docker: oom_score_adj support (#23297)

drivers/docker/config.go

@@ -290,6 +290,13 @@ var (
 			hclspec.NewLiteral(`5`),
 		),
 
+		// oom_score_adj is the positive integer that can be used to mark the task as
+		// more likely to be OOM killed
+		"oom_score_adj": hclspec.NewDefault(
+			hclspec.NewAttr("oom_score_adj", "number", false),
+			hclspec.NewLiteral(`0`),
+		),
+
 		// the duration that the driver will wait for activity from the Docker engine during an image pull
 		// before canceling the request
 		"pull_activity_timeout": hclspec.NewDefault(
@@ -392,6 +399,7 @@ var (
 		"mounts":          hclspec.NewBlockList("mounts", mountBodySpec),
 		"network_aliases": hclspec.NewAttr("network_aliases", "list(string)", false),
 		"network_mode":    hclspec.NewAttr("network_mode", "string", false),
+		"oom_score_adj":   hclspec.NewAttr("oom_score_adj", "number", false),
 		"runtime":         hclspec.NewAttr("runtime", "string", false),
 		"pids_limit":      hclspec.NewAttr("pids_limit", "number", false),
 		"pid_mode":        hclspec.NewAttr("pid_mode", "string", false),
@@ -469,6 +477,7 @@ type TaskConfig struct {
 	Mounts                  []DockerMount      `codec:"mount"`
 	NetworkAliases          []string           `codec:"network_aliases"`
 	NetworkMode             string             `codec:"network_mode"`
+	OOMScoreAdj             int                `codec:"oom_score_adj"`
 	Runtime                 string             `codec:"runtime"`
 	PidsLimit               int64              `codec:"pids_limit"`
 	PidMode                 string             `codec:"pid_mode"`
@@ -660,6 +669,7 @@ type DriverConfig struct {
 	PullActivityTimeout           string        `codec:"pull_activity_timeout"`
 	PidsLimit                     int64         `codec:"pids_limit"`
 	pullActivityTimeoutDuration   time.Duration `codec:"-"`
+	OOMScoreAdj                   int           `codec:"oom_score_adj"`
 	ExtraLabels                   []string      `codec:"extra_labels"`
 	Logging                       LoggingConfig `codec:"logging"`
 

drivers/docker/config_test.go

@@ -314,6 +314,7 @@ config {
   ]
   network_aliases = ["redis"]
   network_mode = "host"
+  oom_score_adj = 1000
   pids_limit = 2000
 	pid_mode = "host"
 	ports = ["http", "https"]
@@ -475,6 +476,7 @@ config {
 		},
 		NetworkAliases: []string{"redis"},
 		NetworkMode:    "host",
+		OOMScoreAdj:    1000,
 		PidsLimit:      2000,
 		PidMode:        "host",
 		Ports:          []string{"http", "https"},
@@ -720,6 +722,35 @@ func TestConfig_DriverConfig_ContainerExistsAttempts(t *testing.T) {
 	}
 }
 
+func TestConfig_DriverConfig_OOMScoreAdj(t *testing.T) {
+	ci.Parallel(t)
+
+	cases := []struct {
+		name     string
+		config   string
+		expected int
+	}{
+		{
+			name:     "default",
+			config:   `{}`,
+			expected: 0,
+		},
+		{
+			name:     "set explicitly",
+			config:   `{ oom_score_adj = 1001 }`,
+			expected: 1001,
+		},
+	}
+
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			var tc DriverConfig
+			hclutils.NewConfigParser(configSpec).ParseHCL(t, "config "+c.config, &tc)
+			must.Eq(t, c.expected, tc.OOMScoreAdj)
+		})
+	}
+}
+
 func TestConfig_DriverConfig_InfraImagePullTimeout(t *testing.T) {
 	ci.Parallel(t)
 

drivers/docker/driver.go

@@ -990,8 +990,9 @@ func (d *Driver) createContainerConfig(task *drivers.TaskConfig, driverConfig *T
 	hostConfig := &docker.HostConfig{
 		// do not set cgroup parent anymore
 
-		Memory:            memory,            // hard limit
-		MemoryReservation: memoryReservation, // soft limit
+		Memory:            memory,                   // hard limit
+		MemoryReservation: memoryReservation,        // soft limit
+		OomScoreAdj:       driverConfig.OOMScoreAdj, // ignored on platforms other than linux
 
 		CPUShares:  task.Resources.LinuxResources.CPUShares,
 		CPUSetCPUs: task.Resources.LinuxResources.CpusetCpus,