Docker runtimes

drivers/docker/config.go

@@ -251,7 +251,11 @@ var (
 			hclspec.NewAttr("nvidia_runtime", "string", false),
 			hclspec.NewLiteral(`"nvidia"`),
 		),
-
+		// list of docker runtimes allowed to be used
+		"allowed_runtimes": hclspec.NewDefault(
+			hclspec.NewAttr("allowed_runtimes", "list(string)", false),
+			hclspec.NewLiteral(`["runc", "nvidia"]`),
+		),
 		// image to use when creating a network namespace parent container
 		"infra_image": hclspec.NewDefault(
 			hclspec.NewAttr("infra_image", "string", false),
@@ -341,6 +345,7 @@ var (
 		})),
 		"network_aliases": hclspec.NewAttr("network_aliases", "list(string)", false),
 		"network_mode":    hclspec.NewAttr("network_mode", "string", false),
+		"runtime":         hclspec.NewAttr("runtime", "string", false),
 		"pids_limit":      hclspec.NewAttr("pids_limit", "number", false),
 		"pid_mode":        hclspec.NewAttr("pid_mode", "string", false),
 		"port_map":        hclspec.NewAttr("port_map", "list(map(number))", false),
@@ -404,6 +409,7 @@ type TaskConfig struct {
 	Mounts            []DockerMount      `codec:"mounts"`
 	NetworkAliases    []string           `codec:"network_aliases"`
 	NetworkMode       string             `codec:"network_mode"`
+	Runtime           string             `codec:"runtime"`
 	PidsLimit         int64              `codec:"pids_limit"`
 	PidMode           string             `codec:"pid_mode"`
 	PortMap           hclutils.MapStrInt `codec:"port_map"`
@@ -572,6 +578,9 @@ type DriverConfig struct {
 	DisableLogCollection        bool          `codec:"disable_log_collection"`
 	PullActivityTimeout         string        `codec:"pull_activity_timeout"`
 	pullActivityTimeoutDuration time.Duration `codec:"-"`
+
+	AllowedRuntimesList []string            `codec:"allowed_runtimes"`
+	allowedRuntimes     map[string]struct{} `codec:"-"`
 }
 
 type AuthConfig struct {
@@ -657,6 +666,11 @@ func (d *Driver) SetConfig(c *base.Config) error {
 		d.config.pullActivityTimeoutDuration = dur
 	}
 
+	d.config.allowedRuntimes = make(map[string]struct{}, len(d.config.AllowedRuntimesList))
+	for _, r := range d.config.AllowedRuntimesList {
+		d.config.allowedRuntimes[r] = struct{}{}
+	}
+
 	if c.AgentConfig != nil {
 		d.clientConfig = c.AgentConfig.Driver
 	}

drivers/docker/config_test.go

@@ -270,6 +270,7 @@ config {
   }
   privileged = true
   readonly_rootfs = true
+  runtime = "runc"
   security_opt = [
     "credentialspec=file://gmsaUser.json"
   ],
@@ -398,6 +399,7 @@ config {
 		},
 		Privileged:     true,
 		ReadonlyRootfs: true,
+		Runtime:        "runc",
 		SecurityOpt: []string{
 			"credentialspec=file://gmsaUser.json",
 		},
@@ -551,3 +553,34 @@ func TestConfig_DriverConfig_PullActivityTimeout(t *testing.T) {
 		})
 	}
 }
+
+func TestConfig_DriverConfig_AllowedRuntimes(t *testing.T) {
+	cases := []struct {
+		name     string
+		config   string
+		expected map[string]struct{}
+	}{
+		{
+			name:     "pure default",
+			config:   `{}`,
+			expected: map[string]struct{}{"runc": struct{}{}, "nvidia": struct{}{}},
+		},
+		{
+			name:     "custom",
+			config:   `{ allowed_runtimes = ["runc", "firecracker"]}`,
+			expected: map[string]struct{}{"runc": struct{}{}, "firecracker": struct{}{}},
+		},
+	}
+
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			var tc map[string]interface{}
+			hclutils.NewConfigParser(configSpec).ParseHCL(t, "config "+c.config, &tc)
+
+			dh := dockerDriverHarness(t, tc)
+			d := dh.Impl().(*Driver)
+			require.Equal(t, c.expected, d.config.allowedRuntimes)
+		})
+	}
+
+}

drivers/docker/driver.go

@@ -739,6 +739,20 @@ func (d *Driver) createContainerConfig(task *drivers.TaskConfig, driverConfig *T
 		config.WorkingDir = driverConfig.WorkDir
 	}
 
+	containerRuntime := driverConfig.Runtime
+	if _, ok := task.DeviceEnv[nvidiaVisibleDevices]; ok {
+		if !d.gpuRuntime {
+			return c, fmt.Errorf("requested docker-runtime %q was not found", d.config.GPURuntimeName)
+		}
+		if containerRuntime != "" && containerRuntime != d.config.GPURuntimeName {
+			return c, fmt.Errorf("conflicting runtime requests: gpu runtime %q conflicts with task runtime %q", d.config.GPURuntimeName, containerRuntime)
+		}
+		containerRuntime = d.config.GPURuntimeName
+	}
+	if _, ok := d.config.allowedRuntimes[containerRuntime]; !ok && containerRuntime != "" {
+		return c, fmt.Errorf("requested runtime is not allowed: %q", containerRuntime)
+	}
+
 	hostConfig := &docker.HostConfig{
 		Memory:    task.Resources.LinuxResources.MemoryLimitBytes,
 		CPUShares: task.Resources.LinuxResources.CPUShares,
@@ -752,13 +766,8 @@ func (d *Driver) createContainerConfig(task *drivers.TaskConfig, driverConfig *T
 		VolumeDriver: driverConfig.VolumeDriver,
 
 		PidsLimit: &driverConfig.PidsLimit,
-	}
 
-	if _, ok := task.DeviceEnv[nvidiaVisibleDevices]; ok {
-		if !d.gpuRuntime {
-			return c, fmt.Errorf("requested docker-runtime %q was not found", d.config.GPURuntimeName)
-		}
-		hostConfig.Runtime = d.config.GPURuntimeName
+		Runtime: containerRuntime,
 	}
 
 	// Calculate CPU Quota

drivers/docker/driver_test.go

@@ -1029,6 +1029,29 @@ func TestDockerDriver_SecurityOptFromFile(t *testing.T) {
 	require.Contains(t, container.HostConfig.SecurityOpt[0], "reboot")
 }
 
+func TestDockerDriver_Runtime(t *testing.T) {
+	if !tu.IsCI() {
+		t.Parallel()
+	}
+	testutil.DockerCompatible(t)
+
+	task, cfg, ports := dockerTask(t)
+	defer freeport.Return(ports)
+	cfg.Runtime = "runc"
+	require.NoError(t, task.EncodeConcreteDriverConfig(cfg))
+
+	client, d, handle, cleanup := dockerSetup(t, task)
+	defer cleanup()
+	require.NoError(t, d.WaitUntilStarted(task.ID, 5*time.Second))
+
+	container, err := client.InspectContainer(handle.containerID)
+	if err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	require.Exactly(t, cfg.Runtime, container.HostConfig.Runtime)
+}
+
 func TestDockerDriver_CreateContainerConfig(t *testing.T) {
 	t.Parallel()
 
@@ -1053,6 +1076,70 @@ func TestDockerDriver_CreateContainerConfig(t *testing.T) {
 	require.Equal(t, containerName, c.Name)
 }
 
+func TestDockerDriver_CreateContainerConfig_RuntimeConflict(t *testing.T) {
+	t.Parallel()
+
+	task, cfg, ports := dockerTask(t)
+	defer freeport.Return(ports)
+	task.DeviceEnv[nvidia.NvidiaVisibleDevices] = "GPU_UUID_1"
+
+	require.NoError(t, task.EncodeConcreteDriverConfig(cfg))
+
+	dh := dockerDriverHarness(t, nil)
+	driver := dh.Impl().(*Driver)
+	driver.gpuRuntime = true
+
+	// Should error if a runtime was explicitly set that doesn't match gpu runtime
+	cfg.Runtime = "nvidia"
+	c, err := driver.createContainerConfig(task, cfg, "org/repo:0.1")
+	require.NoError(t, err)
+	require.Equal(t, "nvidia", c.HostConfig.Runtime)
+
+	cfg.Runtime = "custom"
+	_, err = driver.createContainerConfig(task, cfg, "org/repo:0.1")
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "conflicting runtime requests")
+}
+
+func TestDockerDriver_CreateContainerConfig_ChecksAllowedRuntimes(t *testing.T) {
+	t.Parallel()
+
+	dh := dockerDriverHarness(t, nil)
+	driver := dh.Impl().(*Driver)
+	driver.gpuRuntime = true
+	driver.config.allowedRuntimes = map[string]struct{}{
+		"runc":   struct{}{},
+		"custom": struct{}{},
+	}
+
+	allowedRuntime := []string{
+		"", // default always works
+		"runc",
+		"custom",
+	}
+
+	task, cfg, ports := dockerTask(t)
+	defer freeport.Return(ports)
+	require.NoError(t, task.EncodeConcreteDriverConfig(cfg))
+
+	for _, runtime := range allowedRuntime {
+		t.Run(runtime, func(t *testing.T) {
+			cfg.Runtime = runtime
+			c, err := driver.createContainerConfig(task, cfg, "org/repo:0.1")
+			require.NoError(t, err)
+			require.Equal(t, runtime, c.HostConfig.Runtime)
+		})
+	}
+
+	t.Run("not allowed: denied", func(t *testing.T) {
+		cfg.Runtime = "denied"
+		_, err := driver.createContainerConfig(task, cfg, "org/repo:0.1")
+		require.Error(t, err)
+		require.Contains(t, err.Error(), "runtime is not allowed")
+	})
+
+}
+
 func TestDockerDriver_CreateContainerConfig_User(t *testing.T) {
 	t.Parallel()
 

website/pages/docs/drivers/docker.mdx

@@ -406,6 +406,16 @@ The `docker` driver supports the following configuration in the job spec. Only
 - `readonly_rootfs` - (Optional) `true` or `false` (default). Mount
   the container's filesystem as read only.
 
+- `runtime` - (Optional) A string representing a configured runtime to pass to docker.
+  This is equivilent to the --runtime argument in the docker CLI
+  For example, to use gvisor:
+
+  ```hcl
+  config {
+    runtime = "runsc"
+  }
+  ```
+
 - `pids_limit` - (Optional) An integer value that specifies the pid limit for
   the container. Defaults to unlimited.