package main
import (
+ "bytes"
+ "context"
"encoding/json"
"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"
- "github.com/curoverse/dockerclient"
"io"
"io/ioutil"
"log"
"sync"
"syscall"
"time"
+
+ "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"
+
+ dockertypes "github.com/docker/docker/api/types"
+ dockercontainer "github.com/docker/docker/api/types/container"
+ dockernetwork "github.com/docker/docker/api/types/network"
+ dockerclient "github.com/docker/docker/client"
)
// IArvadosClient is the minimal Arvados API methods used by crunch-run.
Get(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
Update(resourceType string, uuid string, parameters arvadosclient.Dict, output interface{}) error
Call(method, resourceType, uuid, action string, parameters arvadosclient.Dict, output interface{}) error
+ CallRaw(method string, resourceType string, uuid string, action string, parameters arvadosclient.Dict) (reader io.ReadCloser, err error)
Discovery(key string) (interface{}, error)
}
// ThinDockerClient is the minimal Docker client interface used by crunch-run.
type ThinDockerClient interface {
- StopContainer(id string, timeout int) error
- InspectImage(id string) (*dockerclient.ImageInfo, error)
- LoadImage(reader io.Reader) error
- CreateContainer(config *dockerclient.ContainerConfig, name string, authConfig *dockerclient.AuthConfig) (string, error)
- StartContainer(id string, config *dockerclient.HostConfig) error
- AttachContainer(id string, options *dockerclient.AttachOptions) (io.ReadCloser, error)
- Wait(id string) <-chan dockerclient.WaitResult
- RemoveImage(name string, force bool) ([]*dockerclient.ImageDelete, error)
+ ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error)
+ ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
+ networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error)
+ ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error
+ ContainerStop(ctx context.Context, container string, timeout *time.Duration) error
+ ContainerWait(ctx context.Context, container string) (int64, error)
+ ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error)
+ ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error)
+ ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error)
+}
+
+// ThinDockerClientProxy is a proxy implementation of ThinDockerClient
+// that executes the docker requests on dockerclient.Client
+type ThinDockerClientProxy struct {
+ Docker *dockerclient.Client
+}
+
+// ContainerAttach invokes dockerclient.Client.ContainerAttach
+func (proxy ThinDockerClientProxy) ContainerAttach(ctx context.Context, container string, options dockertypes.ContainerAttachOptions) (dockertypes.HijackedResponse, error) {
+ return proxy.Docker.ContainerAttach(ctx, container, options)
+}
+
+// ContainerCreate invokes dockerclient.Client.ContainerCreate
+func (proxy ThinDockerClientProxy) ContainerCreate(ctx context.Context, config *dockercontainer.Config, hostConfig *dockercontainer.HostConfig,
+ networkingConfig *dockernetwork.NetworkingConfig, containerName string) (dockercontainer.ContainerCreateCreatedBody, error) {
+ return proxy.Docker.ContainerCreate(ctx, config, hostConfig, networkingConfig, containerName)
+}
+
+// ContainerStart invokes dockerclient.Client.ContainerStart
+func (proxy ThinDockerClientProxy) ContainerStart(ctx context.Context, container string, options dockertypes.ContainerStartOptions) error {
+ return proxy.Docker.ContainerStart(ctx, container, options)
+}
+
+// ContainerStop invokes dockerclient.Client.ContainerStop
+func (proxy ThinDockerClientProxy) ContainerStop(ctx context.Context, container string, timeout *time.Duration) error {
+ return proxy.Docker.ContainerStop(ctx, container, timeout)
+}
+
+// ContainerWait invokes dockerclient.Client.ContainerWait
+func (proxy ThinDockerClientProxy) ContainerWait(ctx context.Context, container string) (int64, error) {
+ return proxy.Docker.ContainerWait(ctx, container)
+}
+
+// ImageInspectWithRaw invokes dockerclient.Client.ImageInspectWithRaw
+func (proxy ThinDockerClientProxy) ImageInspectWithRaw(ctx context.Context, image string) (dockertypes.ImageInspect, []byte, error) {
+ return proxy.Docker.ImageInspectWithRaw(ctx, image)
+}
+
+// ImageLoad invokes dockerclient.Client.ImageLoad
+func (proxy ThinDockerClientProxy) ImageLoad(ctx context.Context, input io.Reader, quiet bool) (dockertypes.ImageLoadResponse, error) {
+ return proxy.Docker.ImageLoad(ctx, input, quiet)
+}
+
+// ImageRemove invokes dockerclient.Client.ImageRemove
+func (proxy ThinDockerClientProxy) ImageRemove(ctx context.Context, image string, options dockertypes.ImageRemoveOptions) ([]dockertypes.ImageDeleteResponseItem, error) {
+ return proxy.Docker.ImageRemove(ctx, image, options)
}
// ContainerRunner is the main stateful struct used for a single execution of a
ArvClient IArvadosClient
Kc IKeepClient
arvados.Container
- dockerclient.ContainerConfig
- dockerclient.HostConfig
+ ContainerConfig dockercontainer.Config
+ dockercontainer.HostConfig
token string
ContainerID string
ExitCode *int
loggingDone chan bool
CrunchLog *ThrottledLogger
Stdout io.WriteCloser
- Stderr *ThrottledLogger
+ Stderr io.WriteCloser
LogCollection *CollectionWriter
LogsPDH *string
RunArvMount
HostOutputDir string
CleanupTempDir []string
Binds []string
+ Volumes map[string]struct{}
OutputPDH *string
- CancelLock sync.Mutex
- Cancelled bool
SigChan chan os.Signal
ArvMountExit chan error
finalState string
- trashLifetime time.Duration
statLogger io.WriteCloser
statReporter *crunchstat.Reporter
// parent to be X" feature even on sites where the "specify
// cgroup parent" feature breaks.
setCgroupParent string
+
+ cStateLock sync.Mutex
+ cStarted bool // StartContainer() succeeded
+ cCancelled bool // StopContainer() invoked
+
+ enableNetwork string // one of "default" or "always"
+ networkMode string // passed through to HostConfig.NetworkMode
}
// SetupSignals sets up signal handling to gracefully terminate the underlying
signal.Notify(runner.SigChan, syscall.SIGINT)
signal.Notify(runner.SigChan, syscall.SIGQUIT)
- go func(sig <-chan os.Signal) {
- for range sig {
- if !runner.Cancelled {
- runner.CancelLock.Lock()
- runner.Cancelled = true
- if runner.ContainerID != "" {
- runner.Docker.StopContainer(runner.ContainerID, 10)
- }
- runner.CancelLock.Unlock()
- }
- }
+ go func(sig chan os.Signal) {
+ <-sig
+ runner.stop()
+ signal.Stop(sig)
}(runner.SigChan)
}
+// stop the underlying Docker container.
+func (runner *ContainerRunner) stop() {
+ runner.cStateLock.Lock()
+ defer runner.cStateLock.Unlock()
+ if runner.cCancelled {
+ return
+ }
+ runner.cCancelled = true
+ if runner.cStarted {
+ timeout := time.Duration(10)
+ err := runner.Docker.ContainerStop(context.TODO(), runner.ContainerID, &(timeout))
+ if err != nil {
+ log.Printf("StopContainer failed: %s", err)
+ }
+ }
+}
+
// LoadImage determines the docker image id from the container record and
// checks if it is available in the local Docker image store. If not, it loads
// the image from Keep.
runner.CrunchLog.Printf("Using Docker image id '%s'", imageID)
- _, err = runner.Docker.InspectImage(imageID)
+ _, _, err = runner.Docker.ImageInspectWithRaw(context.TODO(), imageID)
if err != nil {
runner.CrunchLog.Print("Loading Docker image from keep")
return fmt.Errorf("While creating ManifestFileReader for container image: %v", err)
}
- err = runner.Docker.LoadImage(readCloser)
+ response, err := runner.Docker.ImageLoad(context.TODO(), readCloser, false)
if err != nil {
return fmt.Errorf("While loading container image into Docker: %v", err)
}
+ response.Body.Close()
} else {
runner.CrunchLog.Print("Docker image is available")
}
collectionPaths := []string{}
runner.Binds = nil
+ runner.Volumes = make(map[string]struct{})
needCertMount := true
var binds []string
for _, bind := range binds {
mnt := runner.Container.Mounts[bind]
- if bind == "stdout" {
+ if bind == "stdout" || bind == "stderr" {
// Is it a "file" mount kind?
if mnt.Kind != "file" {
- return fmt.Errorf("Unsupported mount kind '%s' for stdout. Only 'file' is supported.", mnt.Kind)
+ return fmt.Errorf("Unsupported mount kind '%s' for %s. Only 'file' is supported.", mnt.Kind, bind)
}
// Does path start with OutputPath?
prefix += "/"
}
if !strings.HasPrefix(mnt.Path, prefix) {
- return fmt.Errorf("Stdout path does not start with OutputPath: %s, %s", mnt.Path, prefix)
+ return fmt.Errorf("%s path does not start with OutputPath: %s, %s", strings.Title(bind), mnt.Path, prefix)
+ }
+ }
+
+ if bind == "stdin" {
+ // Is it a "collection" mount kind?
+ if mnt.Kind != "collection" && mnt.Kind != "json" {
+ return fmt.Errorf("Unsupported mount kind '%s' for stdin. Only 'collection' or 'json' are supported.", mnt.Kind)
}
}
}
switch {
- case mnt.Kind == "collection":
+ case mnt.Kind == "collection" && bind != "stdin":
var src string
if mnt.UUID != "" && mnt.PortableDataHash != "" {
return fmt.Errorf("Cannot specify both 'uuid' and 'portable_data_hash' for a collection mount")
if mnt.Writable {
return fmt.Errorf("Can never write to a collection specified by portable data hash")
}
+ idx := strings.Index(mnt.PortableDataHash, "/")
+ if idx > 0 {
+ mnt.Path = path.Clean(mnt.PortableDataHash[idx:])
+ mnt.PortableDataHash = mnt.PortableDataHash[0:idx]
+ runner.Container.Mounts[bind] = mnt
+ }
src = fmt.Sprintf("%s/by_id/%s", runner.ArvMountPoint, mnt.PortableDataHash)
+ if mnt.Path != "" && mnt.Path != "." {
+ if strings.HasPrefix(mnt.Path, "./") {
+ mnt.Path = mnt.Path[2:]
+ } else if strings.HasPrefix(mnt.Path, "/") {
+ mnt.Path = mnt.Path[1:]
+ }
+ src += "/" + mnt.Path
+ }
} else {
src = fmt.Sprintf("%s/tmp%d", runner.ArvMountPoint, tmpcount)
arvMountCmd = append(arvMountCmd, "--mount-tmp")
runner.Binds = append(runner.Binds, fmt.Sprintf("%s:%s", runner.HostOutputDir, bind))
case mnt.Kind == "tmp":
- runner.Binds = append(runner.Binds, bind)
+ runner.Volumes[bind] = struct{}{}
case mnt.Kind == "json":
jsondata, err := json.Marshal(mnt.Content)
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.
+type infoCommand struct {
+ label string
+ cmd []string
+}
+
+// Gather node information and store it on the log for debugging
+// purposes.
+func (runner *ContainerRunner) LogNodeInfo() (err error) {
+ w := runner.NewLogWriter("node-info")
+ logger := log.New(w, "node-info", 0)
+
+ commands := []infoCommand{
+ infoCommand{
+ label: "Host Information",
+ cmd: []string{"uname", "-a"},
+ },
+ infoCommand{
+ label: "CPU Information",
+ cmd: []string{"cat", "/proc/cpuinfo"},
+ },
+ infoCommand{
+ label: "Memory Information",
+ cmd: []string{"cat", "/proc/meminfo"},
+ },
+ infoCommand{
+ label: "Disk Space",
+ cmd: []string{"df", "-m", "/", os.TempDir()},
+ },
+ infoCommand{
+ label: "Disk INodes",
+ cmd: []string{"df", "-i", "/", os.TempDir()},
+ },
+ }
+
+ // Run commands with informational output to be logged.
+ var out []byte
+ for _, command := range commands {
+ out, err = exec.Command(command.cmd[0], command.cmd[1:]...).CombinedOutput()
+ if err != nil {
+ return fmt.Errorf("While running command %q: %v",
+ command.cmd, err)
+ }
+ logger.Println(command.label)
+ for _, line := range strings.Split(string(out), "\n") {
+ logger.Println(" ", line)
+ }
+ }
+
+ err = w.Close()
+ if err != nil {
+ return fmt.Errorf("While closing node-info logs: %v", err)
+ }
+ return nil
+}
+
+// Get and save the raw JSON container record from the API server
+func (runner *ContainerRunner) LogContainerRecord() (err error) {
+ w := &ArvLogWriter{
+ ArvClient: runner.ArvClient,
+ UUID: runner.Container.UUID,
+ loggingStream: "container",
+ writeCloser: runner.LogCollection.Open("container.json"),
+ }
+
+ // Get Container record JSON from the API Server
+ reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
+ if err != nil {
+ return fmt.Errorf("While retrieving container record from the API server: %v", err)
+ }
+ defer reader.Close()
+ // Read the API server response as []byte
+ json_bytes, err := ioutil.ReadAll(reader)
+ if err != nil {
+ return fmt.Errorf("While reading container record API server response: %v", err)
+ }
+ // Decode the JSON []byte
+ var cr map[string]interface{}
+ if err = json.Unmarshal(json_bytes, &cr); err != nil {
+ return fmt.Errorf("While decoding the container record JSON response: %v", err)
+ }
+ // Re-encode it using indentation to improve readability
+ enc := json.NewEncoder(w)
+ enc.SetIndent("", " ")
+ if err = enc.Encode(cr); err != nil {
+ return fmt.Errorf("While logging the JSON container record: %v", err)
+ }
+ err = w.Close()
+ if err != nil {
+ return fmt.Errorf("While closing container.json log: %v", err)
+ }
+ return nil
+}
+
+// AttachStreams connects the docker container stdin, stdout and stderr logs
+// to the Arvados logger which logs to Keep and the API server logs table.
func (runner *ContainerRunner) AttachStreams() (err error) {
runner.CrunchLog.Print("Attaching container streams")
- var containerReader io.Reader
- containerReader, err = runner.Docker.AttachContainer(runner.ContainerID,
- &dockerclient.AttachOptions{Stream: true, Stdout: true, Stderr: true})
+ // If stdin mount is provided, attach it to the docker container
+ var stdinRdr keepclient.Reader
+ var stdinJson []byte
+ if stdinMnt, ok := runner.Container.Mounts["stdin"]; ok {
+ if stdinMnt.Kind == "collection" {
+ var stdinColl arvados.Collection
+ collId := stdinMnt.UUID
+ if collId == "" {
+ collId = stdinMnt.PortableDataHash
+ }
+ err = runner.ArvClient.Get("collections", collId, nil, &stdinColl)
+ if err != nil {
+ return fmt.Errorf("While getting stding collection: %v", err)
+ }
+
+ stdinRdr, err = runner.Kc.ManifestFileReader(manifest.Manifest{Text: stdinColl.ManifestText}, stdinMnt.Path)
+ if os.IsNotExist(err) {
+ return fmt.Errorf("stdin collection path not found: %v", stdinMnt.Path)
+ } else if err != nil {
+ return fmt.Errorf("While getting stdin collection path %v: %v", stdinMnt.Path, err)
+ }
+ } else if stdinMnt.Kind == "json" {
+ stdinJson, err = json.Marshal(stdinMnt.Content)
+ if err != nil {
+ return fmt.Errorf("While encoding stdin json data: %v", err)
+ }
+ }
+ }
+
+ stdinUsed := stdinRdr != nil || len(stdinJson) != 0
+ response, err := runner.Docker.ContainerAttach(context.TODO(), runner.ContainerID,
+ dockertypes.ContainerAttachOptions{Stream: true, Stdin: stdinUsed, Stdout: true, Stderr: true})
if err != nil {
return fmt.Errorf("While attaching container stdout/stderr streams: %v", err)
}
runner.loggingDone = make(chan bool)
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]
- if subdirs != "" {
- st, err := os.Stat(runner.HostOutputDir)
- if err != nil {
- return fmt.Errorf("While Stat on temp dir: %v", err)
- }
- stdoutPath := path.Join(runner.HostOutputDir, subdirs)
- err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
- if err != nil {
- return fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
- }
- }
- }
- stdoutFile, err := os.Create(path.Join(runner.HostOutputDir, stdoutPath))
+ stdoutFile, err := runner.getStdoutFile(stdoutMnt.Path)
if err != nil {
- return fmt.Errorf("While creating stdout file: %v", err)
+ return err
}
runner.Stdout = stdoutFile
} else {
runner.Stdout = NewThrottledLogger(runner.NewLogWriter("stdout"))
}
- runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
- go runner.ProcessDockerAttach(containerReader)
+ if stderrMnt, ok := runner.Container.Mounts["stderr"]; ok {
+ stderrFile, err := runner.getStdoutFile(stderrMnt.Path)
+ if err != nil {
+ return err
+ }
+ runner.Stderr = stderrFile
+ } else {
+ runner.Stderr = NewThrottledLogger(runner.NewLogWriter("stderr"))
+ }
+
+ if stdinRdr != nil {
+ go func() {
+ _, err := io.Copy(response.Conn, stdinRdr)
+ if err != nil {
+ runner.CrunchLog.Print("While writing stdin collection to docker container %q", err)
+ runner.stop()
+ }
+ stdinRdr.Close()
+ response.CloseWrite()
+ }()
+ } else if len(stdinJson) != 0 {
+ go func() {
+ _, err := io.Copy(response.Conn, bytes.NewReader(stdinJson))
+ if err != nil {
+ runner.CrunchLog.Print("While writing stdin json to docker container %q", err)
+ runner.stop()
+ }
+ response.CloseWrite()
+ }()
+ }
+
+ go runner.ProcessDockerAttach(response.Reader)
return nil
}
+func (runner *ContainerRunner) getStdoutFile(mntPath string) (*os.File, error) {
+ stdoutPath := mntPath[len(runner.Container.OutputPath):]
+ index := strings.LastIndex(stdoutPath, "/")
+ if index > 0 {
+ subdirs := stdoutPath[:index]
+ if subdirs != "" {
+ st, err := os.Stat(runner.HostOutputDir)
+ if err != nil {
+ return nil, fmt.Errorf("While Stat on temp dir: %v", err)
+ }
+ stdoutPath := path.Join(runner.HostOutputDir, subdirs)
+ err = os.MkdirAll(stdoutPath, st.Mode()|os.ModeSetgid|0777)
+ if err != nil {
+ return nil, fmt.Errorf("While MkdirAll %q: %v", stdoutPath, err)
+ }
+ }
+ }
+ stdoutFile, err := os.Create(path.Join(runner.HostOutputDir, stdoutPath))
+ if err != nil {
+ return nil, fmt.Errorf("While creating file %q: %v", stdoutPath, err)
+ }
+
+ return stdoutFile, nil
+}
+
// CreateContainer creates the docker container.
func (runner *ContainerRunner) CreateContainer() error {
runner.CrunchLog.Print("Creating Docker container")
for k, v := range runner.Container.Environment {
runner.ContainerConfig.Env = append(runner.ContainerConfig.Env, k+"="+v)
}
+
+ runner.ContainerConfig.Volumes = runner.Volumes
+
+ runner.HostConfig = dockercontainer.HostConfig{
+ Binds: runner.Binds,
+ Cgroup: dockercontainer.CgroupSpec(runner.setCgroupParent),
+ LogConfig: dockercontainer.LogConfig{
+ Type: "none",
+ },
+ }
+
if wantAPI := runner.Container.RuntimeConstraints.API; wantAPI != nil && *wantAPI {
tok, err := runner.ContainerToken()
if err != nil {
"ARVADOS_API_HOST="+os.Getenv("ARVADOS_API_HOST"),
"ARVADOS_API_HOST_INSECURE="+os.Getenv("ARVADOS_API_HOST_INSECURE"),
)
- runner.ContainerConfig.NetworkDisabled = false
+ runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
} else {
- runner.ContainerConfig.NetworkDisabled = true
+ if runner.enableNetwork == "always" {
+ runner.HostConfig.NetworkMode = dockercontainer.NetworkMode(runner.networkMode)
+ } else {
+ runner.HostConfig.NetworkMode = dockercontainer.NetworkMode("none")
+ }
}
- var err error
- runner.ContainerID, err = runner.Docker.CreateContainer(&runner.ContainerConfig, "", nil)
+ _, stdinUsed := runner.Container.Mounts["stdin"]
+ runner.ContainerConfig.OpenStdin = stdinUsed
+ runner.ContainerConfig.StdinOnce = stdinUsed
+ runner.ContainerConfig.AttachStdin = stdinUsed
+ runner.ContainerConfig.AttachStdout = true
+ runner.ContainerConfig.AttachStderr = true
+
+ createdBody, err := runner.Docker.ContainerCreate(context.TODO(), &runner.ContainerConfig, &runner.HostConfig, nil, runner.Container.UUID)
if err != nil {
return fmt.Errorf("While creating container: %v", err)
}
- runner.HostConfig = dockerclient.HostConfig{
- Binds: runner.Binds,
- CgroupParent: runner.setCgroupParent,
- LogConfig: dockerclient.LogConfig{
- Type: "none",
- },
- }
+ runner.ContainerID = createdBody.ID
return runner.AttachStreams()
}
// StartContainer starts the docker container created by CreateContainer.
func (runner *ContainerRunner) StartContainer() error {
runner.CrunchLog.Printf("Starting Docker container id '%s'", runner.ContainerID)
- err := runner.Docker.StartContainer(runner.ContainerID, &runner.HostConfig)
+ runner.cStateLock.Lock()
+ defer runner.cStateLock.Unlock()
+ if runner.cCancelled {
+ return ErrCancelled
+ }
+ err := runner.Docker.ContainerStart(context.TODO(), runner.ContainerID,
+ dockertypes.ContainerStartOptions{})
if err != nil {
return fmt.Errorf("could not start container: %v", err)
}
+ runner.cStarted = true
return nil
}
func (runner *ContainerRunner) WaitFinish() error {
runner.CrunchLog.Print("Waiting for container to finish")
- result := runner.Docker.Wait(runner.ContainerID)
- wr := <-result
- if wr.Error != nil {
- return fmt.Errorf("While waiting for container to finish: %v", wr.Error)
+ waitDocker, err := runner.Docker.ContainerWait(context.TODO(), runner.ContainerID)
+ if err != nil {
+ return fmt.Errorf("container wait: %v", err)
+ }
+
+ runner.CrunchLog.Printf("Container exited with code: %v", waitDocker)
+ code := int(waitDocker)
+ runner.ExitCode = &code
+
+ waitMount := runner.ArvMountExit
+ select {
+ case err := <-waitMount:
+ runner.CrunchLog.Printf("arv-mount exited before container finished: %v", err)
+ waitMount = nil
+ runner.stop()
+ default:
}
- runner.ExitCode = &wr.ExitCode
// wait for stdout/stderr to complete
<-runner.loggingDone
_, err = os.Stat(collectionMetafile)
if err != nil {
// Regular directory
- cw := CollectionWriter{runner.Kc, nil, sync.Mutex{}}
+ cw := CollectionWriter{0, runner.Kc, nil, nil, sync.Mutex{}}
manifestText, err = cw.WriteTree(runner.HostOutputDir, runner.CrunchLog.Logger)
if err != nil {
return fmt.Errorf("While uploading output files: %v", err)
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 {
// Save output
var response arvados.Collection
- manifestText := manifest.Manifest{Text: manifestText}.NormalizedText()
+ manifest := manifest.Manifest{Text: manifestText}
+ manifestText = manifest.Extract(".", ".").Text
err = runner.ArvClient.Create("collections",
arvadosclient.Dict{
+ "ensure_unique_name": true,
"collection": arvadosclient.Dict{
- "trash_at": time.Now().Add(runner.trashLifetime).Format(time.RFC3339),
+ "is_trashed": true,
"name": "output for " + runner.Container.UUID,
"manifest_text": manifestText}},
&response)
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)
+ mft := manifest.Manifest{Text: collection.ManifestText}
+ extracted := mft.Extract(mnt.Path, bindSuffix)
+ if extracted.Err != nil {
+ return "", fmt.Errorf("Error parsing manifest for %v: %v", mnt.PortableDataHash, extracted.Err.Error())
}
- runner.trashLifetime = time.Duration(tl.(float64)) * time.Second
+ return extracted.Text, nil
}
func (runner *ContainerRunner) CleanupDirs() {
// 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.Container.UUID,
- "crunch-run", nil})
+ runner.CrunchLog = NewThrottledLogger(&ArvLogWriter{ArvClient: runner.ArvClient,
+ UUID: runner.Container.UUID, loggingStream: "crunch-run", writeCloser: nil})
if runner.LogsPDH != nil {
// If we have already assigned something to LogsPDH,
var response arvados.Collection
err = runner.ArvClient.Create("collections",
arvadosclient.Dict{
+ "ensure_unique_name": true,
"collection": arvadosclient.Dict{
- "trash_at": time.Now().Add(runner.trashLifetime).Format(time.RFC3339),
+ "is_trashed": true,
"name": "logs for " + runner.Container.UUID,
"manifest_text": mt}},
&response)
// UpdateContainerRunning updates the container state to "Running"
func (runner *ContainerRunner) UpdateContainerRunning() error {
- runner.CancelLock.Lock()
- defer runner.CancelLock.Unlock()
- if runner.Cancelled {
+ runner.cStateLock.Lock()
+ defer runner.cStateLock.Unlock()
+ if runner.cCancelled {
return ErrCancelled
}
return runner.ArvClient.Update("containers", runner.Container.UUID,
// IsCancelled returns the value of Cancelled, with goroutine safety.
func (runner *ContainerRunner) IsCancelled() bool {
- runner.CancelLock.Lock()
- defer runner.CancelLock.Unlock()
- return runner.Cancelled
+ runner.cStateLock.Lock()
+ defer runner.cStateLock.Unlock()
+ return runner.cCancelled
}
// NewArvLogWriter creates an ArvLogWriter
func (runner *ContainerRunner) NewArvLogWriter(name string) io.WriteCloser {
- return &ArvLogWriter{runner.ArvClient, runner.Container.UUID, name, runner.LogCollection.Open(name + ".txt")}
+ return &ArvLogWriter{ArvClient: runner.ArvClient, UUID: runner.Container.UUID, loggingStream: name,
+ writeCloser: runner.LogCollection.Open(name + ".txt")}
}
// Run the full container lifecycle.
return
}
+ // Gather and record node information
+ err = runner.LogNodeInfo()
+ if err != nil {
+ return
+ }
+ // Save container.json record on log collection
+ err = runner.LogContainerRecord()
+ if err != nil {
+ return
+ }
+
runner.StartCrunchstat()
if runner.IsCancelled() {
cr.NewLogWriter = cr.NewArvLogWriter
cr.RunArvMount = cr.ArvMountCmd
cr.MkTempDir = ioutil.TempDir
- cr.LogCollection = &CollectionWriter{kc, nil, sync.Mutex{}}
+ cr.LogCollection = &CollectionWriter{0, kc, nil, nil, sync.Mutex{}}
cr.Container.UUID = containerUUID
cr.CrunchLog = NewThrottledLogger(cr.NewLogWriter("crunch-run"))
cr.CrunchLog.Immediate = log.New(os.Stderr, containerUUID+" ", 0)
- cr.loadDiscoveryVars()
+
+ loadLogThrottleParams(api)
+
return cr
}
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")
+ enableNetwork := flag.String("container-enable-networking", "default",
+ `Specify if networking should be enabled for container. One of 'default', 'always':
+ default: only enable networking if container requests it.
+ always: containers always have networking enabled
+ `)
+ networkMode := flag.String("container-network-mode", "default",
+ `Set networking mode for container. Corresponds to Docker network mode (--net).
+ `)
flag.Parse()
containerId := flag.Arg(0)
}
kc.Retries = 4
- var docker *dockerclient.DockerClient
- docker, err = dockerclient.NewDockerClient("unix:///var/run/docker.sock", nil)
+ var docker *dockerclient.Client
+ // API version 1.21 corresponds to Docker 1.9, which is currently the
+ // minimum version we want to support.
+ docker, err = dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
if err != nil {
log.Fatalf("%s: %v", containerId, err)
}
- cr := NewContainerRunner(api, kc, docker, containerId)
+ dockerClientProxy := ThinDockerClientProxy{Docker: docker}
+
+ cr := NewContainerRunner(api, kc, dockerClientProxy, containerId)
cr.statInterval = *statInterval
cr.cgroupRoot = *cgroupRoot
cr.expectCgroupParent = *cgroupParent
+ cr.enableNetwork = *enableNetwork
+ cr.networkMode = *networkMode
if *cgroupParentSubsystem != "" {
p := findCgroup(*cgroupParentSubsystem)
cr.setCgroupParent = p
projects / arvados.git / blobdiff