X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/0ac69cee6889be8337d147a75829596f19075fa5..85c684122fd678dc24035d58236dd5734aed50e5:/services/crunch-run/crunchrun.go diff --git a/services/crunch-run/crunchrun.go b/services/crunch-run/crunchrun.go index d508ffb3c7..f73ec73466 100644 --- a/services/crunch-run/crunchrun.go +++ b/services/crunch-run/crunchrun.go @@ -5,6 +5,8 @@ import ( "errors" "flag" "fmt" + "git.curoverse.com/arvados.git/lib/crunchstat" + "git.curoverse.com/arvados.git/sdk/go/arvados" "git.curoverse.com/arvados.git/sdk/go/arvadosclient" "git.curoverse.com/arvados.git/sdk/go/keepclient" "git.curoverse.com/arvados.git/sdk/go/manifest" @@ -16,6 +18,8 @@ import ( "os/exec" "os/signal" "path" + "path/filepath" + "sort" "strings" "sync" "syscall" @@ -26,8 +30,9 @@ import ( type IArvadosClient interface { Create(resourceType string, parameters arvadosclient.Dict, output interface{}) error Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error - Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) (err error) - Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) (err error) + Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error + Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error + Discovery(key string) (interface{}, error) } // ErrCancelled is the error returned when the container is cancelled. @@ -36,48 +41,7 @@ var ErrCancelled = errors.New("Cancelled") // IKeepClient is the minimal Keep API methods used by crunch-run. type IKeepClient interface { PutHB(hash string, buf []byte) (string, int, error) - ManifestFileReader(m manifest.Manifest, filename string) (keepclient.ReadCloserWithLen, error) -} - -// Mount describes the mount points to create inside the container. -type Mount struct { - Kind string `json:"kind"` - Writable bool `json:"writable"` - PortableDataHash string `json:"portable_data_hash"` - UUID string `json:"uuid"` - DeviceType string `json:"device_type"` - Path string `json:"path"` -} - -// Collection record returned by the API server. -type CollectionRecord struct { - ManifestText string `json:"manifest_text"` - PortableDataHash string `json:"portable_data_hash"` -} - -type RuntimeConstraints struct { - API *bool -} - -// ContainerRecord is the container record returned by the API server. -type ContainerRecord struct { - UUID string `json:"uuid"` - Command []string `json:"command"` - ContainerImage string `json:"container_image"` - Cwd string `json:"cwd"` - Environment map[string]string `json:"environment"` - Mounts map[string]Mount `json:"mounts"` - OutputPath string `json:"output_path"` - Priority int `json:"priority"` - RuntimeConstraints RuntimeConstraints `json:"runtime_constraints"` - State string `json:"state"` - Output string `json:"output"` -} - -// APIClientAuthorization is an arvados#api_client_authorization resource. -type APIClientAuthorization struct { - UUID string `json:"uuid"` - APIToken string `json:"api_token"` + ManifestFileReader(m manifest.Manifest, filename string) (keepclient.Reader, error) } // NewLogWriter is a factory function to create a new log writer. @@ -105,7 +69,7 @@ type ContainerRunner struct { Docker ThinDockerClient ArvClient IArvadosClient Kc IKeepClient - ContainerRecord + arvados.Container dockerclient.ContainerConfig dockerclient.HostConfig token string @@ -131,6 +95,25 @@ type ContainerRunner struct { SigChan chan os.Signal ArvMountExit chan error finalState string + trashLifetime time.Duration + + statLogger io.WriteCloser + statReporter *crunchstat.Reporter + statInterval time.Duration + cgroupRoot string + // What we expect the container's cgroup parent to be. + expectCgroupParent string + // What we tell docker to use as the container's cgroup + // parent. Note: Ideally we would use the same field for both + // expectCgroupParent and setCgroupParent, and just make it + // default to "docker". However, when using docker < 1.10 with + // systemd, specifying a non-empty cgroup parent (even the + // default value "docker") hits a docker bug + // (https://github.com/docker/docker/issues/17126). Using two + // separate fields makes it possible to use the "expect cgroup + // parent to be X" feature even on sites where the "specify + // cgroup parent" feature breaks. + setCgroupParent string } // SetupSignals sets up signal handling to gracefully terminate the underlying @@ -142,7 +125,7 @@ func (runner *ContainerRunner) SetupSignals() { signal.Notify(runner.SigChan, syscall.SIGQUIT) go func(sig <-chan os.Signal) { - for _ = range sig { + for range sig { if !runner.Cancelled { runner.CancelLock.Lock() runner.Cancelled = true @@ -160,10 +143,10 @@ func (runner *ContainerRunner) SetupSignals() { // the image from Keep. func (runner *ContainerRunner) LoadImage() (err error) { - runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.ContainerRecord.ContainerImage) + runner.CrunchLog.Printf("Fetching Docker image from collection '%s'", runner.Container.ContainerImage) - var collection CollectionRecord - err = runner.ArvClient.Get("collections", runner.ContainerRecord.ContainerImage, nil, &collection) + var collection arvados.Collection + err = runner.ArvClient.Get("collections", runner.Container.ContainerImage, nil, &collection) if err != nil { return fmt.Errorf("While getting container image collection: %v", err) } @@ -257,8 +240,15 @@ func (runner *ContainerRunner) ArvMountCmd(arvMountCmd []string, token string) ( return c, nil } +func (runner *ContainerRunner) SetupArvMountPoint(prefix string) (err error) { + if runner.ArvMountPoint == "" { + runner.ArvMountPoint, err = runner.MkTempDir("", prefix) + } + return +} + func (runner *ContainerRunner) SetupMounts() (err error) { - runner.ArvMountPoint, err = runner.MkTempDir("", "keep") + err = runner.SetupArvMountPoint("keep") if err != nil { return fmt.Errorf("While creating keep mount temp dir: %v", err) } @@ -268,10 +258,23 @@ func (runner *ContainerRunner) SetupMounts() (err error) { pdhOnly := true tmpcount := 0 arvMountCmd := []string{"--foreground", "--allow-other", "--read-write"} + + if runner.Container.RuntimeConstraints.KeepCacheRAM > 0 { + arvMountCmd = append(arvMountCmd, "--file-cache", fmt.Sprintf("%d", runner.Container.RuntimeConstraints.KeepCacheRAM)) + } + collectionPaths := []string{} runner.Binds = nil + needCertMount := true + + var binds []string + for bind, _ := range runner.Container.Mounts { + binds = append(binds, bind) + } + sort.Strings(binds) - for bind, mnt := range runner.ContainerRecord.Mounts { + for _, bind := range binds { + mnt := runner.Container.Mounts[bind] if bind == "stdout" { // Is it a "file" mount kind? if mnt.Kind != "file" { @@ -279,7 +282,7 @@ func (runner *ContainerRunner) SetupMounts() (err error) { } // Does path start with OutputPath? - prefix := runner.ContainerRecord.OutputPath + prefix := runner.Container.OutputPath if !strings.HasSuffix(prefix, "/") { prefix += "/" } @@ -288,7 +291,18 @@ func (runner *ContainerRunner) SetupMounts() (err error) { } } - if mnt.Kind == "collection" { + if bind == "/etc/arvados/ca-certificates.crt" { + needCertMount = false + } + + if strings.HasPrefix(bind, runner.Container.OutputPath+"/") && bind != runner.Container.OutputPath+"/" { + if mnt.Kind != "collection" { + return fmt.Errorf("Only mount points of kind 'collection' are supported underneath the output_path: %v", bind) + } + } + + switch { + case mnt.Kind == "collection": var src string if mnt.UUID != "" && mnt.PortableDataHash != "" { return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount") @@ -311,33 +325,57 @@ func (runner *ContainerRunner) SetupMounts() (err error) { tmpcount += 1 } if mnt.Writable { - if bind == runner.ContainerRecord.OutputPath { + if bind == runner.Container.OutputPath { runner.HostOutputDir = src + } else if strings.HasPrefix(bind, runner.Container.OutputPath+"/") { + return fmt.Errorf("Writable mount points are not permitted underneath the output_path: %v", bind) } runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", src, bind)) } else { runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", src, bind)) } collectionPaths = append(collectionPaths, src) - } else if mnt.Kind == "tmp" { - if bind == runner.ContainerRecord.OutputPath { - runner.HostOutputDir, err = runner.MkTempDir("", "") - if err != nil { - return fmt.Errorf("While creating mount temp dir: %v", err) - } - st, staterr := os.Stat(runner.HostOutputDir) - if staterr != nil { - return fmt.Errorf("While Stat on temp dir: %v", staterr) - } - err = os.Chmod(runner.HostOutputDir, st.Mode()|os.ModeSetgid|0777) - if staterr != nil { - return fmt.Errorf("While Chmod temp dir: %v", err) - } - runner.CleanupTempDir = append(runner.CleanupTempDir, runner.HostOutputDir) - runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", runner.HostOutputDir, bind)) - } else { - runner.Binds = append(runner.Binds, bind) + + case mnt.Kind == "tmp" && bind == runner.Container.OutputPath: + runner.HostOutputDir, err = runner.MkTempDir("", "") + if err != nil { + return fmt.Errorf("While creating mount temp dir: %v", err) + } + st, staterr := os.Stat(runner.HostOutputDir) + if staterr != nil { + return fmt.Errorf("While Stat on temp dir: %v", staterr) + } + err = os.Chmod(runner.HostOutputDir, st.Mode()|os.ModeSetgid|0777) + if staterr != nil { + return fmt.Errorf("While Chmod temp dir: %v", err) + } + runner.CleanupTempDir = append(runner.CleanupTempDir, runner.HostOutputDir) + runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", runner.HostOutputDir, bind)) + + case mnt.Kind == "tmp": + runner.Binds = append(runner.Binds, bind) + + case mnt.Kind == "json": + jsondata, err := json.Marshal(mnt.Content) + if err != nil { + return fmt.Errorf("encoding json data: %v", err) + } + // Create a tempdir with a single file + // (instead of just a tempfile): this way we + // can ensure the file is world-readable + // inside the container, without having to + // make it world-readable on the docker host. + tmpdir, err := runner.MkTempDir("", "") + if err != nil { + return fmt.Errorf("creating temp dir: %v", err) } + runner.CleanupTempDir = append(runner.CleanupTempDir, tmpdir) + tmpfn := filepath.Join(tmpdir, "mountdata.json") + err = ioutil.WriteFile(tmpfn, jsondata, 0644) + if err != nil { + return fmt.Errorf("writing temp file: %v", err) + } + runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s:ro", tmpfn, bind)) } } @@ -345,6 +383,16 @@ func (runner *ContainerRunner) SetupMounts() (err error) { return fmt.Errorf("Output path does not correspond to a writable mount point") } + if wantAPI := runner.Container.RuntimeConstraints.API; needCertMount && wantAPI != nil && *wantAPI { + for _, certfile := range arvadosclient.CertFiles { + _, err := os.Stat(certfile) + if err == nil { + runner.Binds = append(runner.Binds, fmt.Sprintf("%s:/etc/arvados/ca-certificates.crt:ro", certfile)) + break + } + } + } + if pdhOnly { arvMountCmd = append(arvMountCmd, "--mount-by-pdh", "by_id") } else { @@ -406,6 +454,14 @@ func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) { runner.CrunchLog.Printf("While closing stderr logs: %v", closeerr) } + if runner.statReporter != nil { + runner.statReporter.Stop() + closeerr = runner.statLogger.Close() + if closeerr != nil { + runner.CrunchLog.Printf("While closing crunchstat logs: %v", closeerr) + } + } + runner.loggingDone <- true close(runner.loggingDone) return @@ -413,6 +469,18 @@ func (runner *ContainerRunner) ProcessDockerAttach(containerReader io.Reader) { } } +func (runner *ContainerRunner) StartCrunchstat() { + runner.statLogger = NewThrottledLogger(runner.NewLogWriter("crunchstat")) + runner.statReporter = &crunchstat.Reporter{ + CID: runner.ContainerID, + Logger: log.New(runner.statLogger, "", 0), + CgroupParent: runner.expectCgroupParent, + CgroupRoot: runner.cgroupRoot, + PollPeriod: runner.statInterval, + } + runner.statReporter.Start() +} + // AttachLogs connects the docker container stdout and stderr logs to the // Arvados logger which logs to Keep and the API server logs table. func (runner *ContainerRunner) AttachStreams() (err error) { @@ -428,8 +496,8 @@ func (runner *ContainerRunner) AttachStreams() (err error) { runner.loggingDone = make(chan bool) - if stdoutMnt, ok := runner.ContainerRecord.Mounts["stdout"]; ok { - stdoutPath := stdoutMnt.Path[len(runner.ContainerRecord.OutputPath):] + if stdoutMnt, ok := runner.Container.Mounts["stdout"]; ok { + stdoutPath := stdoutMnt.Path[len(runner.Container.OutputPath):] index := strings.LastIndex(stdoutPath, "/") if index > 0 { subdirs := stdoutPath[:index] @@ -464,15 +532,15 @@ func (runner *ContainerRunner) AttachStreams() (err error) { func (runner *ContainerRunner) CreateContainer() error { runner.CrunchLog.Print("Creating Docker container") - runner.ContainerConfig.Cmd = runner.ContainerRecord.Command - if runner.ContainerRecord.Cwd != "." { - runner.ContainerConfig.WorkingDir = runner.ContainerRecord.Cwd + runner.ContainerConfig.Cmd = runner.Container.Command + if runner.Container.Cwd != "." { + runner.ContainerConfig.WorkingDir = runner.Container.Cwd } - for k, v := range runner.ContainerRecord.Environment { + for k, v := range runner.Container.Environment { runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v) } - if wantAPI := runner.ContainerRecord.RuntimeConstraints.API; wantAPI != nil && *wantAPI { + if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI { tok, err := runner.ContainerToken() if err != nil { return err @@ -493,8 +561,13 @@ func (runner *ContainerRunner) CreateContainer() error { return fmt.Errorf("While creating container: %v", err) } - runner.HostConfig = dockerclient.HostConfig{Binds: runner.Binds, - LogConfig: dockerclient.LogConfig{Type: "none"}} + runner.HostConfig = dockerclient.HostConfig{ + Binds: runner.Binds, + CgroupParent: runner.setCgroupParent, + LogConfig: dockerclient.LogConfig{ + Type: "none", + }, + } return runner.AttachStreams() } @@ -533,6 +606,21 @@ func (runner *ContainerRunner) CaptureOutput() error { return nil } + if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI { + // Output may have been set directly by the container, so + // refresh the container record to check. + err := runner.ArvClient.Get("containers", runner.Container.UUID, + nil, &runner.Container) + if err != nil { + return err + } + if runner.Container.Output != "" { + // Container output is already set. + runner.OutputPDH = &runner.Container.Output + return nil + } + } + if runner.HostOutputDir == "" { return nil } @@ -561,7 +649,7 @@ func (runner *ContainerRunner) CaptureOutput() error { } defer file.Close() - rec := CollectionRecord{} + var rec arvados.Collection err = json.NewDecoder(file).Decode(&rec) if err != nil { return fmt.Errorf("While reading FUSE metafile: %v", err) @@ -569,20 +657,107 @@ func (runner *ContainerRunner) CaptureOutput() error { manifestText = rec.ManifestText } - var response CollectionRecord + // Pre-populate output from the configured mount points + var binds []string + for bind, _ := range runner.Container.Mounts { + binds = append(binds, bind) + } + sort.Strings(binds) + + for _, bind := range binds { + mnt := runner.Container.Mounts[bind] + + bindSuffix := strings.TrimPrefix(bind, runner.Container.OutputPath) + + if bindSuffix == bind || len(bindSuffix) <= 0 { + // either does not start with OutputPath or is OutputPath itself + continue + } + + if strings.HasPrefix(bindSuffix, "/") == false { + bindSuffix = "/" + bindSuffix + } + + if mnt.ExcludeFromOutput == true { + continue + } + + idx := strings.Index(mnt.PortableDataHash, "/") + if idx > 0 { + mnt.Path = mnt.PortableDataHash[idx:] + mnt.PortableDataHash = mnt.PortableDataHash[0:idx] + } + + // append to manifest_text + m, err := runner.getCollectionManifestForPath(mnt, bindSuffix) + if err != nil { + return err + } + + manifestText = manifestText + m + } + + // Save output + var response arvados.Collection + manifestText := manifest.Manifest{Text: manifestText}.NormalizedText() err = runner.ArvClient.Create("collections", arvadosclient.Dict{ "collection": arvadosclient.Dict{ + "trash_at": time.Now().Add(runner.trashLifetime).Format(time.RFC3339), + "name": "output for " + runner.Container.UUID, "manifest_text": manifestText}}, &response) if err != nil { return fmt.Errorf("While creating output collection: %v", err) } + runner.OutputPDH = &response.PortableDataHash + return nil +} + +var outputCollections = make(map[string]arvados.Collection) + +// Fetch the collection for the mnt.PortableDataHash +// Return the manifest_text fragment corresponding to the specified mnt.Path +// after making any required updates. +// Ex: +// If mnt.Path is not specified, +// return the entire manifest_text after replacing any "." with bindSuffix +// If mnt.Path corresponds to one stream, +// return the manifest_text for that stream after replacing that stream name with bindSuffix +// Otherwise, check if a filename in any one stream is being sought. Return the manifest_text +// for that stream after replacing stream name with bindSuffix minus the last word +// and the file name with last word of the bindSuffix +// Allowed path examples: +// "path":"/" +// "path":"/subdir1" +// "path":"/subdir1/subdir2" +// "path":"/subdir/filename" etc +func (runner *ContainerRunner) getCollectionManifestForPath(mnt arvados.Mount, bindSuffix string) (string, error) { + collection := outputCollections[mnt.PortableDataHash] + if collection.PortableDataHash == "" { + err := runner.ArvClient.Get("collections", mnt.PortableDataHash, nil, &collection) + if err != nil { + return "", fmt.Errorf("While getting collection for %v: %v", mnt.PortableDataHash, err) + } + outputCollections[mnt.PortableDataHash] = collection + } - runner.OutputPDH = new(string) - *runner.OutputPDH = response.PortableDataHash + if collection.ManifestText == "" { + runner.CrunchLog.Printf("No manifest text for collection %v", collection.PortableDataHash) + return "", nil + } - return nil + manifest := manifest.Manifest{Text: collection.ManifestText} + manifestText := manifest.ManifestTextForPath(mnt.Path, bindSuffix) + return manifestText, nil +} + +func (runner *ContainerRunner) loadDiscoveryVars() { + tl, err := runner.ArvClient.Discovery("defaultTrashLifetime") + if err != nil { + log.Fatalf("getting defaultTrashLifetime from discovery document: %s", err) + } + runner.trashLifetime = time.Duration(tl.(float64)) * time.Second } func (runner *ContainerRunner) CleanupDirs() { @@ -616,7 +791,7 @@ func (runner *ContainerRunner) CommitLogs() error { // point, but re-open crunch log with ArvClient in case there are any // other further (such as failing to write the log to Keep!) while // shutting down - runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{runner.ArvClient, runner.ContainerRecord.UUID, + runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{runner.ArvClient, runner.Container.UUID, "crunch-run", nil}) if runner.LogsPDH != nil { @@ -633,30 +808,29 @@ func (runner *ContainerRunner) CommitLogs() error { return fmt.Errorf("While creating log manifest: %v", err) } - var response CollectionRecord + var response arvados.Collection err = runner.ArvClient.Create("collections", arvadosclient.Dict{ "collection": arvadosclient.Dict{ - "name": "logs for " + runner.ContainerRecord.UUID, + "trash_at": time.Now().Add(runner.trashLifetime).Format(time.RFC3339), + "name": "logs for " + runner.Container.UUID, "manifest_text": mt}}, &response) if err != nil { return fmt.Errorf("While creating log collection: %v", err) } - runner.LogsPDH = &response.PortableDataHash - return nil } -// UpdateContainerRecordRunning updates the container state to "Running" -func (runner *ContainerRunner) UpdateContainerRecordRunning() error { +// UpdateContainerRunning updates the container state to "Running" +func (runner *ContainerRunner) UpdateContainerRunning() error { runner.CancelLock.Lock() defer runner.CancelLock.Unlock() if runner.Cancelled { return ErrCancelled } - return runner.ArvClient.Update("containers", runner.ContainerRecord.UUID, + return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": arvadosclient.Dict{"state": "Running"}}, nil) } @@ -667,8 +841,8 @@ func (runner *ContainerRunner) ContainerToken() (string, error) { return runner.token, nil } - var auth APIClientAuthorization - err := runner.ArvClient.Call("GET", "containers", runner.ContainerRecord.UUID, "auth", nil, &auth) + var auth arvados.APIClientAuthorization + err := runner.ArvClient.Call("GET", "containers", runner.Container.UUID, "auth", nil, &auth) if err != nil { return "", err } @@ -676,15 +850,15 @@ func (runner *ContainerRunner) ContainerToken() (string, error) { return runner.token, nil } -// UpdateContainerRecordComplete updates the container record state on API +// UpdateContainerComplete updates the container record state on API // server to "Complete" or "Cancelled" -func (runner *ContainerRunner) UpdateContainerRecordFinal() error { +func (runner *ContainerRunner) UpdateContainerFinal() error { update := arvadosclient.Dict{} update["state"] = runner.finalState + if runner.LogsPDH != nil { + update["log"] = *runner.LogsPDH + } if runner.finalState == "Complete" { - if runner.LogsPDH != nil { - update["log"] = *runner.LogsPDH - } if runner.ExitCode != nil { update["exit_code"] = *runner.ExitCode } @@ -692,7 +866,7 @@ func (runner *ContainerRunner) UpdateContainerRecordFinal() error { update["output"] = *runner.OutputPDH } } - return runner.ArvClient.Update("containers", runner.ContainerRecord.UUID, arvadosclient.Dict{"container": update}, nil) + return runner.ArvClient.Update("containers", runner.Container.UUID, arvadosclient.Dict{"container": update}, nil) } // IsCancelled returns the value of Cancelled, with goroutine safety. @@ -704,12 +878,12 @@ func (runner *ContainerRunner) IsCancelled() bool { // NewArvLogWriter creates an ArvLogWriter func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser { - return &ArvLogWriter{runner.ArvClient, runner.ContainerRecord.UUID, name, runner.LogCollection.Open(name + ".txt")} + return &ArvLogWriter{runner.ArvClient, runner.Container.UUID, name, runner.LogCollection.Open(name + ".txt")} } // Run the full container lifecycle. func (runner *ContainerRunner) Run() (err error) { - runner.CrunchLog.Printf("Executing container '%s'", runner.ContainerRecord.UUID) + runner.CrunchLog.Printf("Executing container '%s'", runner.Container.UUID) hostname, hosterr := os.Hostname() if hosterr != nil { @@ -744,7 +918,8 @@ func (runner *ContainerRunner) Run() (err error) { checkErr(err) if runner.finalState == "Queued" { - runner.UpdateContainerRecordFinal() + runner.CrunchLog.Close() + runner.UpdateContainerFinal() return } @@ -756,7 +931,7 @@ func (runner *ContainerRunner) Run() (err error) { checkErr(runner.CaptureOutput()) checkErr(runner.CommitLogs()) - checkErr(runner.UpdateContainerRecordFinal()) + checkErr(runner.UpdateContainerFinal()) // The real log is already closed, but then we opened // a new one in case we needed to log anything while @@ -764,7 +939,7 @@ func (runner *ContainerRunner) Run() (err error) { runner.CrunchLog.Close() }() - err = runner.ArvClient.Get("containers", runner.ContainerRecord.UUID, nil, &runner.ContainerRecord) + err = runner.ArvClient.Get("containers", runner.Container.UUID, nil, &runner.Container) if err != nil { err = fmt.Errorf("While getting container record: %v", err) return @@ -776,6 +951,7 @@ func (runner *ContainerRunner) Run() (err error) { // check for and/or load image err = runner.LoadImage() if err != nil { + runner.finalState = "Cancelled" err = fmt.Errorf("While loading container image: %v", err) return } @@ -783,6 +959,7 @@ func (runner *ContainerRunner) Run() (err error) { // set up FUSE mount and binds err = runner.SetupMounts() if err != nil { + runner.finalState = "Cancelled" err = fmt.Errorf("While setting up mounts: %v", err) return } @@ -792,11 +969,13 @@ func (runner *ContainerRunner) Run() (err error) { return } + runner.StartCrunchstat() + if runner.IsCancelled() { return } - err = runner.UpdateContainerRecordRunning() + err = runner.UpdateContainerRunning() if err != nil { return } @@ -825,17 +1004,27 @@ func NewContainerRunner(api IArvadosClient, cr.RunArvMount = cr.ArvMountCmd cr.MkTempDir = ioutil.TempDir cr.LogCollection = &CollectionWriter{kc, nil, sync.Mutex{}} - cr.ContainerRecord.UUID = containerUUID + cr.Container.UUID = containerUUID cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run")) cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0) + cr.loadDiscoveryVars() return cr } func main() { + statInterval := flag.Duration("crunchstat-interval", 10*time.Second, "sampling period for periodic resource usage reporting") + cgroupRoot := flag.String("cgroup-root", "/sys/fs/cgroup", "path to sysfs cgroup tree") + cgroupParent := flag.String("cgroup-parent", "docker", "name of container's parent cgroup (ignored if -cgroup-parent-subsystem is used)") + cgroupParentSubsystem := flag.String("cgroup-parent-subsystem", "", "use current cgroup for given subsystem as parent cgroup for container") + caCertsPath := flag.String("ca-certs", "", "Path to TLS root certificates") flag.Parse() containerId := flag.Arg(0) + if *caCertsPath != "" { + arvadosclient.CertFiles = []string{*caCertsPath} + } + api, err := arvadosclient.MakeArvadosClient() if err != nil { log.Fatalf("%s: %v", containerId, err) @@ -843,7 +1032,7 @@ func main() { api.Retries = 8 var kc *keepclient.KeepClient - kc, err = keepclient.MakeKeepClient(&api) + kc, err = keepclient.MakeKeepClient(api) if err != nil { log.Fatalf("%s: %v", containerId, err) } @@ -856,6 +1045,14 @@ func main() { } cr := NewContainerRunner(api, kc, docker, containerId) + cr.statInterval = *statInterval + cr.cgroupRoot = *cgroupRoot + cr.expectCgroupParent = *cgroupParent + if *cgroupParentSubsystem != "" { + p := findCgroup(*cgroupParentSubsystem) + cr.setCgroupParent = p + cr.expectCgroupParent = p + } err = cr.Run() if err != nil {