15370: Re-enable docker tests.

[arvados.git] / lib / crunchrun / docker.go

diff --git a/lib/crunchrun/docker.go b/lib/crunchrun/docker.go
index ea9d448215cd1f700bce3bb2b675442c3fc6d231..eee8f1d76a7c4dd86bcf63e8958b38042a432139 100644 (file)
--- a/lib/crunchrun/docker.go
+++ b/lib/crunchrun/docker.go
@@ -8,7 +8,9 @@ import (
        "io"
        "io/ioutil"
        "os"
+       "os/exec"
        "strings"
+       "sync/atomic"
        "time"
 
        "git.arvados.org/arvados.git/sdk/go/arvados"
@@ -27,6 +29,7 @@ type dockerExecutor struct {
        watchdogInterval time.Duration
        dockerclient     *dockerclient.Client
        containerID      string
+       savedIPAddress   atomic.Value
        doneIO           chan struct{}
        errIO            error
 }
@@ -46,7 +49,20 @@ func newDockerExecutor(containerUUID string, logf func(string, ...interface{}),
        }, err
 }
 
-func (e *dockerExecutor) Runtime() string { return "docker" }
+func (e *dockerExecutor) Runtime() string {
+       v, _ := e.dockerclient.ServerVersion(context.Background())
+       info := ""
+       for _, cv := range v.Components {
+               if info != "" {
+                       info += ", "
+               }
+               info += cv.Name + " " + cv.Version
+       }
+       if info == "" {
+               info = "(unknown version)"
+       }
+       return "docker " + info
+}
 
 func (e *dockerExecutor) LoadImage(imageID string, imageTarballPath string, container arvados.Container, arvMountPoint string,
        containerClient *arvados.Client) error {
@@ -108,13 +124,12 @@ func (e *dockerExecutor) config(spec containerSpec) (dockercontainer.Config, doc
        }
        if spec.CUDADeviceCount != 0 {
                var deviceIds []string
-               for _, s := range os.Environ() {
+               if cudaVisibleDevices := os.Getenv("CUDA_VISIBLE_DEVICES"); cudaVisibleDevices != "" {
                        // If a resource manager such as slurm or LSF told
                        // us to select specific devices we need to propagate that.
-                       if strings.HasPrefix(s, "CUDA_VISIBLE_DEVICES=") {
-                               deviceIds = strings.SplitN(strings.SplitN(s, "=", 2)[1], ",")
-                       }
+                       deviceIds = strings.Split(cudaVisibleDevices, ",")
                }
+
                deviceCount := spec.CUDADeviceCount
                if len(deviceIds) > 0 {
                        // Docker won't accept both non-empty
@@ -216,7 +231,6 @@ func (e *dockerExecutor) Wait(ctx context.Context) (int, error) {
        for {
                select {
                case waitBody := <-waitOk:
-                       e.logf("Container exited with code: %v", waitBody.StatusCode)
                        // wait for stdout/stderr to complete
                        <-e.doneIO
                        return int(waitBody.StatusCode), nil
@@ -299,3 +313,34 @@ func (e *dockerExecutor) handleStdoutStderr(stdout, stderr io.Writer, reader io.
 func (e *dockerExecutor) Close() {
        e.dockerclient.ContainerRemove(context.TODO(), e.containerID, dockertypes.ContainerRemoveOptions{Force: true})
 }
+
+func (e *dockerExecutor) InjectCommand(ctx context.Context, detachKeys, username string, usingTTY bool, injectcmd []string) (*exec.Cmd, error) {
+       cmd := exec.CommandContext(ctx, "docker", "exec", "-i", "--detach-keys="+detachKeys, "--user="+username)
+       if usingTTY {
+               cmd.Args = append(cmd.Args, "-t")
+       }
+       cmd.Args = append(cmd.Args, e.containerID)
+       cmd.Args = append(cmd.Args, injectcmd...)
+       return cmd, nil
+}
+
+func (e *dockerExecutor) IPAddress() (string, error) {
+       if ip, ok := e.savedIPAddress.Load().(*string); ok {
+               return *ip, nil
+       }
+       ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(time.Minute))
+       defer cancel()
+       ctr, err := e.dockerclient.ContainerInspect(ctx, e.containerID)
+       if err != nil {
+               return "", fmt.Errorf("cannot get docker container info: %s", err)
+       }
+       ip := ctr.NetworkSettings.IPAddress
+       if ip == "" {
+               // TODO: try to enable networking if it wasn't
+               // already enabled when the container was
+               // created.
+               return "", fmt.Errorf("container has no IP address")
+       }
+       e.savedIPAddress.Store(&ip)
+       return ip, nil
+}