1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
35 "git.arvados.org/arvados.git/lib/cloud"
36 "git.arvados.org/arvados.git/lib/cmd"
37 "git.arvados.org/arvados.git/lib/config"
38 "git.arvados.org/arvados.git/lib/crunchstat"
39 "git.arvados.org/arvados.git/sdk/go/arvados"
40 "git.arvados.org/arvados.git/sdk/go/arvadosclient"
41 "git.arvados.org/arvados.git/sdk/go/ctxlog"
42 "git.arvados.org/arvados.git/sdk/go/keepclient"
43 "git.arvados.org/arvados.git/sdk/go/manifest"
44 "golang.org/x/sys/unix"
49 var arvadosCertPath = "/etc/arvados/ca-certificates.crt"
51 var Command = command{}
53 // ConfigData contains environment variables and (when needed) cluster
54 // configuration, passed from dispatchcloud to crunch-run on stdin.
55 type ConfigData struct {
59 Cluster *arvados.Cluster
62 // IArvadosClient is the minimal Arvados API methods used by crunch-run.
63 type IArvadosClient interface {
64 Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error
65 Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
66 Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
67 Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
68 CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
69 Discovery(key string) (interface{}, error)
72 // ErrCancelled is the error returned when the container is cancelled.
73 var ErrCancelled = errors.New("Cancelled")
75 // IKeepClient is the minimal Keep API methods used by crunch-run.
76 type IKeepClient interface {
77 BlockWrite(context.Context, arvados.BlockWriteOptions) (arvados.BlockWriteResponse, error)
78 ReadAt(locator string, p []byte, off int) (int, error)
79 ManifestFileReader(m manifest.Manifest, filename string) (arvados.File, error)
80 LocalLocator(locator string) (string, error)
81 SetStorageClasses(sc []string)
84 // NewLogWriter is a factory function to create a new log writer.
85 type NewLogWriter func(name string) (io.WriteCloser, error)
87 type RunArvMount func(cmdline []string, tok string) (*exec.Cmd, error)
89 type MkTempDir func(string, string) (string, 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 executor containerExecutor
99 executorStdin io.Closer
100 executorStdout io.Closer
101 executorStderr io.Closer
103 // Dispatcher client is initialized with the Dispatcher token.
104 // This is a privileged token used to manage container status
107 // We have both dispatcherClient and DispatcherArvClient
108 // because there are two different incompatible Arvados Go
109 // SDKs and we have to use both (hopefully this gets fixed in
111 dispatcherClient *arvados.Client
112 DispatcherArvClient IArvadosClient
113 DispatcherKeepClient IKeepClient
115 // Container client is initialized with the Container token
116 // This token controls the permissions of the container, and
117 // must be used for operations such as reading collections.
119 // Same comment as above applies to
120 // containerClient/ContainerArvClient.
121 containerClient *arvados.Client
122 ContainerArvClient IArvadosClient
123 ContainerKeepClient IKeepClient
125 Container arvados.Container
128 NewLogWriter NewLogWriter
129 CrunchLog *ThrottledLogger
132 LogCollection arvados.CollectionFileSystem
134 RunArvMount RunArvMount
139 Volumes map[string]struct{}
141 SigChan chan os.Signal
142 ArvMountExit chan error
143 SecretMounts map[string]arvados.Mount
144 MkArvClient func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error)
147 costStartTime time.Time
150 keepstoreLogger io.WriteCloser
151 keepstoreLogbuf *bufThenWrite
152 statLogger io.WriteCloser
153 statReporter *crunchstat.Reporter
154 hoststatLogger io.WriteCloser
155 hoststatReporter *crunchstat.Reporter
156 statInterval time.Duration
157 // What we tell docker to use as the container's cgroup
159 setCgroupParent string
160 // Fake root dir where crunchstat.Reporter should read OS
161 // files, for testing.
162 crunchstatFakeFS fs.FS
164 cStateLock sync.Mutex
165 cCancelled bool // StopContainer() invoked
167 enableMemoryLimit bool
168 enableNetwork string // one of "default" or "always"
169 networkMode string // "none", "host", or "" -- passed through to executor
170 brokenNodeHook string // script to run if node appears to be broken
171 arvMountLog *ThrottledLogger
173 containerWatchdogInterval time.Duration
177 prices []cloud.InstancePrice
178 pricesLock sync.Mutex
181 // setupSignals sets up signal handling to gracefully terminate the
182 // underlying container and update state when receiving a TERM, INT or
184 func (runner *ContainerRunner) setupSignals() {
185 runner.SigChan = make(chan os.Signal, 1)
186 signal.Notify(runner.SigChan, syscall.SIGTERM)
187 signal.Notify(runner.SigChan, syscall.SIGINT)
188 signal.Notify(runner.SigChan, syscall.SIGQUIT)
190 go func(sig chan os.Signal) {
197 // stop the underlying container.
198 func (runner *ContainerRunner) stop(sig os.Signal) {
199 runner.cStateLock.Lock()
200 defer runner.cStateLock.Unlock()
202 runner.CrunchLog.Printf("caught signal: %v", sig)
204 runner.cCancelled = true
205 runner.CrunchLog.Printf("stopping container")
206 err := runner.executor.Stop()
208 runner.CrunchLog.Printf("error stopping container: %s", err)
212 var errorBlacklist = []string{
213 "(?ms).*[Cc]annot connect to the Docker daemon.*",
214 "(?ms).*oci runtime error.*starting container process.*container init.*mounting.*to rootfs.*no such file or directory.*",
215 "(?ms).*grpc: the connection is unavailable.*",
218 func (runner *ContainerRunner) runBrokenNodeHook() {
219 if runner.brokenNodeHook == "" {
220 path := filepath.Join(lockdir, brokenfile)
221 runner.CrunchLog.Printf("Writing %s to mark node as broken", path)
222 f, err := os.OpenFile(path, os.O_CREATE|os.O_RDWR, 0700)
224 runner.CrunchLog.Printf("Error writing %s: %s", path, err)
229 runner.CrunchLog.Printf("Running broken node hook %q", runner.brokenNodeHook)
231 c := exec.Command(runner.brokenNodeHook)
232 c.Stdout = runner.CrunchLog
233 c.Stderr = runner.CrunchLog
236 runner.CrunchLog.Printf("Error running broken node hook: %v", err)
241 func (runner *ContainerRunner) checkBrokenNode(goterr error) bool {
242 for _, d := range errorBlacklist {
243 if m, e := regexp.MatchString(d, goterr.Error()); m && e == nil {
244 runner.CrunchLog.Printf("Error suggests node is unable to run containers: %v", goterr)
245 runner.runBrokenNodeHook()
252 // LoadImage determines the docker image id from the container record and
253 // checks if it is available in the local Docker image store. If not, it loads
254 // the image from Keep.
255 func (runner *ContainerRunner) LoadImage() (string, error) {
256 runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage)
258 d, err := os.Open(runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage)
263 allfiles, err := d.Readdirnames(-1)
267 var tarfiles []string
268 for _, fnm := range allfiles {
269 if strings.HasSuffix(fnm, ".tar") {
270 tarfiles = append(tarfiles, fnm)
273 if len(tarfiles) == 0 {
274 return "", fmt.Errorf("image collection does not include a .tar image file")
276 if len(tarfiles) > 1 {
277 return "", fmt.Errorf("cannot choose from multiple tar files in image collection: %v", tarfiles)
279 imageID := tarfiles[0][:len(tarfiles[0])-4]
280 imageTarballPath := runner.ArvMountPoint + "/by_id/" + runner.Container.ContainerImage + "/" + imageID + ".tar"
281 runner.CrunchLog.Printf("Using Docker image id %q", imageID)
283 runner.CrunchLog.Print("Loading Docker image from keep")
284 err = runner.executor.LoadImage(imageID, imageTarballPath, runner.Container, runner.ArvMountPoint,
285 runner.containerClient)
293 func (runner *ContainerRunner) ArvMountCmd(cmdline []string, token string) (c *exec.Cmd, err error) {
294 c = exec.Command(cmdline[0], cmdline[1:]...)
296 // Copy our environment, but override ARVADOS_API_TOKEN with
297 // the container auth token.
299 for _, s := range os.Environ() {
300 if !strings.HasPrefix(s, "ARVADOS_API_TOKEN=") {
301 c.Env = append(c.Env, s)
304 c.Env = append(c.Env, "ARVADOS_API_TOKEN="+token)
306 w, err := runner.NewLogWriter("arv-mount")
310 runner.arvMountLog = NewThrottledLogger(w)
311 scanner := logScanner{
314 "Block not found error",
315 "Unhandled exception during FUSE operation",
317 ReportFunc: func(pattern, text string) {
318 runner.updateRuntimeStatus(arvadosclient.Dict{
319 "warning": "arv-mount: " + pattern,
320 "warningDetail": text,
324 c.Stdout = runner.arvMountLog
325 c.Stderr = io.MultiWriter(runner.arvMountLog, os.Stderr, &scanner)
327 runner.CrunchLog.Printf("Running %v", c.Args)
334 statReadme := make(chan bool)
335 runner.ArvMountExit = make(chan error)
340 time.Sleep(100 * time.Millisecond)
341 _, err = os.Stat(fmt.Sprintf("%s/by_id/README", runner.ArvMountPoint))
353 runner.CrunchLog.Printf("Arv-mount exit error: %v", mnterr)
355 runner.ArvMountExit <- mnterr
356 close(runner.ArvMountExit)
362 case err := <-runner.ArvMountExit:
363 runner.ArvMount = nil
371 func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) {
372 if runner.ArvMountPoint == "" {
373 runner.ArvMountPoint, err = runner.MkTempDir(runner.parentTemp, prefix)
378 func copyfile(src string, dst string) (err error) {
379 srcfile, err := os.Open(src)
384 os.MkdirAll(path.Dir(dst), 0777)
386 dstfile, err := os.Create(dst)
390 _, err = io.Copy(dstfile, srcfile)
395 err = srcfile.Close()
396 err2 := dstfile.Close()
409 func (runner *ContainerRunner) SetupMounts() (map[string]bindmount, error) {
410 bindmounts := map[string]bindmount{}
411 err := runner.SetupArvMountPoint("keep")
413 return nil, fmt.Errorf("While creating keep mount temp dir: %v", err)
416 token, err := runner.ContainerToken()
418 return nil, fmt.Errorf("could not get container token: %s", err)
420 runner.CrunchLog.Printf("container token %q", token)
424 arvMountCmd := []string{
428 "--storage-classes", strings.Join(runner.Container.OutputStorageClasses, ","),
429 fmt.Sprintf("--crunchstat-interval=%v", runner.statInterval.Seconds())}
431 if _, isdocker := runner.executor.(*dockerExecutor); isdocker {
432 arvMountCmd = append(arvMountCmd, "--allow-other")
435 if runner.Container.RuntimeConstraints.KeepCacheDisk > 0 {
436 keepcachedir, err := runner.MkTempDir(runner.parentTemp, "keepcache")
438 return nil, fmt.Errorf("while creating keep cache temp dir: %v", err)
440 arvMountCmd = append(arvMountCmd, "--disk-cache", "--disk-cache-dir", keepcachedir, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheDisk))
441 } else if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 {
442 arvMountCmd = append(arvMountCmd, "--ram-cache", "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM))
445 collectionPaths := []string{}
446 needCertMount := true
447 type copyFile struct {
451 var copyFiles []copyFile
454 for bind := range runner.Container.Mounts {
455 binds = append(binds, bind)
457 for bind := range runner.SecretMounts {
458 if _, ok := runner.Container.Mounts[bind]; ok {
459 return nil, fmt.Errorf("secret mount %q conflicts with regular mount", bind)
461 if runner.SecretMounts[bind].Kind != "json" &&
462 runner.SecretMounts[bind].Kind != "text" {
463 return nil, fmt.Errorf("secret mount %q type is %q but only 'json' and 'text' are permitted",
464 bind, runner.SecretMounts[bind].Kind)
466 binds = append(binds, bind)
470 for _, bind := range binds {
471 mnt, notSecret := runner.Container.Mounts[bind]
473 mnt = runner.SecretMounts[bind]
475 if bind == "stdout" || bind == "stderr" {
476 // Is it a "file" mount kind?
477 if mnt.Kind != "file" {
478 return nil, fmt.Errorf("unsupported mount kind '%s' for %s: only 'file' is supported", mnt.Kind, bind)
481 // Does path start with OutputPath?
482 prefix := runner.Container.OutputPath
483 if !strings.HasSuffix(prefix, "/") {
486 if !strings.HasPrefix(mnt.Path, prefix) {
487 return nil, fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
492 // Is it a "collection" mount kind?
493 if mnt.Kind != "collection" && mnt.Kind != "json" {
494 return nil, fmt.Errorf("unsupported mount kind '%s' for stdin: only 'collection' and 'json' are supported", mnt.Kind)
498 if bind == arvadosCertPath {
499 needCertMount = false
502 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" {
503 if mnt.Kind != "collection" && mnt.Kind != "text" && mnt.Kind != "json" {
504 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)
509 case mnt.Kind == "collection" && bind != "stdin":
511 if mnt.UUID != "" && mnt.PortableDataHash != "" {
512 return nil, fmt.Errorf("cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
516 return nil, fmt.Errorf("writing to existing collections currently not permitted")
519 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.UUID)
520 } else if mnt.PortableDataHash != "" {
521 if mnt.Writable && !strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
522 return nil, fmt.Errorf("can never write to a collection specified by portable data hash")
524 idx := strings.Index(mnt.PortableDataHash, "/")
526 mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
527 mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
528 runner.Container.Mounts[bind] = mnt
530 src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
531 if mnt.Path != "" && mnt.Path != "." {
532 if strings.HasPrefix(mnt.Path, "./") {
533 mnt.Path = mnt.Path[2:]
534 } else if strings.HasPrefix(mnt.Path, "/") {
535 mnt.Path = mnt.Path[1:]
537 src += "/" + mnt.Path
540 src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
541 arvMountCmd = append(arvMountCmd, "--mount-tmp", fmt.Sprintf("tmp%d", tmpcount))
545 if bind == runner.Container.OutputPath {
546 runner.HostOutputDir = src
547 bindmounts[bind] = bindmount{HostPath: src}
548 } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") {
549 copyFiles = append(copyFiles, copyFile{src, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
551 bindmounts[bind] = bindmount{HostPath: src}
554 bindmounts[bind] = bindmount{HostPath: src, ReadOnly: true}
556 collectionPaths = append(collectionPaths, src)
558 case mnt.Kind == "tmp":
560 tmpdir, err = runner.MkTempDir(runner.parentTemp, "tmp")
562 return nil, fmt.Errorf("while creating mount temp dir: %v", err)
564 st, staterr := os.Stat(tmpdir)
566 return nil, fmt.Errorf("while Stat on temp dir: %v", staterr)
568 err = os.Chmod(tmpdir, st.Mode()|os.ModeSetgid|0777)
570 return nil, fmt.Errorf("while Chmod temp dir: %v", err)
572 bindmounts[bind] = bindmount{HostPath: tmpdir}
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 nil, fmt.Errorf("encoding json data: %v", err)
585 text, ok := mnt.Content.(string)
587 return nil, 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 nil, fmt.Errorf("creating temp dir: %v", err)
596 tmpfn := filepath.Join(tmpdir, "mountdata."+mnt.Kind)
597 err = ioutil.WriteFile(tmpfn, filedata, 0444)
599 return nil, fmt.Errorf("writing temp file: %v", err)
601 if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && (notSecret || runner.Container.Mounts[runner.Container.OutputPath].Kind != "collection") {
602 // In most cases, if the container
603 // specifies a literal file inside the
604 // output path, we copy it into the
605 // output directory (either a mounted
606 // collection or a staging area on the
607 // host fs). If it's a secret, it will
608 // be skipped when copying output from
609 // staging to Keep later.
610 copyFiles = append(copyFiles, copyFile{tmpfn, runner.HostOutputDir + bind[len(runner.Container.OutputPath):]})
612 // If a secret is outside OutputPath,
613 // we bind mount the secret file
614 // directly just like other mounts. We
615 // also use this strategy when a
616 // secret is inside OutputPath but
617 // OutputPath is a live collection, to
618 // avoid writing the secret to
619 // Keep. Attempting to remove a
620 // bind-mounted secret file from
621 // inside the container will return a
622 // "Device or resource busy" error
623 // that might not be handled well by
624 // the container, which is why we
625 // don't use this strategy when
626 // OutputPath is a staging directory.
627 bindmounts[bind] = bindmount{HostPath: tmpfn, ReadOnly: true}
630 case mnt.Kind == "git_tree":
631 tmpdir, err := runner.MkTempDir(runner.parentTemp, "git_tree")
633 return nil, fmt.Errorf("creating temp dir: %v", err)
635 err = gitMount(mnt).extractTree(runner.containerClient, tmpdir, token)
639 bindmounts[bind] = bindmount{HostPath: tmpdir, ReadOnly: true}
643 if runner.HostOutputDir == "" {
644 return nil, fmt.Errorf("output path does not correspond to a writable mount point")
647 if needCertMount && runner.Container.RuntimeConstraints.API {
648 for _, certfile := range []string{
649 // Populated by caller, or sdk/go/arvados init(), or test suite:
650 os.Getenv("SSL_CERT_FILE"),
651 // Copied from Go 1.21 stdlib (src/crypto/x509/root_linux.go):
652 "/etc/ssl/certs/ca-certificates.crt", // Debian/Ubuntu/Gentoo etc.
653 "/etc/pki/tls/certs/ca-bundle.crt", // Fedora/RHEL 6
654 "/etc/ssl/ca-bundle.pem", // OpenSUSE
655 "/etc/pki/tls/cacert.pem", // OpenELEC
656 "/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem", // CentOS/RHEL 7
657 "/etc/ssl/cert.pem", // Alpine Linux
659 if _, err := os.Stat(certfile); err == nil {
660 bindmounts[arvadosCertPath] = bindmount{HostPath: certfile, ReadOnly: true}
667 // If we are only mounting collections by pdh, make
668 // sure we don't subscribe to websocket events to
669 // avoid putting undesired load on the API server
670 arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id", "--disable-event-listening")
672 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_id")
674 // the by_uuid mount point is used by singularity when writing
675 // out docker images converted to SIF
676 arvMountCmd = append(arvMountCmd, "--mount-by-id", "by_uuid")
677 arvMountCmd = append(arvMountCmd, runner.ArvMountPoint)
679 runner.ArvMount, err = runner.RunArvMount(arvMountCmd, token)
681 return nil, fmt.Errorf("while trying to start arv-mount: %v", err)
683 if runner.hoststatReporter != nil && runner.ArvMount != nil {
684 runner.hoststatReporter.ReportPID("arv-mount", runner.ArvMount.Process.Pid)
687 for _, p := range collectionPaths {
690 return nil, fmt.Errorf("while checking that input files exist: %v", err)
694 for _, cp := range copyFiles {
695 st, err := os.Stat(cp.src)
697 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
700 err = filepath.Walk(cp.src, func(walkpath string, walkinfo os.FileInfo, walkerr error) error {
704 target := path.Join(cp.bind, walkpath[len(cp.src):])
705 if walkinfo.Mode().IsRegular() {
706 copyerr := copyfile(walkpath, target)
710 return os.Chmod(target, walkinfo.Mode()|0777)
711 } else if walkinfo.Mode().IsDir() {
712 mkerr := os.MkdirAll(target, 0777)
716 return os.Chmod(target, walkinfo.Mode()|os.ModeSetgid|0777)
718 return fmt.Errorf("source %q is not a regular file or directory", cp.src)
721 } else if st.Mode().IsRegular() {
722 err = copyfile(cp.src, cp.bind)
724 err = os.Chmod(cp.bind, st.Mode()|0777)
728 return nil, fmt.Errorf("while staging writable file from %q to %q: %v", cp.src, cp.bind, err)
732 return bindmounts, nil
735 func (runner *ContainerRunner) stopHoststat() error {
736 if runner.hoststatReporter == nil {
739 runner.hoststatReporter.Stop()
740 runner.hoststatReporter.LogProcessMemMax(runner.CrunchLog)
741 err := runner.hoststatLogger.Close()
743 return fmt.Errorf("error closing hoststat logs: %v", err)
748 func (runner *ContainerRunner) startHoststat() error {
749 w, err := runner.NewLogWriter("hoststat")
753 runner.hoststatLogger = NewThrottledLogger(w)
754 runner.hoststatReporter = &crunchstat.Reporter{
755 Logger: log.New(runner.hoststatLogger, "", 0),
756 // Our own cgroup is the "host" cgroup, in the sense
757 // that it accounts for resource usage outside the
758 // container. It doesn't count _all_ resource usage on
761 // TODO?: Use the furthest ancestor of our own cgroup
762 // that has stats available. (Currently crunchstat
763 // does not have that capability.)
765 PollPeriod: runner.statInterval,
767 runner.hoststatReporter.Start()
768 runner.hoststatReporter.ReportPID("crunch-run", os.Getpid())
772 func (runner *ContainerRunner) startCrunchstat() error {
773 w, err := runner.NewLogWriter("crunchstat")
777 runner.statLogger = NewThrottledLogger(w)
778 runner.statReporter = &crunchstat.Reporter{
779 Pid: runner.executor.Pid,
780 FS: runner.crunchstatFakeFS,
781 Logger: log.New(runner.statLogger, "", 0),
782 MemThresholds: map[string][]crunchstat.Threshold{
783 "rss": crunchstat.NewThresholdsFromPercentages(runner.Container.RuntimeConstraints.RAM, []int64{90, 95, 99}),
785 PollPeriod: runner.statInterval,
786 TempDir: runner.parentTemp,
787 ThresholdLogger: runner.CrunchLog,
789 runner.statReporter.Start()
793 type infoCommand struct {
798 // LogHostInfo logs info about the current host, for debugging and
799 // accounting purposes. Although it's logged as "node-info", this is
800 // about the environment where crunch-run is actually running, which
801 // might differ from what's described in the node record (see
803 func (runner *ContainerRunner) LogHostInfo() (err error) {
804 w, err := runner.NewLogWriter("node-info")
809 commands := []infoCommand{
811 label: "Host Information",
812 cmd: []string{"uname", "-a"},
815 label: "CPU Information",
816 cmd: []string{"cat", "/proc/cpuinfo"},
819 label: "Memory Information",
820 cmd: []string{"cat", "/proc/meminfo"},
824 cmd: []string{"df", "-m", "/", os.TempDir()},
827 label: "Disk INodes",
828 cmd: []string{"df", "-i", "/", os.TempDir()},
832 // Run commands with informational output to be logged.
833 for _, command := range commands {
834 fmt.Fprintln(w, command.label)
835 cmd := exec.Command(command.cmd[0], command.cmd[1:]...)
838 if err := cmd.Run(); err != nil {
839 err = fmt.Errorf("While running command %q: %v", command.cmd, err)
848 return fmt.Errorf("While closing node-info logs: %v", err)
853 // LogContainerRecord gets and saves the raw JSON container record from the API server
854 func (runner *ContainerRunner) LogContainerRecord() error {
855 logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
856 if !logged && err == nil {
857 err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
862 // LogNodeRecord logs the current host's InstanceType config entry (or
863 // the arvados#node record, if running via crunch-dispatch-slurm).
864 func (runner *ContainerRunner) LogNodeRecord() error {
865 if it := os.Getenv("InstanceType"); it != "" {
866 // Dispatched via arvados-dispatch-cloud. Save
867 // InstanceType config fragment received from
868 // dispatcher on stdin.
869 w, err := runner.LogCollection.OpenFile("node.json", os.O_CREATE|os.O_WRONLY, 0666)
874 _, err = io.WriteString(w, it)
880 // Dispatched via crunch-dispatch-slurm. Look up
881 // apiserver's node record corresponding to
883 hostname := os.Getenv("SLURMD_NODENAME")
885 hostname, _ = os.Hostname()
887 _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
888 // The "info" field has admin-only info when
889 // obtained with a privileged token, and
890 // should not be logged.
891 node, ok := resp.(map[string]interface{})
899 func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
900 writer, err := runner.LogCollection.OpenFile(label+".json", os.O_CREATE|os.O_WRONLY, 0666)
905 ArvClient: runner.DispatcherArvClient,
906 UUID: runner.Container.UUID,
907 loggingStream: label,
911 reader, err := runner.DispatcherArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
913 return false, fmt.Errorf("error getting %s record: %v", label, err)
917 dec := json.NewDecoder(reader)
919 var resp map[string]interface{}
920 if err = dec.Decode(&resp); err != nil {
921 return false, fmt.Errorf("error decoding %s list response: %v", label, err)
923 items, ok := resp["items"].([]interface{})
925 return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
926 } else if len(items) < 1 {
932 // Re-encode it using indentation to improve readability
933 enc := json.NewEncoder(w)
934 enc.SetIndent("", " ")
935 if err = enc.Encode(items[0]); err != nil {
936 return false, fmt.Errorf("error logging %s record: %v", label, err)
940 return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
945 func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
946 stdoutPath := mntPath[len(runner.Container.OutputPath):]
947 index := strings.LastIndex(stdoutPath, "/")
949 subdirs := stdoutPath[:index]
951 st, err := os.Stat(runner.HostOutputDir)
953 return nil, fmt.Errorf("While Stat on temp dir: %v", err)
955 stdoutPath := filepath.Join(runner.HostOutputDir, subdirs)
956 err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
958 return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
962 stdoutFile, err := os.Create(filepath.Join(runner.HostOutputDir, stdoutPath))
964 return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
967 return stdoutFile, nil
970 // CreateContainer creates the docker container.
971 func (runner *ContainerRunner) CreateContainer(imageID string, bindmounts map[string]bindmount) error {
972 var stdin io.ReadCloser = ioutil.NopCloser(bytes.NewReader(nil))
973 if mnt, ok := runner.Container.Mounts["stdin"]; ok {
980 collID = mnt.PortableDataHash
982 path := runner.ArvMountPoint + "/by_id/" + collID + "/" + mnt.Path
983 f, err := os.Open(path)
989 j, err := json.Marshal(mnt.Content)
991 return fmt.Errorf("error encoding stdin json data: %v", err)
993 stdin = ioutil.NopCloser(bytes.NewReader(j))
995 return fmt.Errorf("stdin mount has unsupported kind %q", mnt.Kind)
999 var stdout, stderr io.WriteCloser
1000 if mnt, ok := runner.Container.Mounts["stdout"]; ok {
1001 f, err := runner.getStdoutFile(mnt.Path)
1006 } else if w, err := runner.NewLogWriter("stdout"); err != nil {
1009 stdout = NewThrottledLogger(w)
1012 if mnt, ok := runner.Container.Mounts["stderr"]; ok {
1013 f, err := runner.getStdoutFile(mnt.Path)
1018 } else if w, err := runner.NewLogWriter("stderr"); err != nil {
1021 stderr = NewThrottledLogger(w)
1024 env := runner.Container.Environment
1025 enableNetwork := runner.enableNetwork == "always"
1026 if runner.Container.RuntimeConstraints.API {
1027 enableNetwork = true
1028 tok, err := runner.ContainerToken()
1032 env = map[string]string{}
1033 for k, v := range runner.Container.Environment {
1036 env["ARVADOS_API_TOKEN"] = tok
1037 env["ARVADOS_API_HOST"] = os.Getenv("ARVADOS_API_HOST")
1038 env["ARVADOS_API_HOST_INSECURE"] = os.Getenv("ARVADOS_API_HOST_INSECURE")
1039 env["ARVADOS_KEEP_SERVICES"] = os.Getenv("ARVADOS_KEEP_SERVICES")
1041 workdir := runner.Container.Cwd
1043 // both "" and "." mean default
1046 ram := runner.Container.RuntimeConstraints.RAM
1047 if !runner.enableMemoryLimit {
1050 runner.executorStdin = stdin
1051 runner.executorStdout = stdout
1052 runner.executorStderr = stderr
1054 if runner.Container.RuntimeConstraints.CUDA.DeviceCount > 0 {
1055 nvidiaModprobe(runner.CrunchLog)
1058 return runner.executor.Create(containerSpec{
1060 VCPUs: runner.Container.RuntimeConstraints.VCPUs,
1062 WorkingDir: workdir,
1064 BindMounts: bindmounts,
1065 Command: runner.Container.Command,
1066 EnableNetwork: enableNetwork,
1067 CUDADeviceCount: runner.Container.RuntimeConstraints.CUDA.DeviceCount,
1068 NetworkMode: runner.networkMode,
1069 CgroupParent: runner.setCgroupParent,
1076 // StartContainer starts the docker container created by CreateContainer.
1077 func (runner *ContainerRunner) StartContainer() error {
1078 runner.CrunchLog.Printf("Starting container")
1079 runner.cStateLock.Lock()
1080 defer runner.cStateLock.Unlock()
1081 if runner.cCancelled {
1084 err := runner.executor.Start()
1087 if m, e := regexp.MatchString("(?ms).*(exec|System error).*(no such file or directory|file not found).*", err.Error()); m && e == nil {
1088 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])
1090 return fmt.Errorf("could not start container: %v%s", err, advice)
1095 // WaitFinish waits for the container to terminate, capture the exit code, and
1096 // close the stdout/stderr logging.
1097 func (runner *ContainerRunner) WaitFinish() error {
1098 runner.CrunchLog.Print("Waiting for container to finish")
1099 var timeout <-chan time.Time
1100 if s := runner.Container.SchedulingParameters.MaxRunTime; s > 0 {
1101 timeout = time.After(time.Duration(s) * time.Second)
1103 ctx, cancel := context.WithCancel(context.Background())
1108 runner.CrunchLog.Printf("maximum run time exceeded. Stopping container.")
1110 case <-runner.ArvMountExit:
1111 runner.CrunchLog.Printf("arv-mount exited while container is still running. Stopping container.")
1116 exitcode, err := runner.executor.Wait(ctx)
1118 runner.checkBrokenNode(err)
1121 runner.ExitCode = &exitcode
1124 if exitcode&0x80 != 0 {
1125 // Convert raw exit status (0x80 + signal number) to a
1126 // string to log after the code, like " (signal 101)"
1127 // or " (signal 9, killed)"
1128 sig := syscall.WaitStatus(exitcode).Signal()
1129 if name := unix.SignalName(sig); name != "" {
1130 extra = fmt.Sprintf(" (signal %d, %s)", sig, name)
1132 extra = fmt.Sprintf(" (signal %d)", sig)
1135 runner.CrunchLog.Printf("Container exited with status code %d%s", exitcode, extra)
1136 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1137 "select": []string{"uuid"},
1138 "container": arvadosclient.Dict{"exit_code": exitcode},
1141 runner.CrunchLog.Printf("ignoring error updating exit_code: %s", err)
1145 if err = runner.executorStdin.Close(); err != nil {
1146 err = fmt.Errorf("error closing container stdin: %s", err)
1147 runner.CrunchLog.Printf("%s", err)
1150 if err = runner.executorStdout.Close(); err != nil {
1151 err = fmt.Errorf("error closing container stdout: %s", err)
1152 runner.CrunchLog.Printf("%s", err)
1153 if returnErr == nil {
1157 if err = runner.executorStderr.Close(); err != nil {
1158 err = fmt.Errorf("error closing container stderr: %s", err)
1159 runner.CrunchLog.Printf("%s", err)
1160 if returnErr == nil {
1165 if runner.statReporter != nil {
1166 runner.statReporter.Stop()
1167 runner.statReporter.LogMaxima(runner.CrunchLog, map[string]int64{
1168 "rss": runner.Container.RuntimeConstraints.RAM,
1170 err = runner.statLogger.Close()
1172 runner.CrunchLog.Printf("error closing crunchstat logs: %v", err)
1178 func (runner *ContainerRunner) updateLogs() {
1179 ticker := time.NewTicker(crunchLogUpdatePeriod / 360)
1182 sigusr1 := make(chan os.Signal, 1)
1183 signal.Notify(sigusr1, syscall.SIGUSR1)
1184 defer signal.Stop(sigusr1)
1186 saveAtTime := time.Now().Add(crunchLogUpdatePeriod)
1187 saveAtSize := crunchLogUpdateSize
1193 saveAtTime = time.Now()
1195 runner.logMtx.Lock()
1196 done := runner.LogsPDH != nil
1197 runner.logMtx.Unlock()
1201 size := runner.LogCollection.Size()
1202 if size == savedSize || (time.Now().Before(saveAtTime) && size < saveAtSize) {
1205 saveAtTime = time.Now().Add(crunchLogUpdatePeriod)
1206 saveAtSize = runner.LogCollection.Size() + crunchLogUpdateSize
1207 saved, err := runner.saveLogCollection(false)
1209 runner.CrunchLog.Printf("error updating log collection: %s", err)
1213 err = runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1214 "select": []string{"uuid"},
1215 "container": arvadosclient.Dict{
1216 "log": saved.PortableDataHash,
1220 runner.CrunchLog.Printf("error updating container log to %s: %s", saved.PortableDataHash, err)
1228 var spotInterruptionCheckInterval = 5 * time.Second
1229 var ec2MetadataBaseURL = "http://169.254.169.254"
1231 const ec2TokenTTL = time.Second * 21600
1233 func (runner *ContainerRunner) checkSpotInterruptionNotices() {
1234 type ec2metadata struct {
1235 Action string `json:"action"`
1236 Time time.Time `json:"time"`
1238 runner.CrunchLog.Printf("Checking for spot interruptions every %v using instance metadata at %s", spotInterruptionCheckInterval, ec2MetadataBaseURL)
1239 var metadata ec2metadata
1241 var tokenExp time.Time
1242 check := func() error {
1243 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(time.Minute))
1245 if token == "" || tokenExp.Sub(time.Now()) < time.Minute {
1246 req, err := http.NewRequestWithContext(ctx, http.MethodPut, ec2MetadataBaseURL+"/latest/api/token", nil)
1250 req.Header.Set("X-aws-ec2-metadata-token-ttl-seconds", fmt.Sprintf("%d", int(ec2TokenTTL/time.Second)))
1251 resp, err := http.DefaultClient.Do(req)
1255 defer resp.Body.Close()
1256 if resp.StatusCode != http.StatusOK {
1257 return fmt.Errorf("%s", resp.Status)
1259 newtoken, err := ioutil.ReadAll(resp.Body)
1263 token = strings.TrimSpace(string(newtoken))
1264 tokenExp = time.Now().Add(ec2TokenTTL)
1266 req, err := http.NewRequestWithContext(ctx, http.MethodGet, ec2MetadataBaseURL+"/latest/meta-data/spot/instance-action", nil)
1270 req.Header.Set("X-aws-ec2-metadata-token", token)
1271 resp, err := http.DefaultClient.Do(req)
1275 defer resp.Body.Close()
1276 metadata = ec2metadata{}
1277 switch resp.StatusCode {
1280 case http.StatusNotFound:
1281 // "If Amazon EC2 is not preparing to stop or
1282 // terminate the instance, or if you
1283 // terminated the instance yourself,
1284 // instance-action is not present in the
1285 // instance metadata and you receive an HTTP
1286 // 404 error when you try to retrieve it."
1288 case http.StatusUnauthorized:
1290 return fmt.Errorf("%s", resp.Status)
1292 return fmt.Errorf("%s", resp.Status)
1294 err = json.NewDecoder(resp.Body).Decode(&metadata)
1301 var lastmetadata ec2metadata
1302 for range time.NewTicker(spotInterruptionCheckInterval).C {
1305 runner.CrunchLog.Printf("Error checking spot interruptions: %s", err)
1308 runner.CrunchLog.Printf("Giving up on checking spot interruptions after too many consecutive failures")
1314 if metadata != lastmetadata {
1315 lastmetadata = metadata
1316 text := fmt.Sprintf("Cloud provider scheduled instance %s at %s", metadata.Action, metadata.Time.UTC().Format(time.RFC3339))
1317 runner.CrunchLog.Printf("%s", text)
1318 runner.updateRuntimeStatus(arvadosclient.Dict{
1319 "warning": "preemption notice",
1320 "warningDetail": text,
1321 "preemptionNotice": text,
1323 if proc, err := os.FindProcess(os.Getpid()); err == nil {
1324 // trigger updateLogs
1325 proc.Signal(syscall.SIGUSR1)
1331 func (runner *ContainerRunner) updateRuntimeStatus(status arvadosclient.Dict) {
1332 err := runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1333 "select": []string{"uuid"},
1334 "container": arvadosclient.Dict{
1335 "runtime_status": status,
1339 runner.CrunchLog.Printf("error updating container runtime_status: %s", err)
1343 // CaptureOutput saves data from the container's output directory if
1344 // needed, and updates the container output accordingly.
1345 func (runner *ContainerRunner) CaptureOutput(bindmounts map[string]bindmount) error {
1346 if runner.Container.RuntimeConstraints.API {
1347 // Output may have been set directly by the container, so
1348 // refresh the container record to check.
1349 err := runner.DispatcherArvClient.Get("containers", runner.Container.UUID,
1351 "select": []string{"output"},
1352 }, &runner.Container)
1356 if runner.Container.Output != "" {
1357 // Container output is already set.
1358 runner.OutputPDH = &runner.Container.Output
1363 txt, err := (&copier{
1364 client: runner.containerClient,
1365 keepClient: runner.ContainerKeepClient,
1366 hostOutputDir: runner.HostOutputDir,
1367 ctrOutputDir: runner.Container.OutputPath,
1368 bindmounts: bindmounts,
1369 mounts: runner.Container.Mounts,
1370 secretMounts: runner.SecretMounts,
1371 logger: runner.CrunchLog,
1376 if n := len(regexp.MustCompile(` [0-9a-f]+\+\S*\+R`).FindAllStringIndex(txt, -1)); n > 0 {
1377 runner.CrunchLog.Printf("Copying %d data blocks from remote input collections...", n)
1378 fs, err := (&arvados.Collection{ManifestText: txt}).FileSystem(runner.containerClient, runner.ContainerKeepClient)
1382 txt, err = fs.MarshalManifest(".")
1387 var resp arvados.Collection
1388 err = runner.ContainerArvClient.Create("collections", arvadosclient.Dict{
1389 "ensure_unique_name": true,
1390 "select": []string{"portable_data_hash"},
1391 "collection": arvadosclient.Dict{
1393 "name": "output for " + runner.Container.UUID,
1394 "manifest_text": txt,
1398 return fmt.Errorf("error creating output collection: %v", err)
1400 runner.OutputPDH = &resp.PortableDataHash
1404 func (runner *ContainerRunner) CleanupDirs() {
1405 if runner.ArvMount != nil {
1407 umount := exec.Command("arv-mount", fmt.Sprintf("--unmount-timeout=%d", delay), "--unmount", runner.ArvMountPoint)
1408 umount.Stdout = runner.CrunchLog
1409 umount.Stderr = runner.CrunchLog
1410 runner.CrunchLog.Printf("Running %v", umount.Args)
1411 umnterr := umount.Start()
1414 runner.CrunchLog.Printf("Error unmounting: %v", umnterr)
1415 runner.ArvMount.Process.Kill()
1417 // If arv-mount --unmount gets stuck for any reason, we
1418 // don't want to wait for it forever. Do Wait() in a goroutine
1419 // so it doesn't block crunch-run.
1420 umountExit := make(chan error)
1422 mnterr := umount.Wait()
1424 runner.CrunchLog.Printf("Error unmounting: %v", mnterr)
1426 umountExit <- mnterr
1429 for again := true; again; {
1435 case <-runner.ArvMountExit:
1437 case <-time.After(time.Duration((delay + 1) * int64(time.Second))):
1438 runner.CrunchLog.Printf("Timed out waiting for unmount")
1440 umount.Process.Kill()
1442 runner.ArvMount.Process.Kill()
1446 runner.ArvMount = nil
1449 if runner.ArvMountPoint != "" {
1450 if rmerr := os.Remove(runner.ArvMountPoint); rmerr != nil {
1451 runner.CrunchLog.Printf("While cleaning up arv-mount directory %s: %v", runner.ArvMountPoint, rmerr)
1453 runner.ArvMountPoint = ""
1456 if rmerr := os.RemoveAll(runner.parentTemp); rmerr != nil {
1457 runner.CrunchLog.Printf("While cleaning up temporary directory %s: %v", runner.parentTemp, rmerr)
1461 // CommitLogs posts the collection containing the final container logs.
1462 func (runner *ContainerRunner) CommitLogs() error {
1464 // Hold cStateLock to prevent races on CrunchLog (e.g., stop()).
1465 runner.cStateLock.Lock()
1466 defer runner.cStateLock.Unlock()
1468 runner.CrunchLog.Print(runner.finalState)
1470 if runner.arvMountLog != nil {
1471 runner.arvMountLog.Close()
1473 runner.CrunchLog.Close()
1475 // Closing CrunchLog above allows them to be committed to Keep at this
1476 // point, but re-open crunch log with ArvClient in case there are any
1477 // other further errors (such as failing to write the log to Keep!)
1478 // while shutting down
1479 runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{
1480 ArvClient: runner.DispatcherArvClient,
1481 UUID: runner.Container.UUID,
1482 loggingStream: "crunch-run",
1485 runner.CrunchLog.Immediate = log.New(os.Stderr, runner.Container.UUID+" ", 0)
1488 if runner.keepstoreLogger != nil {
1489 // Flush any buffered logs from our local keepstore
1490 // process. Discard anything logged after this point
1491 // -- it won't end up in the log collection, so
1492 // there's no point writing it to the collectionfs.
1493 runner.keepstoreLogbuf.SetWriter(io.Discard)
1494 runner.keepstoreLogger.Close()
1495 runner.keepstoreLogger = nil
1498 if runner.LogsPDH != nil {
1499 // If we have already assigned something to LogsPDH,
1500 // we must be closing the re-opened log, which won't
1501 // end up getting attached to the container record and
1502 // therefore doesn't need to be saved as a collection
1503 // -- it exists only to send logs to other channels.
1507 saved, err := runner.saveLogCollection(true)
1509 return fmt.Errorf("error saving log collection: %s", err)
1511 runner.logMtx.Lock()
1512 defer runner.logMtx.Unlock()
1513 runner.LogsPDH = &saved.PortableDataHash
1517 // Create/update the log collection. Return value has UUID and
1518 // PortableDataHash fields populated, but others may be blank.
1519 func (runner *ContainerRunner) saveLogCollection(final bool) (response arvados.Collection, err error) {
1520 runner.logMtx.Lock()
1521 defer runner.logMtx.Unlock()
1522 if runner.LogsPDH != nil {
1523 // Already finalized.
1526 updates := arvadosclient.Dict{
1527 "name": "logs for " + runner.Container.UUID,
1529 mt, err1 := runner.LogCollection.MarshalManifest(".")
1531 // Only send updated manifest text if there was no
1533 updates["manifest_text"] = mt
1536 // Even if flushing the manifest had an error, we still want
1537 // to update the log record, if possible, to push the trash_at
1538 // and delete_at times into the future. Details on bug
1541 updates["is_trashed"] = true
1543 // We set trash_at so this collection gets
1544 // automatically cleaned up eventually. It used to be
1545 // 12 hours but we had a situation where the API
1546 // server was down over a weekend but the containers
1547 // kept running such that the log collection got
1548 // trashed, so now we make it 2 weeks. refs #20378
1549 exp := time.Now().Add(time.Duration(24*14) * time.Hour)
1550 updates["trash_at"] = exp
1551 updates["delete_at"] = exp
1553 reqBody := arvadosclient.Dict{
1554 "select": []string{"uuid", "portable_data_hash"},
1555 "collection": updates,
1558 if runner.logUUID == "" {
1559 reqBody["ensure_unique_name"] = true
1560 err2 = runner.DispatcherArvClient.Create("collections", reqBody, &response)
1562 err2 = runner.DispatcherArvClient.Update("collections", runner.logUUID, reqBody, &response)
1565 runner.logUUID = response.UUID
1568 if err1 != nil || err2 != nil {
1569 err = fmt.Errorf("error recording logs: %q, %q", err1, err2)
1574 // UpdateContainerRunning updates the container state to "Running"
1575 func (runner *ContainerRunner) UpdateContainerRunning(logId string) error {
1576 runner.cStateLock.Lock()
1577 defer runner.cStateLock.Unlock()
1578 if runner.cCancelled {
1581 updates := arvadosclient.Dict{
1582 "gateway_address": runner.gateway.Address,
1586 updates["log"] = logId
1588 return runner.DispatcherArvClient.Update(
1590 runner.Container.UUID,
1592 "select": []string{"uuid"},
1593 "container": updates,
1599 // ContainerToken returns the api_token the container (and any
1600 // arv-mount processes) are allowed to use.
1601 func (runner *ContainerRunner) ContainerToken() (string, error) {
1602 if runner.token != "" {
1603 return runner.token, nil
1606 var auth arvados.APIClientAuthorization
1607 err := runner.DispatcherArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth)
1611 runner.token = fmt.Sprintf("v2/%s/%s/%s", auth.UUID, auth.APIToken, runner.Container.UUID)
1612 return runner.token, nil
1615 // UpdateContainerFinal updates the container record state on API
1616 // server to "Complete" or "Cancelled"
1617 func (runner *ContainerRunner) UpdateContainerFinal() error {
1618 update := arvadosclient.Dict{}
1619 update["state"] = runner.finalState
1620 if runner.LogsPDH != nil {
1621 update["log"] = *runner.LogsPDH
1623 if runner.ExitCode != nil {
1624 update["exit_code"] = *runner.ExitCode
1626 update["exit_code"] = nil
1628 if runner.finalState == "Complete" && runner.OutputPDH != nil {
1629 update["output"] = *runner.OutputPDH
1631 update["cost"] = runner.calculateCost(time.Now())
1632 return runner.DispatcherArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{
1633 "select": []string{"uuid"},
1634 "container": update,
1638 // IsCancelled returns the value of Cancelled, with goroutine safety.
1639 func (runner *ContainerRunner) IsCancelled() bool {
1640 runner.cStateLock.Lock()
1641 defer runner.cStateLock.Unlock()
1642 return runner.cCancelled
1645 // NewArvLogWriter creates an ArvLogWriter
1646 func (runner *ContainerRunner) NewArvLogWriter(name string) (io.WriteCloser, error) {
1647 writer, err := runner.LogCollection.OpenFile(name+".txt", os.O_CREATE|os.O_WRONLY, 0666)
1651 return &ArvLogWriter{
1652 ArvClient: runner.DispatcherArvClient,
1653 UUID: runner.Container.UUID,
1654 loggingStream: name,
1655 writeCloser: writer,
1659 // Run the full container lifecycle.
1660 func (runner *ContainerRunner) Run() (err error) {
1661 runner.CrunchLog.Printf("crunch-run %s started", cmd.Version.String())
1662 runner.CrunchLog.Printf("%s", currentUserAndGroups())
1663 v, _ := exec.Command("arv-mount", "--version").CombinedOutput()
1664 runner.CrunchLog.Printf("Using FUSE mount: %s", v)
1665 runner.CrunchLog.Printf("Using container runtime: %s", runner.executor.Runtime())
1666 runner.CrunchLog.Printf("Executing container: %s", runner.Container.UUID)
1667 runner.costStartTime = time.Now()
1669 hostname, hosterr := os.Hostname()
1671 runner.CrunchLog.Printf("Error getting hostname '%v'", hosterr)
1673 runner.CrunchLog.Printf("Executing on host '%s'", hostname)
1676 sigusr2 := make(chan os.Signal, 1)
1677 signal.Notify(sigusr2, syscall.SIGUSR2)
1678 defer signal.Stop(sigusr2)
1680 go runner.handleSIGUSR2(sigusr2)
1682 runner.finalState = "Queued"
1685 runner.CleanupDirs()
1687 runner.CrunchLog.Printf("crunch-run finished")
1688 runner.CrunchLog.Close()
1691 err = runner.fetchContainerRecord()
1695 if runner.Container.State != "Locked" {
1696 return fmt.Errorf("dispatch error detected: container %q has state %q", runner.Container.UUID, runner.Container.State)
1699 var bindmounts map[string]bindmount
1701 // checkErr prints e (unless it's nil) and sets err to
1702 // e (unless err is already non-nil). Thus, if err
1703 // hasn't already been assigned when Run() returns,
1704 // this cleanup func will cause Run() to return the
1705 // first non-nil error that is passed to checkErr().
1706 checkErr := func(errorIn string, e error) {
1710 runner.CrunchLog.Printf("error in %s: %v", errorIn, e)
1714 if runner.finalState == "Complete" {
1715 // There was an error in the finalization.
1716 runner.finalState = "Cancelled"
1720 // Log the error encountered in Run(), if any
1721 checkErr("Run", err)
1723 if runner.finalState == "Queued" {
1724 runner.UpdateContainerFinal()
1728 if runner.IsCancelled() {
1729 runner.finalState = "Cancelled"
1730 // but don't return yet -- we still want to
1731 // capture partial output and write logs
1734 if bindmounts != nil {
1735 checkErr("CaptureOutput", runner.CaptureOutput(bindmounts))
1737 checkErr("stopHoststat", runner.stopHoststat())
1738 checkErr("CommitLogs", runner.CommitLogs())
1739 runner.CleanupDirs()
1740 checkErr("UpdateContainerFinal", runner.UpdateContainerFinal())
1743 runner.setupSignals()
1744 err = runner.startHoststat()
1748 if runner.keepstore != nil {
1749 runner.hoststatReporter.ReportPID("keepstore", runner.keepstore.Process.Pid)
1752 // set up FUSE mount and binds
1753 bindmounts, err = runner.SetupMounts()
1755 runner.finalState = "Cancelled"
1756 err = fmt.Errorf("While setting up mounts: %v", err)
1760 // check for and/or load image
1761 imageID, err := runner.LoadImage()
1763 if !runner.checkBrokenNode(err) {
1764 // Failed to load image but not due to a "broken node"
1765 // condition, probably user error.
1766 runner.finalState = "Cancelled"
1768 err = fmt.Errorf("While loading container image: %v", err)
1772 err = runner.CreateContainer(imageID, bindmounts)
1776 err = runner.LogHostInfo()
1780 err = runner.LogNodeRecord()
1784 err = runner.LogContainerRecord()
1789 if runner.IsCancelled() {
1793 logCollection, err := runner.saveLogCollection(false)
1796 logId = logCollection.PortableDataHash
1798 runner.CrunchLog.Printf("Error committing initial log collection: %v", err)
1800 err = runner.UpdateContainerRunning(logId)
1804 runner.finalState = "Cancelled"
1806 err = runner.startCrunchstat()
1811 err = runner.StartContainer()
1813 runner.checkBrokenNode(err)
1817 err = runner.WaitFinish()
1818 if err == nil && !runner.IsCancelled() {
1819 runner.finalState = "Complete"
1824 // Fetch the current container record (uuid = runner.Container.UUID)
1825 // into runner.Container.
1826 func (runner *ContainerRunner) fetchContainerRecord() error {
1827 reader, err := runner.DispatcherArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
1829 return fmt.Errorf("error fetching container record: %v", err)
1831 defer reader.Close()
1833 dec := json.NewDecoder(reader)
1835 err = dec.Decode(&runner.Container)
1837 return fmt.Errorf("error decoding container record: %v", err)
1841 SecretMounts map[string]arvados.Mount `json:"secret_mounts"`
1844 containerToken, err := runner.ContainerToken()
1846 return fmt.Errorf("error getting container token: %v", err)
1849 runner.ContainerArvClient, runner.ContainerKeepClient,
1850 runner.containerClient, err = runner.MkArvClient(containerToken)
1852 return fmt.Errorf("error creating container API client: %v", err)
1855 runner.ContainerKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1856 runner.DispatcherKeepClient.SetStorageClasses(runner.Container.OutputStorageClasses)
1858 err = runner.ContainerArvClient.Call("GET", "containers", runner.Container.UUID, "secret_mounts", nil, &sm)
1860 if apierr, ok := err.(arvadosclient.APIServerError); !ok || apierr.HttpStatusCode != 404 {
1861 return fmt.Errorf("error fetching secret_mounts: %v", err)
1863 // ok && apierr.HttpStatusCode == 404, which means
1864 // secret_mounts isn't supported by this API server.
1866 runner.SecretMounts = sm.SecretMounts
1871 // NewContainerRunner creates a new container runner.
1872 func NewContainerRunner(dispatcherClient *arvados.Client,
1873 dispatcherArvClient IArvadosClient,
1874 dispatcherKeepClient IKeepClient,
1875 containerUUID string) (*ContainerRunner, error) {
1877 cr := &ContainerRunner{
1878 dispatcherClient: dispatcherClient,
1879 DispatcherArvClient: dispatcherArvClient,
1880 DispatcherKeepClient: dispatcherKeepClient,
1882 cr.NewLogWriter = cr.NewArvLogWriter
1883 cr.RunArvMount = cr.ArvMountCmd
1884 cr.MkTempDir = ioutil.TempDir
1885 cr.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) {
1886 cl, err := arvadosclient.MakeArvadosClient()
1888 return nil, nil, nil, err
1891 kc, err := keepclient.MakeKeepClient(cl)
1893 return nil, nil, nil, err
1895 c2 := arvados.NewClientFromEnv()
1896 c2.AuthToken = token
1897 return cl, kc, c2, nil
1900 cr.LogCollection, err = (&arvados.Collection{}).FileSystem(cr.dispatcherClient, cr.DispatcherKeepClient)
1904 cr.Container.UUID = containerUUID
1905 w, err := cr.NewLogWriter("crunch-run")
1909 cr.CrunchLog = NewThrottledLogger(w)
1910 cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
1912 loadLogThrottleParams(dispatcherArvClient)
1918 func (command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
1919 log := log.New(stderr, "", 0)
1920 flags := flag.NewFlagSet(prog, flag.ContinueOnError)
1921 statInterval := flags.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting")
1922 flags.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree (obsolete, ignored)")
1923 flags.String("cgroup-parent", "docker", "name of container's parent cgroup (obsolete, ignored)")
1924 cgroupParentSubsystem := flags.String("cgroup-parent-subsystem", "", "use current cgroup for given `subsystem` as parent cgroup for container (subsystem argument is only relevant for cgroups v1; in cgroups v2 / unified mode, any non-empty value means use current cgroup); if empty, use the docker daemon's default cgroup parent. See https://doc.arvados.org/install/crunch2-slurm/install-dispatch.html#CrunchRunCommand-cgroups")
1925 caCertsPath := flags.String("ca-certs", "", "Path to TLS root certificates")
1926 detach := flags.Bool("detach", false, "Detach from parent process and run in the background")
1927 stdinConfig := flags.Bool("stdin-config", false, "Load config and environment variables from JSON message on stdin")
1928 configFile := flags.String("config", arvados.DefaultConfigFile, "filename of cluster config file to try loading if -stdin-config=false (default is $ARVADOS_CONFIG)")
1929 sleep := flags.Duration("sleep", 0, "Delay before starting (testing use only)")
1930 kill := flags.Int("kill", -1, "Send signal to an existing crunch-run process for given UUID")
1931 list := flags.Bool("list", false, "List UUIDs of existing crunch-run processes (and notify them to use price data passed on stdin)")
1932 enableMemoryLimit := flags.Bool("enable-memory-limit", true, "tell container runtime to limit container's memory usage")
1933 enableNetwork := flags.String("container-enable-networking", "default", "enable networking \"always\" (for all containers) or \"default\" (for containers that request it)")
1934 networkMode := flags.String("container-network-mode", "default", `Docker network mode for container (use any argument valid for docker --net)`)
1935 memprofile := flags.String("memprofile", "", "write memory profile to `file` after running container")
1936 runtimeEngine := flags.String("runtime-engine", "docker", "container runtime: docker or singularity")
1937 brokenNodeHook := flags.String("broken-node-hook", "", "script to run if node is detected to be broken (for example, Docker daemon is not running)")
1938 flags.Duration("check-containerd", 0, "Ignored. Exists for compatibility with older versions.")
1939 version := flags.Bool("version", false, "Write version information to stdout and exit 0.")
1941 ignoreDetachFlag := false
1942 if len(args) > 0 && args[0] == "-no-detach" {
1943 // This process was invoked by a parent process, which
1944 // has passed along its own arguments, including
1945 // -detach, after the leading -no-detach flag. Strip
1946 // the leading -no-detach flag (it's not recognized by
1947 // flags.Parse()) and ignore the -detach flag that
1950 ignoreDetachFlag = true
1953 if ok, code := cmd.ParseFlags(flags, prog, args, "container-uuid", stderr); !ok {
1955 } else if *version {
1956 fmt.Fprintln(stdout, prog, cmd.Version.String())
1958 } else if !*list && flags.NArg() != 1 {
1959 fmt.Fprintf(stderr, "missing required argument: container-uuid (try -help)\n")
1963 containerUUID := flags.Arg(0)
1966 case *detach && !ignoreDetachFlag:
1967 return Detach(containerUUID, prog, args, stdin, stdout, stderr)
1969 return KillProcess(containerUUID, syscall.Signal(*kill), stdout, stderr)
1971 return ListProcesses(stdin, stdout, stderr)
1974 if len(containerUUID) != 27 {
1975 log.Printf("usage: %s [options] UUID", prog)
1979 var keepstoreLogbuf bufThenWrite
1982 err := json.NewDecoder(stdin).Decode(&conf)
1984 log.Printf("decode stdin: %s", err)
1987 for k, v := range conf.Env {
1988 err = os.Setenv(k, v)
1990 log.Printf("setenv(%q): %s", k, err)
1994 if conf.Cluster != nil {
1995 // ClusterID is missing from the JSON
1996 // representation, but we need it to generate
1997 // a valid config file for keepstore, so we
1998 // fill it using the container UUID prefix.
1999 conf.Cluster.ClusterID = containerUUID[:5]
2002 conf = hpcConfData(containerUUID, *configFile, io.MultiWriter(&keepstoreLogbuf, stderr))
2005 log.Printf("crunch-run %s started", cmd.Version.String())
2008 if *caCertsPath != "" {
2009 os.Setenv("SSL_CERT_FILE", *caCertsPath)
2012 keepstore, err := startLocalKeepstore(conf, io.MultiWriter(&keepstoreLogbuf, stderr))
2017 if keepstore != nil {
2018 defer keepstore.Process.Kill()
2021 api, err := arvadosclient.MakeArvadosClient()
2023 log.Printf("%s: %v", containerUUID, err)
2026 // arvadosclient now interprets Retries=10 to mean
2027 // Timeout=10m, retrying with exponential backoff + jitter.
2030 kc, err := keepclient.MakeKeepClient(api)
2032 log.Printf("%s: %v", containerUUID, err)
2037 cr, err := NewContainerRunner(arvados.NewClientFromEnv(), api, kc, containerUUID)
2043 cr.keepstore = keepstore
2044 if keepstore == nil {
2045 // Log explanation (if any) for why we're not running
2046 // a local keepstore.
2047 var buf bytes.Buffer
2048 keepstoreLogbuf.SetWriter(&buf)
2050 cr.CrunchLog.Printf("%s", strings.TrimSpace(buf.String()))
2052 } else if logWhat := conf.Cluster.Containers.LocalKeepLogsToContainerLog; logWhat == "none" {
2053 cr.CrunchLog.Printf("using local keepstore process (pid %d) at %s", keepstore.Process.Pid, os.Getenv("ARVADOS_KEEP_SERVICES"))
2054 keepstoreLogbuf.SetWriter(io.Discard)
2056 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"))
2057 logwriter, err := cr.NewLogWriter("keepstore")
2062 cr.keepstoreLogger = NewThrottledLogger(logwriter)
2064 var writer io.WriteCloser = cr.keepstoreLogger
2065 if logWhat == "errors" {
2066 writer = &filterKeepstoreErrorsOnly{WriteCloser: writer}
2067 } else if logWhat != "all" {
2068 // should have been caught earlier by
2069 // dispatcher's config loader
2070 log.Printf("invalid value for Containers.LocalKeepLogsToContainerLog: %q", logWhat)
2073 err = keepstoreLogbuf.SetWriter(writer)
2078 cr.keepstoreLogbuf = &keepstoreLogbuf
2081 switch *runtimeEngine {
2083 cr.executor, err = newDockerExecutor(containerUUID, cr.CrunchLog.Printf, cr.containerWatchdogInterval)
2085 cr.executor, err = newSingularityExecutor(cr.CrunchLog.Printf)
2087 cr.CrunchLog.Printf("%s: unsupported RuntimeEngine %q", containerUUID, *runtimeEngine)
2088 cr.CrunchLog.Close()
2092 cr.CrunchLog.Printf("%s: %v", containerUUID, err)
2093 cr.checkBrokenNode(err)
2094 cr.CrunchLog.Close()
2097 defer cr.executor.Close()
2099 cr.brokenNodeHook = *brokenNodeHook
2101 gwAuthSecret := os.Getenv("GatewayAuthSecret")
2102 os.Unsetenv("GatewayAuthSecret")
2103 if gwAuthSecret == "" {
2104 // not safe to run a gateway service without an auth
2106 cr.CrunchLog.Printf("Not starting a gateway server (GatewayAuthSecret was not provided by dispatcher)")
2108 gwListen := os.Getenv("GatewayAddress")
2109 cr.gateway = Gateway{
2111 AuthSecret: gwAuthSecret,
2112 ContainerUUID: containerUUID,
2113 Target: cr.executor,
2115 LogCollection: cr.LogCollection,
2118 // Direct connection won't work, so we use the
2119 // gateway_address field to indicate the
2120 // internalURL of the controller process that
2121 // has the current tunnel connection.
2122 cr.gateway.ArvadosClient = cr.dispatcherClient
2123 cr.gateway.UpdateTunnelURL = func(url string) {
2124 cr.gateway.Address = "tunnel " + url
2125 cr.DispatcherArvClient.Update("containers", containerUUID,
2127 "select": []string{"uuid"},
2128 "container": arvadosclient.Dict{"gateway_address": cr.gateway.Address},
2132 err = cr.gateway.Start()
2134 log.Printf("error starting gateway server: %s", err)
2139 parentTemp, tmperr := cr.MkTempDir("", "crunch-run."+containerUUID+".")
2141 log.Printf("%s: %v", containerUUID, tmperr)
2145 cr.parentTemp = parentTemp
2146 cr.statInterval = *statInterval
2147 cr.enableMemoryLimit = *enableMemoryLimit
2148 cr.enableNetwork = *enableNetwork
2149 cr.networkMode = *networkMode
2150 if *cgroupParentSubsystem != "" {
2151 p, err := findCgroup(os.DirFS("/"), *cgroupParentSubsystem)
2153 log.Printf("fatal: cgroup parent subsystem: %s", err)
2156 cr.setCgroupParent = p
2159 if conf.EC2SpotCheck {
2160 go cr.checkSpotInterruptionNotices()
2165 if *memprofile != "" {
2166 f, err := os.Create(*memprofile)
2168 log.Printf("could not create memory profile: %s", err)
2170 runtime.GC() // get up-to-date statistics
2171 if err := pprof.WriteHeapProfile(f); err != nil {
2172 log.Printf("could not write memory profile: %s", err)
2174 closeerr := f.Close()
2175 if closeerr != nil {
2176 log.Printf("closing memprofile file: %s", err)
2181 log.Printf("%s: %v", containerUUID, runerr)
2187 // Try to load ConfigData in hpc (slurm/lsf) environment. This means
2188 // loading the cluster config from the specified file and (if that
2189 // works) getting the runtime_constraints container field from
2190 // controller to determine # VCPUs so we can calculate KeepBuffers.
2191 func hpcConfData(uuid string, configFile string, stderr io.Writer) ConfigData {
2193 conf.Cluster = loadClusterConfigFile(configFile, stderr)
2194 if conf.Cluster == nil {
2195 // skip loading the container record -- we won't be
2196 // able to start local keepstore anyway.
2199 arv, err := arvadosclient.MakeArvadosClient()
2201 fmt.Fprintf(stderr, "error setting up arvadosclient: %s\n", err)
2204 // arvadosclient now interprets Retries=10 to mean
2205 // Timeout=10m, retrying with exponential backoff + jitter.
2207 var ctr arvados.Container
2208 err = arv.Call("GET", "containers", uuid, "", arvadosclient.Dict{"select": []string{"runtime_constraints"}}, &ctr)
2210 fmt.Fprintf(stderr, "error getting container record: %s\n", err)
2213 if ctr.RuntimeConstraints.VCPUs > 0 {
2214 conf.KeepBuffers = ctr.RuntimeConstraints.VCPUs * conf.Cluster.Containers.LocalKeepBlobBuffersPerVCPU
2219 // Load cluster config file from given path. If an error occurs, log
2220 // the error to stderr and return nil.
2221 func loadClusterConfigFile(path string, stderr io.Writer) *arvados.Cluster {
2222 ldr := config.NewLoader(&bytes.Buffer{}, ctxlog.New(stderr, "plain", "info"))
2224 cfg, err := ldr.Load()
2226 fmt.Fprintf(stderr, "could not load config file %s: %s\n", path, err)
2229 cluster, err := cfg.GetCluster("")
2231 fmt.Fprintf(stderr, "could not use config file %s: %s\n", path, err)
2234 fmt.Fprintf(stderr, "loaded config file %s\n", path)
2238 func startLocalKeepstore(configData ConfigData, logbuf io.Writer) (*exec.Cmd, error) {
2239 if configData.KeepBuffers < 1 {
2240 fmt.Fprintf(logbuf, "not starting a local keepstore process because KeepBuffers=%v in config\n", configData.KeepBuffers)
2243 if configData.Cluster == nil {
2244 fmt.Fprint(logbuf, "not starting a local keepstore process because cluster config file was not loaded\n")
2247 for uuid, vol := range configData.Cluster.Volumes {
2248 if len(vol.AccessViaHosts) > 0 {
2249 fmt.Fprintf(logbuf, "not starting a local keepstore process because a volume (%s) uses AccessViaHosts\n", uuid)
2252 if !vol.ReadOnly && vol.Replication < configData.Cluster.Collections.DefaultReplication {
2253 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)
2258 // Rather than have an alternate way to tell keepstore how
2259 // many buffers to use, etc., when starting it this way, we
2260 // just modify the cluster configuration that we feed it on
2262 ccfg := *configData.Cluster
2263 ccfg.API.MaxKeepBlobBuffers = configData.KeepBuffers
2264 ccfg.Collections.BlobTrash = false
2265 ccfg.Collections.BlobTrashConcurrency = 0
2266 ccfg.Collections.BlobDeleteConcurrency = 0
2268 localaddr := localKeepstoreAddr()
2269 ln, err := net.Listen("tcp", net.JoinHostPort(localaddr, "0"))
2273 _, port, err := net.SplitHostPort(ln.Addr().String())
2279 url := "http://" + net.JoinHostPort(localaddr, port)
2281 fmt.Fprintf(logbuf, "starting keepstore on %s\n", url)
2283 var confJSON bytes.Buffer
2284 err = json.NewEncoder(&confJSON).Encode(arvados.Config{
2285 Clusters: map[string]arvados.Cluster{
2286 ccfg.ClusterID: ccfg,
2292 cmd := exec.Command("/proc/self/exe", "keepstore", "-config=-")
2293 if target, err := os.Readlink(cmd.Path); err == nil && strings.HasSuffix(target, ".test") {
2294 // If we're a 'go test' process, running
2295 // /proc/self/exe would start the test suite in a
2296 // child process, which is not what we want.
2297 cmd.Path, _ = exec.LookPath("go")
2298 cmd.Args = append([]string{"go", "run", "../../cmd/arvados-server"}, cmd.Args[1:]...)
2299 cmd.Env = os.Environ()
2301 cmd.Stdin = &confJSON
2304 cmd.Env = append(cmd.Env,
2306 "ARVADOS_SERVICE_INTERNAL_URL="+url)
2309 return nil, fmt.Errorf("error starting keepstore process: %w", err)
2316 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(time.Second*10))
2318 poll := time.NewTicker(time.Second / 10)
2320 client := http.Client{}
2322 testReq, err := http.NewRequestWithContext(ctx, "GET", url+"/_health/ping", nil)
2323 testReq.Header.Set("Authorization", "Bearer "+configData.Cluster.ManagementToken)
2327 resp, err := client.Do(testReq)
2330 if resp.StatusCode == http.StatusOK {
2335 return nil, fmt.Errorf("keepstore child process exited")
2337 if ctx.Err() != nil {
2338 return nil, fmt.Errorf("timed out waiting for new keepstore process to report healthy")
2341 os.Setenv("ARVADOS_KEEP_SERVICES", url)
2345 // return current uid, gid, groups in a format suitable for logging:
2346 // "crunch-run process has uid=1234(arvados) gid=1234(arvados)
2347 // groups=1234(arvados),114(fuse)"
2348 func currentUserAndGroups() string {
2349 u, err := user.Current()
2351 return fmt.Sprintf("error getting current user ID: %s", err)
2353 s := fmt.Sprintf("crunch-run process has uid=%s(%s) gid=%s", u.Uid, u.Username, u.Gid)
2354 if g, err := user.LookupGroupId(u.Gid); err == nil {
2355 s += fmt.Sprintf("(%s)", g.Name)
2358 if gids, err := u.GroupIds(); err == nil {
2359 for i, gid := range gids {
2364 if g, err := user.LookupGroupId(gid); err == nil {
2365 s += fmt.Sprintf("(%s)", g.Name)
2372 // Return a suitable local interface address for a local keepstore
2373 // service. Currently this is the numerically lowest non-loopback ipv4
2374 // address assigned to a local interface that is not in any of the
2375 // link-local/vpn/loopback ranges 169.254/16, 100.64/10, or 127/8.
2376 func localKeepstoreAddr() string {
2378 // Ignore error (proceed with zero IPs)
2379 addrs, _ := processIPs(os.Getpid())
2380 for addr := range addrs {
2381 ip := net.ParseIP(addr)
2386 if ip.Mask(net.CIDRMask(8, 32)).Equal(net.IPv4(127, 0, 0, 0)) ||
2387 ip.Mask(net.CIDRMask(10, 32)).Equal(net.IPv4(100, 64, 0, 0)) ||
2388 ip.Mask(net.CIDRMask(16, 32)).Equal(net.IPv4(169, 254, 0, 0)) {
2392 ips = append(ips, ip)
2397 sort.Slice(ips, func(ii, jj int) bool {
2398 i, j := ips[ii], ips[jj]
2399 if len(i) != len(j) {
2400 return len(i) < len(j)
2409 return ips[0].String()
2412 func (cr *ContainerRunner) loadPrices() {
2413 buf, err := os.ReadFile(filepath.Join(lockdir, pricesfile))
2415 if !os.IsNotExist(err) {
2416 cr.CrunchLog.Printf("loadPrices: read: %s", err)
2420 var prices []cloud.InstancePrice
2421 err = json.Unmarshal(buf, &prices)
2423 cr.CrunchLog.Printf("loadPrices: decode: %s", err)
2426 cr.pricesLock.Lock()
2427 defer cr.pricesLock.Unlock()
2428 var lastKnown time.Time
2429 if len(cr.prices) > 0 {
2430 lastKnown = cr.prices[0].StartTime
2432 cr.prices = cloud.NormalizePriceHistory(append(prices, cr.prices...))
2433 for i := len(cr.prices) - 1; i >= 0; i-- {
2434 price := cr.prices[i]
2435 if price.StartTime.After(lastKnown) {
2436 cr.CrunchLog.Printf("Instance price changed to %#.3g at %s", price.Price, price.StartTime.UTC())
2441 func (cr *ContainerRunner) calculateCost(now time.Time) float64 {
2442 cr.pricesLock.Lock()
2443 defer cr.pricesLock.Unlock()
2445 // First, make a "prices" slice with the real data as far back
2446 // as it goes, and (if needed) a "since the beginning of time"
2447 // placeholder containing a reasonable guess about what the
2448 // price was between cr.costStartTime and the earliest real
2451 if len(prices) == 0 {
2452 // use price info in InstanceType record initially
2453 // provided by cloud dispatcher
2455 var it arvados.InstanceType
2456 if j := os.Getenv("InstanceType"); j != "" && json.Unmarshal([]byte(j), &it) == nil && it.Price > 0 {
2459 prices = []cloud.InstancePrice{{Price: p}}
2460 } else if prices[len(prices)-1].StartTime.After(cr.costStartTime) {
2461 // guess earlier pricing was the same as the earliest
2462 // price we know about
2463 filler := prices[len(prices)-1]
2464 filler.StartTime = time.Time{}
2465 prices = append(prices, filler)
2468 // Now that our history of price changes goes back at least as
2469 // far as cr.costStartTime, add up the costs for each
2473 for _, ip := range prices {
2474 spanStart := ip.StartTime
2475 if spanStart.After(now) {
2476 // pricing information from the future -- not
2477 // expected from AWS, but possible in
2478 // principle, and exercised by tests.
2482 if spanStart.Before(cr.costStartTime) {
2483 spanStart = cr.costStartTime
2486 cost += ip.Price * spanEnd.Sub(spanStart).Seconds() / 3600
2496 func (runner *ContainerRunner) handleSIGUSR2(sigchan chan os.Signal) {
2499 update := arvadosclient.Dict{
2500 "select": []string{"uuid"},
2501 "container": arvadosclient.Dict{
2502 "cost": runner.calculateCost(time.Now()),
2505 runner.DispatcherArvClient.Update("containers", runner.Container.UUID, update, nil)