X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/76f82a555481de947be1ee70530e61266b7dcbf7..469ea187586ea8017e26874c2d80414ce7571fae:/lib/crunchrun/docker.go

diff --git a/lib/crunchrun/docker.go b/lib/crunchrun/docker.go
index ab00273ef3..eee8f1d76a 100644
--- 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 {
@@ -107,10 +123,40 @@ func (e *dockerExecutor) config(spec containerSpec) (dockercontainer.Config, doc
 		},
 	}
 	if spec.CUDADeviceCount != 0 {
+		var deviceIds []string
+		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.
+			deviceIds = strings.Split(cudaVisibleDevices, ",")
+		}
+
+		deviceCount := spec.CUDADeviceCount
+		if len(deviceIds) > 0 {
+			// Docker won't accept both non-empty
+			// DeviceIDs and a non-zero Count
+			//
+			// (it turns out "Count" is a dumb fallback
+			// that just allocates device 0, 1, 2, ...,
+			// Count-1)
+			deviceCount = 0
+		}
+
+		// Capabilities are confusing.  The driver has generic
+		// capabilities "gpu" and "nvidia" but then there's
+		// additional capabilities "compute" and "utility"
+		// that are passed to nvidia-container-cli.
+		//
+		// "compute" means include the CUDA libraries and
+		// "utility" means include the CUDA utility programs
+		// (like nvidia-smi).
+		//
+		// https://github.com/moby/moby/blob/7b9275c0da707b030e62c96b679a976f31f929d3/daemon/nvidia_linux.go#L37
+		// https://github.com/containerd/containerd/blob/main/contrib/nvidia/nvidia.go
 		hostCfg.Resources.DeviceRequests = append(hostCfg.Resources.DeviceRequests, dockercontainer.DeviceRequest{
 			Driver:       "nvidia",
-			Count:        spec.CUDADeviceCount,
-			Capabilities: [][]string{[]string{"gpu", "nvidia", "compute"}},
+			Count:        deviceCount,
+			DeviceIDs:    deviceIds,
+			Capabilities: [][]string{[]string{"gpu", "nvidia", "compute", "utility"}},
 		})
 	}
 	for path, mount := range spec.BindMounts {
@@ -185,7 +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
@@ -268,3 +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
+}