import (
"bytes"
- "context"
"encoding/json"
"errors"
"flag"
"git.curoverse.com/arvados.git/sdk/go/arvadosclient"
"git.curoverse.com/arvados.git/sdk/go/keepclient"
"git.curoverse.com/arvados.git/sdk/go/manifest"
+ "golang.org/x/net/context"
dockertypes "github.com/docker/docker/api/types"
dockercontainer "github.com/docker/docker/api/types/container"
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, condition dockercontainer.WaitCondition) (<-chan dockercontainer.ContainerWaitOKBody, <-chan error) {
- return proxy.Docker.ContainerWait(ctx, container, condition)
-}
-
-// 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
// container.
type ContainerRunner struct {
cmd []string
}
-// LogNodeInfo gathers node information and store it on the log for debugging
-// purposes.
-func (runner *ContainerRunner) LogNodeInfo() (err error) {
+// LogHostInfo logs info about the current host, for debugging and
+// accounting purposes. Although it's logged as "node-info", this is
+// about the environment where crunch-run is actually running, which
+// might differ from what's described in the node record (see
+// LogNodeRecord).
+func (runner *ContainerRunner) LogHostInfo() (err error) {
w := runner.NewLogWriter("node-info")
commands := []infoCommand{
}
// LogContainerRecord gets and saves the raw JSON container record from the API server
-func (runner *ContainerRunner) LogContainerRecord() (err error) {
+func (runner *ContainerRunner) LogContainerRecord() error {
+ logged, err := runner.logAPIResponse("container", "containers", map[string]interface{}{"filters": [][]string{{"uuid", "=", runner.Container.UUID}}}, nil)
+ if !logged && err == nil {
+ err = fmt.Errorf("error: no container record found for %s", runner.Container.UUID)
+ }
+ return err
+}
+
+// LogNodeRecord logs arvados#node record corresponding to the current host.
+func (runner *ContainerRunner) LogNodeRecord() error {
+ hostname := os.Getenv("SLURMD_NODENAME")
+ if hostname == "" {
+ hostname, _ = os.Hostname()
+ }
+ _, err := runner.logAPIResponse("node", "nodes", map[string]interface{}{"filters": [][]string{{"hostname", "=", hostname}}}, func(resp interface{}) {
+ // The "info" field has admin-only info when obtained
+ // with a privileged token, and should not be logged.
+ node, ok := resp.(map[string]interface{})
+ if ok {
+ delete(node, "info")
+ }
+ })
+ return err
+}
+
+func (runner *ContainerRunner) logAPIResponse(label, path string, params map[string]interface{}, munge func(interface{})) (logged bool, err error) {
w := &ArvLogWriter{
ArvClient: runner.ArvClient,
UUID: runner.Container.UUID,
- loggingStream: "container",
- writeCloser: runner.LogCollection.Open("container.json"),
+ loggingStream: label,
+ writeCloser: runner.LogCollection.Open(label + ".json"),
}
- // Get Container record JSON from the API Server
- reader, err := runner.ArvClient.CallRaw("GET", "containers", runner.Container.UUID, "", nil)
+ reader, err := runner.ArvClient.CallRaw("GET", path, "", "", arvadosclient.Dict(params))
if err != nil {
- return fmt.Errorf("While retrieving container record from the API server: %v", err)
+ return false, fmt.Errorf("error getting %s record: %v", label, err)
}
defer reader.Close()
dec := json.NewDecoder(reader)
dec.UseNumber()
- var cr map[string]interface{}
- if err = dec.Decode(&cr); err != nil {
- return fmt.Errorf("While decoding the container record JSON response: %v", err)
+ var resp map[string]interface{}
+ if err = dec.Decode(&resp); err != nil {
+ return false, fmt.Errorf("error decoding %s list response: %v", label, err)
+ }
+ items, ok := resp["items"].([]interface{})
+ if !ok {
+ return false, fmt.Errorf("error decoding %s list response: no \"items\" key in API list response", label)
+ } else if len(items) < 1 {
+ return false, nil
+ }
+ if munge != nil {
+ munge(items[0])
}
// 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)
+ if err = enc.Encode(items[0]); err != nil {
+ return false, fmt.Errorf("error logging %s record: %v", label, err)
}
err = w.Close()
if err != nil {
- return fmt.Errorf("While closing container.json log: %v", err)
+ return false, fmt.Errorf("error closing %s.json in log collection: %v", label, err)
}
- return nil
+ return true, nil
}
// AttachStreams connects the docker container stdin, stdout and stderr logs
if err != nil {
return
}
-
- // Gather and record node information
- err = runner.LogNodeInfo()
+ err = runner.LogHostInfo()
+ if err != nil {
+ return
+ }
+ err = runner.LogNodeRecord()
if err != nil {
return
}
- // Save container.json record on log collection
err = runner.LogContainerRecord()
if err != nil {
return
// API version 1.21 corresponds to Docker 1.9, which is currently the
// minimum version we want to support.
docker, dockererr := dockerclient.NewClient(dockerclient.DefaultDockerHost, "1.21", nil, nil)
- dockerClientProxy := ThinDockerClientProxy{Docker: docker}
-
- cr := NewContainerRunner(api, kc, dockerClientProxy, containerId)
+ cr := NewContainerRunner(api, kc, docker, containerId)
if dockererr != nil {
cr.CrunchLog.Printf("%s: %v", containerId, dockererr)
cr.checkBrokenNode(dockererr)
projects / arvados.git / blobdiff