakeKeepClient(cl)
1751 return nil, nil, nil, err
1753 c2 := arvados.NewClientFromEnv()
1754 c2.AuthToken = token
1755 return cl, kc, c2, nil
1758 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1762 cr.Container.UUID = containerUUID
1763 w, err := cr.NewLogWriter("crunch-run")
1767 cr.CrunchLog = NewThrottledLogger(w)
1768 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1770 loadLogThrottleParams(dispatcherArvClient)
1776 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1777 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1778 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1779 cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1780 cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1781 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1782 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1783 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1784 stdinEnv := flags.Bool("stdin-env", false, "Load environment variables from JSON message on stdin")
1785 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1786 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1787 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1788 enableNetwork := flags.String("container-enable-networking", "default",
1789 `Specify if networking should be enabled for container. One of 'default', 'always':
1790 default: only enable networking if container requests it.
1791 always: containers always have networking enabled
1793 networkMode := flags.String("container-network-mode", "default",
1794 `Set networking mode for container. Corresponds to Docker network mode (--net).
1796 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1797 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1798 containerRunner := flags.String("container-runner", "docker",
1799 `Specify the container runner. available options: docker, singularity.
1801 ignoreDetachFlag := false
1802 if len(args) > 0 && args[0] == "-no-detach" {
1803 // This process was invoked by a parent process, which
1804 // has passed along its own arguments, including
1805 // -detach, after the leading -no-detach flag. Strip
1806 // the leading -no-detach flag (it's not recognized by
1807 // flags.Parse()) and ignore the -detach flag that
1810 ignoreDetachFlag = true
1813 if err := flags.Parse(args); err == flag.ErrHelp {
1815 } else if err != nil {
1820 if *stdinEnv && !ignoreDetachFlag {
1821 // Load env vars on stdin if asked (but not in a
1822 // detached child process, in which case stdin is
1824 err := loadEnv(os.Stdin)
1831 containerID := flags.Arg(0)
1834 case *detach && !ignoreDetachFlag:
1835 return Detach(containerID, prog, args, os.Stdout, os.Stderr)
1837 return KillProcess(containerID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1839 return ListProcesses(os.Stdout, os.Stderr)
1842 if containerID == "" {
1843 log.Printf("usage: %s [options] UUID", prog)
1847 log.Printf("crunch-run %s started", cmd.Version.String())
1850 if *caCertsPath != "" {
1851 arvadosclient.CertFiles = []string{*caCertsPath}
1854 api, err := arvadosclient.MakeArvadosClient()
1856 log.Printf("%s: %v", containerID, err)
1861 kc, kcerr := keepclient.MakeKeepClient(api)
1863 log.Printf("%s: %v", containerID, kcerr)
1866 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1869 var cr *ContainerRunner
1870 if *containerRunner == "docker" {
1871 // API version 1.21 corresponds to Docker 1.9, which is currently the
1872 // minimum version we want to support.
1873 docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
1875 cr, err = NewContainerRunner(arvados.NewClientFromEnv(), api, kc, adapter(docker), containerID)
1880 if dockererr != nil {
1881 cr.CrunchLog.Printf("%s: %v", containerID, dockererr)
1882 cr.checkBrokenNode(dockererr)
1883 cr.CrunchLog.Close()
1889 singularity, singularityerr := NewSingularityClient()
1891 cr, err = NewContainerRunner(arvados.NewClientFromEnv(), api, kc, singularity, containerID)
1897 if singularityerr != nil {
1898 cr.CrunchLog.Printf("%s: %v", containerID, singularityerr)
1899 //cr.checkBrokenNode(singularityrerr) //
1900 cr.CrunchLog.Close()
1904 cr.gateway = Gateway{
1905 Address: os.Getenv("GatewayAddress"),
1906 AuthSecret: os.Getenv("GatewayAuthSecret"),
1907 ContainerUUID: containerID,
1908 DockerContainerID: &cr.ContainerID,
1911 os.Unsetenv("GatewayAuthSecret")
1912 if cr.gateway.Address != "" {
1913 err = cr.gateway.Start()
1915 log.Printf("error starting gateway server: %s", err)
1920 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerID+".")
1922 log.Printf("%s: %v", containerID, tmperr)
1926 cr.parentTemp = parentTemp
1927 cr.statInterval = *statInterval
1928 cr.cgroupRoot = *cgroupRoot
1929 cr.expectCgroupParent = *cgroupParent
1930 cr.enableNetwork = *enableNetwork
1931 cr.networkMode = *networkMode
1932 if *cgroupParentSubsystem != "" {
1933 p := findCgroup(*cgroupParentSubsystem)
1934 cr.setCgroupParent = p
1935 cr.expectCgroupParent = p
1940 if *memprofile != "" {
1941 f, err := os.Create(*memprofile)
1943 log.Printf("could not create memory profile: %s", err)
1945 runtime.GC() // get up-to-date statistics
1946 if err := pprof.WriteHeapProfile(f); err != nil {
1947 log.Printf("could not write memory profile: %s", err)
1949 closeerr := f.Close()
1950 if closeerr != nil {
1951 log.Printf("closing memprofile file: %s", err)
1956 log.Printf("%s: %v", containerID, runerr)
1962 func loadEnv(rdr io.Reader) error {
1963 buf, err := ioutil.ReadAll(rdr)
1965 return fmt.Errorf("read stdin: %s", err)
1967 var env map[string]string
1968 err = json.Unmarshal(buf, &env)
1970 return fmt.Errorf("decode stdin: %s", err)
1972 for k, v := range env {
1973 err = os.Setenv(k, v)
1975 return fmt.Errorf("setenv(%q): %s", k, err)