1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
33 "git.arvados.org/arvados.git/lib/cmd"
34 "git.arvados.org/arvados.git/lib/crunchstat"
35 "git.arvados.org/arvados.git/sdk/go/arvados"
36 "git.arvados.org/arvados.git/sdk/go/arvadosclient"
37 "git.arvados.org/arvados.git/sdk/go/keepclient"
38 "git.arvados.org/arvados.git/sdk/go/manifest"
39 "golang.org/x/sys/unix"
44 var Command = command{}
46 // ConfigData contains environment variables and (when needed) cluster
47 // configuration, passed from dispatchcloud to crunch-run on stdin.
48 type ConfigData struct {
51 Cluster *arvados.Cluster
54 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
55 type IArvadosClient interface {
56 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
57 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
58 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
59 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
60 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
61 Discovery(key string) (interface{}, error)
64 // ErrCancelled is the error returned when the container is cancelled.
65 var ErrCancelled = errors.New("Cancelled")
67 // IKeepClient is the minimal Keep API methods used by crunch-run.
68 type IKeepClient interface {
69 BlockWrite(context.Context, arvados.BlockWriteOptions) (arvados.BlockWriteResponse, error)
70 ReadAt(locator string, p []byte, off int) (int, error)
71 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
72 LocalLocator(locator string) (string, error)
74 SetStorageClasses(sc []string)
77 // NewLogWriter is a factory function to create a new log writer.
78 type NewLogWriter func(name string) (io.WriteCloser, error)
80 type RunArvMount func(cmdline []string, tok string) (*exec.Cmd, error)
82 type MkTempDir func(string, string) (string, error)
84 type PsProcess interface {
85 CmdlineSlice() ([]string, error)
88 // ContainerRunner is the main stateful struct used for a single execution of a
90 type ContainerRunner struct {
91 executor containerExecutor
92 executorStdin io.Closer
93 executorStdout io.Closer
94 executorStderr io.Closer
96 // Dispatcher client is initialized with the Dispatcher token.
97 // This is a privileged token used to manage container status
100 // We have both dispatcherClient and DispatcherArvClient
101 // because there are two different incompatible Arvados Go
102 // SDKs and we have to use both (hopefully this gets fixed in
104 dispatcherClient *arvados.Client
105 DispatcherArvClient IArvadosClient
106 DispatcherKeepClient IKeepClient
108 // Container client is initialized with the Container token
109 // This token controls the permissions of the container, and
110 // must be used for operations such as reading collections.
112 // Same comment as above applies to
113 // containerClient/ContainerArvClient.
114 containerClient *arvados.Client
115 ContainerArvClient IArvadosClient
116 ContainerKeepClient IKeepClient
118 Container arvados.Container
121 NewLogWriter NewLogWriter
122 CrunchLog *ThrottledLogger
125 LogCollection arvados.CollectionFileSystem
127 RunArvMount RunArvMount
132 Volumes map[string]struct{}
134 SigChan chan os.Signal
135 ArvMountExit chan error
136 SecretMounts map[string]arvados.Mount
137 MkArvClient func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error)
141 keepstoreLogger io.WriteCloser
142 keepstoreLogbuf *bufThenWrite
143 statLogger io.WriteCloser
144 statReporter *crunchstat.Reporter
145 hoststatLogger io.WriteCloser
146 hoststatReporter *crunchstat.Reporter
147 statInterval time.Duration
149 // What we expect the container's cgroup parent to be.
150 expectCgroupParent string
151 // What we tell docker to use as the container's cgroup
152 // parent. Note: Ideally we would use the same field for both
153 // expectCgroupParent and setCgroupParent, and just make it
154 // default to "docker". However, when using docker < 1.10 with
155 // systemd, specifying a non-empty cgroup parent (even the
156 // default value "docker") hits a docker bug
157 // (https://github.com/docker/docker/issues/17126). Using two
158 // separate fields makes it possible to use the "expect cgroup
159 // parent to be X" feature even on sites where the "specify
160 // cgroup parent" feature breaks.
161 setCgroupParent string
163 cStateLock sync.Mutex
164 cCancelled bool // StopContainer() invoked
166 enableMemoryLimit bool
167 enableNetwork string // one of "default" or "always"
168 networkMode string // "none", "host", or "" -- passed through to executor
169 arvMountLog *ThrottledLogger
171 containerWatchdogInterval time.Duration
176 // setupSignals sets up signal handling to gracefully terminate the
177 // underlying container and update state when receiving a TERM, INT or
179 func (runner *ContainerRunner) setupSignals() {
180 runner.SigChan = make(chan os.Signal, 1)
181 signal.Notify(runner.SigChan, syscall.SIGTERM)
182 signal.Notify(runner.SigChan, syscall.SIGINT)
183 signal.Notify(runner.SigChan, syscall.SIGQUIT)
185 go func(sig chan os.Signal) {
192 // stop the underlying container.
193 func (runner *ContainerRunner) stop(sig os.Signal) {
194 runner.cStateLock.Lock()
195 defer runner.cStateLock.Unlock()
197 runner.CrunchLog.Printf("caught signal: %v", sig)
199 runner.cCancelled = true
200 runner.CrunchLog.Printf("stopping container")
201 err := runner.executor.Stop()
203 runner.CrunchLog.Printf("error stopping container: %s", err)
207 var errorBlacklist = []string{
208 "(?ms).*[Cc]annot connect to the Docker daemon.*",
209 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
210 "(?ms).*grpc: the connection is unavailable.*",
212 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)")
214 func (runner *ContainerRunner) runBrokenNodeHook() {
215 if *brokenNodeHook == "" {
216 path := filepath.Join(lockdir, brokenfile)
217 runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
218 f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
220 runner.CrunchLog.Printf("Error writing %s: %s", path, err)
225 runner.CrunchLog.Printf("Running broken node hook %q", *brokenNodeHook)
227 c := exec.Command(*brokenNodeHook)
228 c.Stdout = runner.CrunchLog
229 c.Stderr = runner.CrunchLog
232 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
237 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
238 for _, d := range errorBlacklist {
239 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
240 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
241 runner.runBrokenNodeHook()
248 // LoadImage determines the docker image id from the container record and
249 // checks if it is available in the local Docker image store. If not, it loads
250 // the image from Keep.
251 func (runner *ContainerRunner) LoadImage() (string, error) {
252 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
254 d, err := os.Open(runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage)
259 allfiles, err := d.Readdirnames(-1)
263 var tarfiles []string
264 for _, fnm := range allfiles {
265 if strings.HasSuffix(fnm, ".tar") {
266 tarfiles = append(tarfiles, fnm)
269 if len(tarfiles) == 0 {
270 return "", fmt.Errorf("image collection does not include a .tar image file")
272 if len(tarfiles) > 1 {
273 return "", fmt.Errorf("cannot choose from multiple tar files in image collection: %v", tarfiles)
275 imageID := tarfiles[0][:len(tarfiles[0])-4]
276 imageTarballPath := runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage + "/" + imageID + ".tar"
277 runner.CrunchLog.Printf("Using Docker image id %q", imageID)
279 runner.CrunchLog.Print("Loading Docker image from keep")
280 err = runner.executor.LoadImage(imageID, imageTarballPath, runner.Container, runner.ArvMountPoint,
281 runner.containerClient)
289 func (runner *ContainerRunner) ArvMountCmd(cmdline []string, token string) (c *exec.Cmd, err error) {
290 c = exec.Command(cmdline[0], cmdline[1:]...)
292 // Copy our environment, but override ARVADOS_API_TOKEN with
293 // the container auth token.
295 for _, s := range os.Environ() {
296 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
297 c.Env = append(c.Env, s)
300 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
302 w, err := runner.NewLogWriter("arv-mount")
306 runner.arvMountLog = NewThrottledLogger(w)
307 scanner := logScanner{
310 "Block not found error",
311 "Unhandled exception during FUSE operation",
313 ReportFunc: runner.reportArvMountWarning,
315 c.Stdout = runner.arvMountLog
316 c.Stderr = io.MultiWriter(runner.arvMountLog, os.Stderr, &scanner)
318 runner.CrunchLog.Printf("Running %v", c.Args)
325 statReadme := make(chan bool)
326 runner.ArvMountExit = make(chan error)
331 time.Sleep(100 * time.Millisecond)
332 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
344 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
346 runner.ArvMountExit <- mnterr
347 close(runner.ArvMountExit)
353 case err := <-runner.ArvMountExit:
354 runner.ArvMount = nil
362 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
363 if runner.ArvMountPoint == "" {
364 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
369 func copyfile(src string, dst string) (err error) {
370 srcfile, err := os.Open(src)
375 os.MkdirAll(path.Dir(dst), 0777)
377 dstfile, err := os.Create(dst)
381 _, err = io.Copy(dstfile, srcfile)
386 err = srcfile.Close()
387 err2 := dstfile.Close()
400 func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
401 bindmounts := map[string]bindmount{}
402 err := runner.SetupArvMountPoint("keep")
404 return nil, fmt.Errorf("While creating keep mount temp dir: %v", err)
407 token, err := runner.ContainerToken()
409 return nil, fmt.Errorf("could not get container token: %s", err)
411 runner.CrunchLog.Printf("container token %q", token)
415 arvMountCmd := []string{
419 "--storage-classes", strings.Join(runner.Container.OutputStorageClasses, ","),
420 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
422 if runner.executor.Runtime() == "docker" {
423 arvMountCmd = append(arvMountCmd, "--allow-other")
426 if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
427 arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
430 collectionPaths := []string{}
431 needCertMount := true
432 type copyFile struct {
436 var copyFiles []copyFile
439 for bind := range runner.Container.Mounts {
440 binds = append(binds, bind)
442 for bind := range runner.SecretMounts {
443 if _, ok := runner.Container.Mounts[bind]; ok {
444 return nil, fmt.Errorf("secret mount %q conflicts with regular mount", bind)
446 if runner.SecretMounts[bind].Kind != "json" &&
447 runner.SecretMounts[bind].Kind != "text" {
448 return nil, fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
449 bind, runner.SecretMounts[bind].Kind)
451 binds = append(binds, bind)
455 for _, bind := range binds {
456 mnt, ok := runner.Container.Mounts[bind]
458 mnt = runner.SecretMounts[bind]
460 if bind == "stdout" || bind == "stderr" {
461 // Is it a "file" mount kind?
462 if mnt.Kind != "file" {
463 return nil, fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
466 // Does path start with OutputPath?
467 prefix := runner.Container.OutputPath
468 if !strings.HasSuffix(prefix, "/") {
471 if !strings.HasPrefix(mnt.Path, prefix) {
472 return nil, fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
477 // Is it a "collection" mount kind?
478 if mnt.Kind != "collection" && mnt.Kind != "json" {
479 return nil, fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
483 if bind == "/etc/arvados/ca-certificates.crt" {
484 needCertMount = false
487 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
488 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
489 return nil, fmt.Errorf("only mount points of kind 'collection', 'text' or 'json' are supported underneath the output_path for %q, was %q", bind, mnt.Kind)
494 case mnt.Kind == "collection" && bind != "stdin":
496 if mnt.UUID != "" && mnt.PortableDataHash != "" {
497 return nil, fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
501 return nil, fmt.Errorf("writing to existing collections currently not permitted")
504 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
505 } else if mnt.PortableDataHash != "" {
506 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
507 return nil, fmt.Errorf("can never write to a collection specified by portable data hash")
509 idx := strings.Index(mnt.PortableDataHash, "/")
511 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
512 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
513 runner.Container.Mounts[bind] = mnt
515 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
516 if mnt.Path != "" && mnt.Path != "." {
517 if strings.HasPrefix(mnt.Path, "./") {
518 mnt.Path = mnt.Path[2:]
519 } else if strings.HasPrefix(mnt.Path, "/") {
520 mnt.Path = mnt.Path[1:]
522 src += "/" + mnt.Path
525 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
526 arvMountCmd = append(arvMountCmd, "--mount-tmp")
527 arvMountCmd = append(arvMountCmd, fmt.Sprintf("tmp%d", tmpcount))
531 if bind == runner.Container.OutputPath {
532 runner.HostOutputDir = src
533 bindmounts[bind] = bindmount{HostPath: src}
534 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
535 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
537 bindmounts[bind] = bindmount{HostPath: src}
540 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
542 collectionPaths = append(collectionPaths, src)
544 case mnt.Kind == "tmp":
546 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
548 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
550 st, staterr := os.Stat(tmpdir)
552 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
554 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
556 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
558 bindmounts[bind] = bindmount{HostPath: tmpdir}
559 if bind == runner.Container.OutputPath {
560 runner.HostOutputDir = tmpdir
563 case mnt.Kind == "json" || mnt.Kind == "text":
565 if mnt.Kind == "json" {
566 filedata, err = json.Marshal(mnt.Content)
568 return nil, fmt.Errorf("encoding json data: %v", err)
571 text, ok := mnt.Content.(string)
573 return nil, fmt.Errorf("content for mount %q must be a string", bind)
575 filedata = []byte(text)
578 tmpdir, err := runner.MkTempDir(runner.parentTemp, mnt.Kind)
580 return nil, fmt.Errorf("creating temp dir: %v", err)
582 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
583 err = ioutil.WriteFile(tmpfn, filedata, 0444)
585 return nil, fmt.Errorf("writing temp file: %v", err)
587 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
588 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
590 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
593 case mnt.Kind == "git_tree":
594 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
596 return nil, fmt.Errorf("creating temp dir: %v", err)
598 err = gitMount(mnt).extractTree(runner.ContainerArvClient, tmpdir, token)
602 bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
606 if runner.HostOutputDir == "" {
607 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
610 if needCertMount && runner.Container.RuntimeConstraints.API {
611 for _, certfile := range arvadosclient.CertFiles {
612 _, err := os.Stat(certfile)
614 bindmounts["/etc/arvados/ca-certificates.crt"] = bindmount{HostPath: certfile, ReadOnly: true}
621 // If we are only mounting collections by pdh, make
622 // sure we don't subscribe to websocket events to
623 // avoid putting undesired load on the API server
624 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id", "--disable-event-listening")
626 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
628 // the by_uuid mount point is used by singularity when writing
629 // out docker images converted to SIF
630 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
631 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
633 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
635 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
638 for _, p := range collectionPaths {
641 return nil, fmt.Errorf("while checking that input files exist: %v", err)
645 for _, cp := range copyFiles {
646 st, err := os.Stat(cp.src)
648 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
651 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
655 target := path.Join(cp.bind, walkpath[len(cp.src):])
656 if walkinfo.Mode().IsRegular() {
657 copyerr := copyfile(walkpath, target)
661 return os.Chmod(target, walkinfo.Mode()|0777)
662 } else if walkinfo.Mode().IsDir() {
663 mkerr := os.MkdirAll(target, 0777)
667 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
669 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
672 } else if st.Mode().IsRegular() {
673 err = copyfile(cp.src, cp.bind)
675 err = os.Chmod(cp.bind, st.Mode()|0777)
679 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
683 return bindmounts, nil
686 func (runner *ContainerRunner) stopHoststat() error {
687 if runner.hoststatReporter == nil {
690 runner.hoststatReporter.Stop()
691 err := runner.hoststatLogger.Close()
693 return fmt.Errorf("error closing hoststat logs: %v", err)
698 func (runner *ContainerRunner) startHoststat() error {
699 w, err := runner.NewLogWriter("hoststat")
703 runner.hoststatLogger = NewThrottledLogger(w)
704 runner.hoststatReporter = &crunchstat.Reporter{
705 Logger: log.New(runner.hoststatLogger, "", 0),
706 CgroupRoot: runner.cgroupRoot,
707 PollPeriod: runner.statInterval,
709 runner.hoststatReporter.Start()
713 func (runner *ContainerRunner) startCrunchstat() error {
714 w, err := runner.NewLogWriter("crunchstat")
718 runner.statLogger = NewThrottledLogger(w)
719 runner.statReporter = &crunchstat.Reporter{
720 CID: runner.executor.CgroupID(),
721 Logger: log.New(runner.statLogger, "", 0),
722 CgroupParent: runner.expectCgroupParent,
723 CgroupRoot: runner.cgroupRoot,
724 PollPeriod: runner.statInterval,
725 TempDir: runner.parentTemp,
727 runner.statReporter.Start()
731 type infoCommand struct {
736 // LogHostInfo logs info about the current host, for debugging and
737 // accounting purposes. Although it's logged as "node-info", this is
738 // about the environment where crunch-run is actually running, which
739 // might differ from what's described in the node record (see
741 func (runner *ContainerRunner) LogHostInfo() (err error) {
742 w, err := runner.NewLogWriter("node-info")
747 commands := []infoCommand{
749 label: "Host Information",
750 cmd: []string{"uname", "-a"},
753 label: "CPU Information",
754 cmd: []string{"cat", "/proc/cpuinfo"},
757 label: "Memory Information",
758 cmd: []string{"cat", "/proc/meminfo"},
762 cmd: []string{"df", "-m", "/", os.TempDir()},
765 label: "Disk INodes",
766 cmd: []string{"df", "-i", "/", os.TempDir()},
770 // Run commands with informational output to be logged.
771 for _, command := range commands {
772 fmt.Fprintln(w, command.label)
773 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
776 if err := cmd.Run(); err != nil {
777 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
786 return fmt.Errorf("While closing node-info logs: %v", err)
791 // LogContainerRecord gets and saves the raw JSON container record from the API server
792 func (runner *ContainerRunner) LogContainerRecord() error {
793 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
794 if !logged && err == nil {
795 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
800 // LogNodeRecord logs the current host's InstanceType config entry (or
801 // the arvados#node record, if running via crunch-dispatch-slurm).
802 func (runner *ContainerRunner) LogNodeRecord() error {
803 if it := os.Getenv("InstanceType"); it != "" {
804 // Dispatched via arvados-dispatch-cloud. Save
805 // InstanceType config fragment received from
806 // dispatcher on stdin.
807 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
812 _, err = io.WriteString(w, it)
818 // Dispatched via crunch-dispatch-slurm. Look up
819 // apiserver's node record corresponding to
821 hostname := os.Getenv("SLURMD_NODENAME")
823 hostname, _ = os.Hostname()
825 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
826 // The "info" field has admin-only info when
827 // obtained with a privileged token, and
828 // should not be logged.
829 node, ok := resp.(map[string]interface{})
837 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
838 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
843 ArvClient: runner.DispatcherArvClient,
844 UUID: runner.Container.UUID,
845 loggingStream: label,
849 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
851 return false, fmt.Errorf("error getting %s record: %v", label, err)
855 dec := json.NewDecoder(reader)
857 var resp map[string]interface{}
858 if err = dec.Decode(&resp); err != nil {
859 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
861 items, ok := resp["items"].([]interface{})
863 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
864 } else if len(items) < 1 {
870 // Re-encode it using indentation to improve readability
871 enc := json.NewEncoder(w)
872 enc.SetIndent("", " ")
873 if err = enc.Encode(items[0]); err != nil {
874 return false, fmt.Errorf("error logging %s record: %v", label, err)
878 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
883 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
884 stdoutPath := mntPath[len(runner.Container.OutputPath):]
885 index := strings.LastIndex(stdoutPath, "/")
887 subdirs := stdoutPath[:index]
889 st, err := os.Stat(runner.HostOutputDir)
891 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
893 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
894 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
896 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
900 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
902 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
905 return stdoutFile, nil
908 // CreateContainer creates the docker container.
909 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
910 var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
911 if mnt, ok := runner.Container.Mounts["stdin"]; ok {
918 collID = mnt.PortableDataHash
920 path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
921 f, err := os.Open(path)
927 j, err := json.Marshal(mnt.Content)
929 return fmt.Errorf("error encoding stdin json data: %v", err)
931 stdin = ioutil.NopCloser(bytes.NewReader(j))
933 return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
937 var stdout, stderr io.WriteCloser
938 if mnt, ok := runner.Container.Mounts["stdout"]; ok {
939 f, err := runner.getStdoutFile(mnt.Path)
944 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
947 stdout = NewThrottledLogger(w)
950 if mnt, ok := runner.Container.Mounts["stderr"]; ok {
951 f, err := runner.getStdoutFile(mnt.Path)
956 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
959 stderr = NewThrottledLogger(w)
962 env := runner.Container.Environment
963 enableNetwork := runner.enableNetwork == "always"
964 if runner.Container.RuntimeConstraints.API {
966 tok, err := runner.ContainerToken()
970 env = map[string]string{}
971 for k, v := range runner.Container.Environment {
974 env["ARVADOS_API_TOKEN"] = tok
975 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
976 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
978 workdir := runner.Container.Cwd
980 // both "" and "." mean default
983 ram := runner.Container.RuntimeConstraints.RAM
984 if !runner.enableMemoryLimit {
987 runner.executorStdin = stdin
988 runner.executorStdout = stdout
989 runner.executorStderr = stderr
991 if runner.Container.RuntimeConstraints.CUDA.DeviceCount > 0 {
992 nvidiaModprobe(runner.CrunchLog)
995 return runner.executor.Create(containerSpec{
997 VCPUs: runner.Container.RuntimeConstraints.VCPUs,
1001 BindMounts: bindmounts,
1002 Command: runner.Container.Command,
1003 EnableNetwork: enableNetwork,
1004 CUDADeviceCount: runner.Container.RuntimeConstraints.CUDA.DeviceCount,
1005 NetworkMode: runner.networkMode,
1006 CgroupParent: runner.setCgroupParent,
1013 // StartContainer starts the docker container created by CreateContainer.
1014 func (runner *ContainerRunner) StartContainer() error {
1015 runner.CrunchLog.Printf("Starting container")
1016 runner.cStateLock.Lock()
1017 defer runner.cStateLock.Unlock()
1018 if runner.cCancelled {
1021 err := runner.executor.Start()
1024 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1025 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])
1027 return fmt.Errorf("could not start container: %v%s", err, advice)
1032 // WaitFinish waits for the container to terminate, capture the exit code, and
1033 // close the stdout/stderr logging.
1034 func (runner *ContainerRunner) WaitFinish() error {
1035 runner.CrunchLog.Print("Waiting for container to finish")
1036 var timeout <-chan time.Time
1037 if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1038 timeout = time.After(time.Duration(s) * time.Second)
1040 ctx, cancel := context.WithCancel(context.Background())
1045 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1047 case <-runner.ArvMountExit:
1048 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1053 exitcode, err := runner.executor.Wait(ctx)
1055 runner.checkBrokenNode(err)
1058 runner.ExitCode = &exitcode
1061 if exitcode&0x80 != 0 {
1062 // Convert raw exit status (0x80 + signal number) to a
1063 // string to log after the code, like " (signal 101)"
1064 // or " (signal 9, killed)"
1065 sig := syscall.WaitStatus(exitcode).Signal()
1066 if name := unix.SignalName(sig); name != "" {
1067 extra = fmt.Sprintf(" (signal %d, %s)", sig, name)
1069 extra = fmt.Sprintf(" (signal %d)", sig)
1072 runner.CrunchLog.Printf("Container exited with status code %d%s", exitcode, extra)
1075 if err = runner.executorStdin.Close(); err != nil {
1076 err = fmt.Errorf("error closing container stdin: %s", err)
1077 runner.CrunchLog.Printf("%s", err)
1080 if err = runner.executorStdout.Close(); err != nil {
1081 err = fmt.Errorf("error closing container stdout: %s", err)
1082 runner.CrunchLog.Printf("%s", err)
1083 if returnErr == nil {
1087 if err = runner.executorStderr.Close(); err != nil {
1088 err = fmt.Errorf("error closing container stderr: %s", err)
1089 runner.CrunchLog.Printf("%s", err)
1090 if returnErr == nil {
1095 if runner.statReporter != nil {
1096 runner.statReporter.Stop()
1097 err = runner.statLogger.Close()
1099 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1105 func (runner *ContainerRunner) updateLogs() {
1106 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1109 sigusr1 := make(chan os.Signal, 1)
1110 signal.Notify(sigusr1, syscall.SIGUSR1)
1111 defer signal.Stop(sigusr1)
1113 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1114 saveAtSize := crunchLogUpdateSize
1120 saveAtTime = time.Now()
1122 runner.logMtx.Lock()
1123 done := runner.LogsPDH != nil
1124 runner.logMtx.Unlock()
1128 size := runner.LogCollection.Size()
1129 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1132 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1133 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1134 saved, err := runner.saveLogCollection(false)
1136 runner.CrunchLog.Printf("error updating log collection: %s", err)
1140 var updated arvados.Container
1141 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1142 "container": arvadosclient.Dict{"log": saved.PortableDataHash},
1145 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1153 func (runner *ContainerRunner) reportArvMountWarning(pattern, text string) {
1154 var updated arvados.Container
1155 err := runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1156 "container": arvadosclient.Dict{
1157 "runtime_status": arvadosclient.Dict{
1158 "warning": "arv-mount: " + pattern,
1159 "warningDetail": text,
1164 runner.CrunchLog.Printf("error updating container runtime_status: %s", err)
1168 // CaptureOutput saves data from the container's output directory if
1169 // needed, and updates the container output accordingly.
1170 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1171 if runner.Container.RuntimeConstraints.API {
1172 // Output may have been set directly by the container, so
1173 // refresh the container record to check.
1174 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1175 nil, &runner.Container)
1179 if runner.Container.Output != "" {
1180 // Container output is already set.
1181 runner.OutputPDH = &runner.Container.Output
1186 txt, err := (&copier{
1187 client: runner.containerClient,
1188 arvClient: runner.ContainerArvClient,
1189 keepClient: runner.ContainerKeepClient,
1190 hostOutputDir: runner.HostOutputDir,
1191 ctrOutputDir: runner.Container.OutputPath,
1192 bindmounts: bindmounts,
1193 mounts: runner.Container.Mounts,
1194 secretMounts: runner.SecretMounts,
1195 logger: runner.CrunchLog,
1200 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1201 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1202 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1206 txt, err = fs.MarshalManifest(".")
1211 var resp arvados.Collection
1212 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1213 "ensure_unique_name": true,
1214 "collection": arvadosclient.Dict{
1216 "name": "output for " + runner.Container.UUID,
1217 "manifest_text": txt,
1221 return fmt.Errorf("error creating output collection: %v", err)
1223 runner.OutputPDH = &resp.PortableDataHash
1227 func (runner *ContainerRunner) CleanupDirs() {
1228 if runner.ArvMount != nil {
1230 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1231 umount.Stdout = runner.CrunchLog
1232 umount.Stderr = runner.CrunchLog
1233 runner.CrunchLog.Printf("Running %v", umount.Args)
1234 umnterr := umount.Start()
1237 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1238 runner.ArvMount.Process.Kill()
1240 // If arv-mount --unmount gets stuck for any reason, we
1241 // don't want to wait for it forever. Do Wait() in a goroutine
1242 // so it doesn't block crunch-run.
1243 umountExit := make(chan error)
1245 mnterr := umount.Wait()
1247 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1249 umountExit <- mnterr
1252 for again := true; again; {
1258 case <-runner.ArvMountExit:
1260 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1261 runner.CrunchLog.Printf("Timed out waiting for unmount")
1263 umount.Process.Kill()
1265 runner.ArvMount.Process.Kill()
1269 runner.ArvMount = nil
1272 if runner.ArvMountPoint != "" {
1273 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1274 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1276 runner.ArvMountPoint = ""
1279 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1280 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1284 // CommitLogs posts the collection containing the final container logs.
1285 func (runner *ContainerRunner) CommitLogs() error {
1287 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1288 runner.cStateLock.Lock()
1289 defer runner.cStateLock.Unlock()
1291 runner.CrunchLog.Print(runner.finalState)
1293 if runner.arvMountLog != nil {
1294 runner.arvMountLog.Close()
1296 runner.CrunchLog.Close()
1298 // Closing CrunchLog above allows them to be committed to Keep at this
1299 // point, but re-open crunch log with ArvClient in case there are any
1300 // other further errors (such as failing to write the log to Keep!)
1301 // while shutting down
1302 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1303 ArvClient: runner.DispatcherArvClient,
1304 UUID: runner.Container.UUID,
1305 loggingStream: "crunch-run",
1308 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1311 if runner.keepstoreLogger != nil {
1312 // Flush any buffered logs from our local keepstore
1313 // process. Discard anything logged after this point
1314 // -- it won't end up in the log collection, so
1315 // there's no point writing it to the collectionfs.
1316 runner.keepstoreLogbuf.SetWriter(io.Discard)
1317 runner.keepstoreLogger.Close()
1318 runner.keepstoreLogger = nil
1321 if runner.LogsPDH != nil {
1322 // If we have already assigned something to LogsPDH,
1323 // we must be closing the re-opened log, which won't
1324 // end up getting attached to the container record and
1325 // therefore doesn't need to be saved as a collection
1326 // -- it exists only to send logs to other channels.
1330 saved, err := runner.saveLogCollection(true)
1332 return fmt.Errorf("error saving log collection: %s", err)
1334 runner.logMtx.Lock()
1335 defer runner.logMtx.Unlock()
1336 runner.LogsPDH = &saved.PortableDataHash
1340 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1341 runner.logMtx.Lock()
1342 defer runner.logMtx.Unlock()
1343 if runner.LogsPDH != nil {
1344 // Already finalized.
1347 updates := arvadosclient.Dict{
1348 "name": "logs for " + runner.Container.UUID,
1350 mt, err1 := runner.LogCollection.MarshalManifest(".")
1352 // Only send updated manifest text if there was no
1354 updates["manifest_text"] = mt
1357 // Even if flushing the manifest had an error, we still want
1358 // to update the log record, if possible, to push the trash_at
1359 // and delete_at times into the future. Details on bug
1362 updates["is_trashed"] = true
1364 exp := time.Now().Add(crunchLogUpdatePeriod * 24)
1365 updates["trash_at"] = exp
1366 updates["delete_at"] = exp
1368 reqBody := arvadosclient.Dict{"collection": updates}
1370 if runner.logUUID == "" {
1371 reqBody["ensure_unique_name"] = true
1372 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1374 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1377 runner.logUUID = response.UUID
1380 if err1 != nil || err2 != nil {
1381 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1386 // UpdateContainerRunning updates the container state to "Running"
1387 func (runner *ContainerRunner) UpdateContainerRunning() error {
1388 runner.cStateLock.Lock()
1389 defer runner.cStateLock.Unlock()
1390 if runner.cCancelled {
1393 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID,
1394 arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running", "gateway_address": runner.gateway.Address}}, nil)
1397 // ContainerToken returns the api_token the container (and any
1398 // arv-mount processes) are allowed to use.
1399 func (runner *ContainerRunner) ContainerToken() (string, error) {
1400 if runner.token != "" {
1401 return runner.token, nil
1404 var auth arvados.APIClientAuthorization
1405 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1409 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1410 return runner.token, nil
1413 // UpdateContainerFinal updates the container record state on API
1414 // server to "Complete" or "Cancelled"
1415 func (runner *ContainerRunner) UpdateContainerFinal() error {
1416 update := arvadosclient.Dict{}
1417 update["state"] = runner.finalState
1418 if runner.LogsPDH != nil {
1419 update["log"] = *runner.LogsPDH
1421 if runner.finalState == "Complete" {
1422 if runner.ExitCode != nil {
1423 update["exit_code"] = *runner.ExitCode
1425 if runner.OutputPDH != nil {
1426 update["output"] = *runner.OutputPDH
1429 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil)
1432 // IsCancelled returns the value of Cancelled, with goroutine safety.
1433 func (runner *ContainerRunner) IsCancelled() bool {
1434 runner.cStateLock.Lock()
1435 defer runner.cStateLock.Unlock()
1436 return runner.cCancelled
1439 // NewArvLogWriter creates an ArvLogWriter
1440 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1441 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1445 return &ArvLogWriter{
1446 ArvClient: runner.DispatcherArvClient,
1447 UUID: runner.Container.UUID,
1448 loggingStream: name,
1449 writeCloser: writer,
1453 // Run the full container lifecycle.
1454 func (runner *ContainerRunner) Run() (err error) {
1455 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1456 runner.CrunchLog.Printf("Executing container '%s' using %s runtime", runner.Container.UUID, runner.executor.Runtime())
1458 hostname, hosterr := os.Hostname()
1460 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1462 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1465 runner.finalState = "Queued"
1468 runner.CleanupDirs()
1470 runner.CrunchLog.Printf("crunch-run finished")
1471 runner.CrunchLog.Close()
1474 err = runner.fetchContainerRecord()
1478 if runner.Container.State != "Locked" {
1479 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1482 var bindmounts map[string]bindmount
1484 // checkErr prints e (unless it's nil) and sets err to
1485 // e (unless err is already non-nil). Thus, if err
1486 // hasn't already been assigned when Run() returns,
1487 // this cleanup func will cause Run() to return the
1488 // first non-nil error that is passed to checkErr().
1489 checkErr := func(errorIn string, e error) {
1493 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1497 if runner.finalState == "Complete" {
1498 // There was an error in the finalization.
1499 runner.finalState = "Cancelled"
1503 // Log the error encountered in Run(), if any
1504 checkErr("Run", err)
1506 if runner.finalState == "Queued" {
1507 runner.UpdateContainerFinal()
1511 if runner.IsCancelled() {
1512 runner.finalState = "Cancelled"
1513 // but don't return yet -- we still want to
1514 // capture partial output and write logs
1517 if bindmounts != nil {
1518 checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1520 checkErr("stopHoststat", runner.stopHoststat())
1521 checkErr("CommitLogs", runner.CommitLogs())
1522 runner.CleanupDirs()
1523 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1526 runner.setupSignals()
1527 err = runner.startHoststat()
1532 // set up FUSE mount and binds
1533 bindmounts, err = runner.SetupMounts()
1535 runner.finalState = "Cancelled"
1536 err = fmt.Errorf("While setting up mounts: %v", err)
1540 // check for and/or load image
1541 imageID, err := runner.LoadImage()
1543 if !runner.checkBrokenNode(err) {
1544 // Failed to load image but not due to a "broken node"
1545 // condition, probably user error.
1546 runner.finalState = "Cancelled"
1548 err = fmt.Errorf("While loading container image: %v", err)
1552 err = runner.CreateContainer(imageID, bindmounts)
1556 err = runner.LogHostInfo()
1560 err = runner.LogNodeRecord()
1564 err = runner.LogContainerRecord()
1569 if runner.IsCancelled() {
1573 err = runner.UpdateContainerRunning()
1577 runner.finalState = "Cancelled"
1579 err = runner.startCrunchstat()
1584 err = runner.StartContainer()
1586 runner.checkBrokenNode(err)
1590 err = runner.WaitFinish()
1591 if err == nil && !runner.IsCancelled() {
1592 runner.finalState = "Complete"
1597 // Fetch the current container record (uuid = runner.Container.UUID)
1598 // into runner.Container.
1599 func (runner *ContainerRunner) fetchContainerRecord() error {
1600 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1602 return fmt.Errorf("error fetching container record: %v", err)
1604 defer reader.Close()
1606 dec := json.NewDecoder(reader)
1608 err = dec.Decode(&runner.Container)
1610 return fmt.Errorf("error decoding container record: %v", err)
1614 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1617 containerToken, err := runner.ContainerToken()
1619 return fmt.Errorf("error getting container token: %v", err)
1622 runner.ContainerArvClient, runner.ContainerKeepClient,
1623 runner.containerClient, err = runner.MkArvClient(containerToken)
1625 return fmt.Errorf("error creating container API client: %v", err)
1628 runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1629 runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1631 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1633 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1634 return fmt.Errorf("error fetching secret_mounts: %v", err)
1636 // ok && apierr.HttpStatusCode == 404, which means
1637 // secret_mounts isn't supported by this API server.
1639 runner.SecretMounts = sm.SecretMounts
1644 // NewContainerRunner creates a new container runner.
1645 func NewContainerRunner(dispatcherClient *arvados.Client,
1646 dispatcherArvClient IArvadosClient,
1647 dispatcherKeepClient IKeepClient,
1648 containerUUID string) (*ContainerRunner, error) {
1650 cr := &ContainerRunner{
1651 dispatcherClient: dispatcherClient,
1652 DispatcherArvClient: dispatcherArvClient,
1653 DispatcherKeepClient: dispatcherKeepClient,
1655 cr.NewLogWriter = cr.NewArvLogWriter
1656 cr.RunArvMount = cr.ArvMountCmd
1657 cr.MkTempDir = ioutil.TempDir
1658 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1659 cl, err := arvadosclient.MakeArvadosClient()
1661 return nil, nil, nil, err
1664 kc, err := keepclient.MakeKeepClient(cl)
1666 return nil, nil, nil, err
1668 c2 := arvados.NewClientFromEnv()
1669 c2.AuthToken = token
1670 return cl, kc, c2, nil
1673 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1677 cr.Container.UUID = containerUUID
1678 w, err := cr.NewLogWriter("crunch-run")
1682 cr.CrunchLog = NewThrottledLogger(w)
1683 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1685 loadLogThrottleParams(dispatcherArvClient)
1691 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1692 log := log.New(stderr, "", 0)
1693 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1694 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1695 cgroupRoot := flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree")
1696 cgroupParent := flags.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)")
1697 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container")
1698 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1699 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1700 stdinConfig := flags.Bool("stdin-config", false, "Load config and environment variables from JSON message on stdin")
1701 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1702 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1703 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes")
1704 enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1705 enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1706 networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1707 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1708 runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1709 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1711 ignoreDetachFlag := false
1712 if len(args) > 0 && args[0] == "-no-detach" {
1713 // This process was invoked by a parent process, which
1714 // has passed along its own arguments, including
1715 // -detach, after the leading -no-detach flag. Strip
1716 // the leading -no-detach flag (it's not recognized by
1717 // flags.Parse()) and ignore the -detach flag that
1720 ignoreDetachFlag = true
1723 if ok, code := cmd.ParseFlags(flags, prog, args, "container-uuid", stderr); !ok {
1725 } else if !*list && flags.NArg() != 1 {
1726 fmt.Fprintf(stderr, "missing required argument: container-uuid (try -help)\n")
1730 containerUUID := flags.Arg(0)
1733 case *detach && !ignoreDetachFlag:
1734 return Detach(containerUUID, prog, args, os.Stdin, os.Stdout, os.Stderr)
1736 return KillProcess(containerUUID, syscall.Signal(*kill), os.Stdout, os.Stderr)
1738 return ListProcesses(os.Stdout, os.Stderr)
1741 if len(containerUUID) != 27 {
1742 log.Printf("usage: %s [options] UUID", prog)
1748 err := json.NewDecoder(stdin).Decode(&conf)
1750 log.Printf("decode stdin: %s", err)
1753 for k, v := range conf.Env {
1754 err = os.Setenv(k, v)
1756 log.Printf("setenv(%q): %s", k, err)
1760 if conf.Cluster != nil {
1761 // ClusterID is missing from the JSON
1762 // representation, but we need it to generate
1763 // a valid config file for keepstore, so we
1764 // fill it using the container UUID prefix.
1765 conf.Cluster.ClusterID = containerUUID[:5]
1769 log.Printf("crunch-run %s started", cmd.Version.String())
1772 if *caCertsPath != "" {
1773 arvadosclient.CertFiles = []string{*caCertsPath}
1776 var keepstoreLogbuf bufThenWrite
1777 keepstore, err := startLocalKeepstore(conf, io.MultiWriter(&keepstoreLogbuf, stderr))
1782 if keepstore != nil {
1783 defer keepstore.Process.Kill()
1786 api, err := arvadosclient.MakeArvadosClient()
1788 log.Printf("%s: %v", containerUUID, err)
1793 kc, err := keepclient.MakeKeepClient(api)
1795 log.Printf("%s: %v", containerUUID, err)
1798 kc.BlockCache = &keepclient.BlockCache{MaxBlocks: 2}
1801 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
1807 if keepstore == nil {
1808 // Log explanation (if any) for why we're not running
1809 // a local keepstore.
1810 var buf bytes.Buffer
1811 keepstoreLogbuf.SetWriter(&buf)
1813 cr.CrunchLog.Printf("%s", strings.TrimSpace(buf.String()))
1815 } else if logWhat := conf.Cluster.Containers.LocalKeepLogsToContainerLog; logWhat == "none" {
1816 cr.CrunchLog.Printf("using local keepstore process (pid %d) at %s", keepstore.Process.Pid, os.Getenv("ARVADOS_KEEP_SERVICES"))
1817 keepstoreLogbuf.SetWriter(io.Discard)
1819 cr.CrunchLog.Printf("using local keepstore process (pid %d) at %s, writing logs to keepstore.txt in log collection", keepstore.Process.Pid, os.Getenv("ARVADOS_KEEP_SERVICES"))
1820 logwriter, err := cr.NewLogWriter("keepstore")
1825 cr.keepstoreLogger = NewThrottledLogger(logwriter)
1827 var writer io.WriteCloser = cr.keepstoreLogger
1828 if logWhat == "errors" {
1829 writer = &filterKeepstoreErrorsOnly{WriteCloser: writer}
1830 } else if logWhat != "all" {
1831 // should have been caught earlier by
1832 // dispatcher's config loader
1833 log.Printf("invalid value for Containers.LocalKeepLogsToContainerLog: %q", logWhat)
1836 err = keepstoreLogbuf.SetWriter(writer)
1841 cr.keepstoreLogbuf = &keepstoreLogbuf
1844 switch *runtimeEngine {
1846 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
1848 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
1850 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
1851 cr.CrunchLog.Close()
1855 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
1856 cr.checkBrokenNode(err)
1857 cr.CrunchLog.Close()
1860 defer cr.executor.Close()
1862 gwAuthSecret := os.Getenv("GatewayAuthSecret")
1863 os.Unsetenv("GatewayAuthSecret")
1864 if gwAuthSecret == "" {
1865 // not safe to run a gateway service without an auth
1867 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
1868 } else if gwListen := os.Getenv("GatewayAddress"); gwListen == "" {
1869 // dispatcher did not tell us which external IP
1870 // address to advertise --> no gateway service
1871 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAddress was not provided by dispatcher)")
1872 } else if de, ok := cr.executor.(*dockerExecutor); ok {
1873 cr.gateway = Gateway{
1875 AuthSecret: gwAuthSecret,
1876 ContainerUUID: containerUUID,
1877 DockerContainerID: &de.containerID,
1879 ContainerIPAddress: dockerContainerIPAddress(&de.containerID),
1881 err = cr.gateway.Start()
1883 log.Printf("error starting gateway server: %s", err)
1888 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
1890 log.Printf("%s: %v", containerUUID, tmperr)
1894 cr.parentTemp = parentTemp
1895 cr.statInterval = *statInterval
1896 cr.cgroupRoot = *cgroupRoot
1897 cr.expectCgroupParent = *cgroupParent
1898 cr.enableMemoryLimit = *enableMemoryLimit
1899 cr.enableNetwork = *enableNetwork
1900 cr.networkMode = *networkMode
1901 if *cgroupParentSubsystem != "" {
1902 p := findCgroup(*cgroupParentSubsystem)
1903 cr.setCgroupParent = p
1904 cr.expectCgroupParent = p
1909 if *memprofile != "" {
1910 f, err := os.Create(*memprofile)
1912 log.Printf("could not create memory profile: %s", err)
1914 runtime.GC() // get up-to-date statistics
1915 if err := pprof.WriteHeapProfile(f); err != nil {
1916 log.Printf("could not write memory profile: %s", err)
1918 closeerr := f.Close()
1919 if closeerr != nil {
1920 log.Printf("closing memprofile file: %s", err)
1925 log.Printf("%s: %v", containerUUID, runerr)
1931 func startLocalKeepstore(configData ConfigData, logbuf io.Writer) (*exec.Cmd, error) {
1932 if configData.Cluster == nil || configData.KeepBuffers < 1 {
1935 for uuid, vol := range configData.Cluster.Volumes {
1936 if len(vol.AccessViaHosts) > 0 {
1937 fmt.Fprintf(logbuf, "not starting a local keepstore process because a volume (%s) uses AccessViaHosts\n", uuid)
1940 if !vol.ReadOnly && vol.Replication < configData.Cluster.Collections.DefaultReplication {
1941 fmt.Fprintf(logbuf, "not starting a local keepstore process because a writable volume (%s) has replication less than Collections.DefaultReplication (%d < %d)\n", uuid, vol.Replication, configData.Cluster.Collections.DefaultReplication)
1946 // Rather than have an alternate way to tell keepstore how
1947 // many buffers to use when starting it this way, we just
1948 // modify the cluster configuration that we feed it on stdin.
1949 configData.Cluster.API.MaxKeepBlobBuffers = configData.KeepBuffers
1951 ln, err := net.Listen("tcp", "localhost:0")
1955 _, port, err := net.SplitHostPort(ln.Addr().String())
1961 url := "http://localhost:" + port
1963 fmt.Fprintf(logbuf, "starting keepstore on %s\n", url)
1965 var confJSON bytes.Buffer
1966 err = json.NewEncoder(&confJSON).Encode(arvados.Config{
1967 Clusters: map[string]arvados.Cluster{
1968 configData.Cluster.ClusterID: *configData.Cluster,
1974 cmd := exec.Command("/proc/self/exe", "keepstore", "-config=-")
1975 if target, err := os.Readlink(cmd.Path); err == nil && strings.HasSuffix(target, ".test") {
1976 // If we're a 'go test' process, running
1977 // /proc/self/exe would start the test suite in a
1978 // child process, which is not what we want.
1979 cmd.Path, _ = exec.LookPath("go")
1980 cmd.Args = append([]string{"go", "run", "../../cmd/arvados-server"}, cmd.Args[1:]...)
1981 cmd.Env = os.Environ()
1983 cmd.Stdin = &confJSON
1986 cmd.Env = append(cmd.Env,
1988 "ARVADOS_SERVICE_INTERNAL_URL="+url)
1991 return nil, fmt.Errorf("error starting keepstore process: %w", err)
1998 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(time.Second*10))
2000 poll := time.NewTicker(time.Second / 10)
2002 client := http.Client{}
2004 testReq, err := http.NewRequestWithContext(ctx, "GET", url+"/_health/ping", nil)
2005 testReq.Header.Set("Authorization", "Bearer "+configData.Cluster.ManagementToken)
2009 resp, err := client.Do(testReq)
2012 if resp.StatusCode == http.StatusOK {
2017 return nil, fmt.Errorf("keepstore child process exited")
2019 if ctx.Err() != nil {
2020 return nil, fmt.Errorf("timed out waiting for new keepstore process to report healthy")
2023 os.Setenv("ARVADOS_KEEP_SERVICES", url)