drivers/raw_exec: enable configuring raw_exec task to have no memory …

…limit (#19670)

* drivers/raw_exec: enable configuring raw_exec task to have no memory limit

This PR makes it possible to configure a raw_exec task to not have an
upper memory limit, which is how the driver would behave pre-1.7.

This is done by setting memory_max = -1. The cluster (or node pool) must
have memory oversubscription enabled.

* cl: add cl

.changelog/19670.txt

@@ -0,0 +1,3 @@
+```release-note:improvement
+drivers: Enable configuring a raw_exec task to not have an upper memory limit
+```

drivers/shared/executor/executor_universal_linux.go

@@ -1,6 +1,8 @@
 // Copyright (c) HashiCorp, Inc.
 // SPDX-License-Identifier: MPL-2.0
 
+//go:build linux
+
 package executor
 
 import (
@@ -20,6 +22,12 @@ import (
 	"golang.org/x/sys/unix"
 )
 
+const (
+	// memoryNoLimit is a sentinel value for memory_max that indicates the
+	// raw_exec driver should not enforce a maximum memory limit
+	memoryNoLimit = -1
+)
+
 // setCmdUser takes a user id as a string and looks up the user, and sets the command
 // to execute as that user.
 func setCmdUser(cmd *exec.Cmd, userid string) error {
@@ -226,7 +234,11 @@ func (e *UniversalExecutor) configureCG2(cgroup string, command *ExecCommand) {
 	// write memory cgroup files
 	memHard, memSoft := e.computeMemory(command)
 	ed := cgroupslib.OpenPath(cgroup)
-	_ = ed.Write("memory.max", strconv.FormatInt(memHard, 10))
+	if memHard == memoryNoLimit {
+		_ = ed.Write("memory.max", "max")
+	} else {
+		_ = ed.Write("memory.max", strconv.FormatInt(memHard, 10))
+	}
 	if memSoft > 0 {
 		ed = cgroupslib.OpenPath(cgroup)
 		_ = ed.Write("memory.low", strconv.FormatInt(memSoft, 10))
@@ -264,17 +276,29 @@ func (*UniversalExecutor) computeCPU(command *ExecCommand) uint64 {
 	return cpuWeight
 }
 
+func mbToBytes(n int64) int64 {
+	return n * 1024 * 1024
+}
+
 // computeMemory returns the hard and soft memory limits for the task
 func (*UniversalExecutor) computeMemory(command *ExecCommand) (int64, int64) {
 	mem := command.Resources.NomadResources.Memory
 	memHard, memSoft := mem.MemoryMaxMB, mem.MemoryMB
-	if memHard <= 0 {
+
+	switch memHard {
+	case 0:
+		// typical case where 'memory' is the hard limit
 		memHard = mem.MemoryMB
-		memSoft = 0
+		return mbToBytes(memHard), 0
+	case memoryNoLimit:
+		// special oversub case where 'memory' is soft limit and there is no
+		// hard limit - helping re-create old raw_exec behavior
+		return memoryNoLimit, mbToBytes(memSoft)
+	default:
+		// typical oversub case where 'memory' is soft limit and 'memory_max'
+		// is hard limit
+		return mbToBytes(memHard), mbToBytes(memSoft)
 	}
-	memHardBytes := memHard * 1024 * 1024
-	memSoftBytes := memSoft * 1024 * 1024
-	return memHardBytes, memSoftBytes
 }
 
 // withNetworkIsolation calls the passed function the network namespace `spec`

drivers/shared/executor/executor_universal_linux_test.go

@@ -0,0 +1,68 @@
+// Copyright (c) HashiCorp, Inc.
+// SPDX-License-Identifier: MPL-2.0
+
+//go:build linux
+
+package executor
+
+import (
+	"fmt"
+	"testing"
+
+	"github.com/hashicorp/nomad/nomad/structs"
+	"github.com/hashicorp/nomad/plugins/drivers"
+	"github.com/shoenig/test/must"
+)
+
+func Test_computeMemory(t *testing.T) {
+	cases := []struct {
+		memory    int64
+		memoryMax int64
+		expSoft   int64
+		expHard   int64
+	}{
+		{
+			// typical case; only 'memory' is set and that is used as the hard
+			// memory limit
+			memory:    100,
+			memoryMax: 0,
+			expSoft:   0,
+			expHard:   mbToBytes(100),
+		},
+		{
+			// oversub case; both 'memory' and 'memory_max' are set and used as
+			// the soft and hard memory limits
+			memory:    100,
+			memoryMax: 200,
+			expSoft:   mbToBytes(100),
+			expHard:   mbToBytes(200),
+		},
+		{
+			// special oversub case; 'memory' is set and 'memory_max' is set to
+			// -1; which indicates there should be no hard limit (i.e. -1 / max)
+			memory:    100,
+			memoryMax: memoryNoLimit,
+			expSoft:   mbToBytes(100),
+			expHard:   memoryNoLimit,
+		},
+	}
+
+	for _, tc := range cases {
+		name := fmt.Sprintf("(%d,%d)", tc.memory, tc.memoryMax)
+		t.Run(name, func(t *testing.T) {
+			command := &ExecCommand{
+				Resources: &drivers.Resources{
+					NomadResources: &structs.AllocatedTaskResources{
+						Memory: structs.AllocatedMemoryResources{
+							MemoryMB:    tc.memory,
+							MemoryMaxMB: tc.memoryMax,
+						},
+					},
+				},
+			}
+			hard, soft := (*UniversalExecutor)(nil).computeMemory(command)
+			must.Eq(t, tc.expSoft, soft)
+			must.Eq(t, tc.expHard, hard)
+		})
+	}
+}

e2e/rawexec/input/oversubmax.hcl

@@ -0,0 +1,38 @@
+# Copyright (c) HashiCorp, Inc.
+# SPDX-License-Identifier: BUSL-1.1
+
+job "oversubmax" {
+  type = "batch"
+
+  constraint {
+    attribute = "${attr.kernel.name}"
+    value     = "linux"
+  }
+
+  group "group" {
+    reschedule {
+      attempts  = 0
+      unlimited = false
+    }
+
+    restart {
+      attempts = 0
+      mode     = "fail"
+    }
+
+    task "cat" {
+      driver = "raw_exec"
+
+      config {
+        command = "bash"
+        args    = ["-c", "cat /sys/fs/cgroup/$(cat /proc/self/cgroup | cut -d':' -f3)/memory.{low,max}"]
+      }
+
+      resources {
+        cpu        = 100
+        memory     = 64
+        memory_max = -1 # unlimited
+      }
+    }
+  }
+}

e2e/rawexec/rawexec_test.go

@@ -4,6 +4,7 @@
 package rawexec
 
 import (
+	"regexp"
 	"testing"
 
 	"github.com/hashicorp/nomad/e2e/v3/cluster3"
@@ -18,6 +19,8 @@ func TestRawExec(t *testing.T) {
 	)
 
 	t.Run("testOomAdj", testOomAdj)
+	t.Run("testOversubMemory", testOversubMemory)
+	t.Run("testOversubMemoryUnlimited", testOversubMemoryUnlimited)
 }
 
 func testOomAdj(t *testing.T) {
@@ -27,3 +30,21 @@ func testOomAdj(t *testing.T) {
 	logs := job.TaskLogs("group", "cat")
 	must.StrContains(t, logs.Stdout, "0")
 }
+
+func testOversubMemory(t *testing.T) {
+	job, cleanup := jobs3.Submit(t, "./input/oversub.hcl")
+	t.Cleanup(cleanup)
+
+	logs := job.TaskLogs("group", "cat")
+	must.StrContains(t, logs.Stdout, "134217728") // 128 mb memory_max
+}
+
+func testOversubMemoryUnlimited(t *testing.T) {
+	job, cleanup := jobs3.Submit(t, "./input/oversubmax.hcl")
+	t.Cleanup(cleanup)
+
+	// will print memory.low then memory.max
+	logs := job.TaskLogs("group", "cat")
+	logsRe := regexp.MustCompile(`67108864\s+max`)
+	must.RegexMatch(t, logsRe, logs.Stdout)
+}

nomad/structs/structs.go

@@ -2466,6 +2466,12 @@ func (r *Resources) DiskInBytes() int64 {
 	return int64(r.DiskMB * BytesInMegabyte)
 }
 
+const (
+	// memoryNoLimit is a sentinel value indicating there is no upper hard
+	// memory limit
+	memoryNoLimit = -1
+)
+
 func (r *Resources) Validate() error {
 	var mErr multierror.Error
 
@@ -2488,7 +2494,9 @@ func (r *Resources) Validate() error {
 		}
 	}
 
-	if r.MemoryMaxMB != 0 && r.MemoryMaxMB < r.MemoryMB {
+	// ensure memory_max is greater than memory, unless it is set to 0 or -1 which
+	// are both sentinel values
+	if (r.MemoryMaxMB != 0 && r.MemoryMaxMB != memoryNoLimit) && r.MemoryMaxMB < r.MemoryMB {
 		mErr.Errors = append(mErr.Errors, fmt.Errorf("MemoryMaxMB value (%d) should be larger than MemoryMB value (%d)", r.MemoryMaxMB, r.MemoryMB))
 	}
 

nomad/structs/structs_test.go

@@ -2139,6 +2139,14 @@ func TestTask_Validate_Resources(t *testing.T) {
 			},
 			err: "MemoryMaxMB value (10) should be larger than MemoryMB value (200",
 		},
+		{
+			name: "memory max no limit",
+			res: &Resources{
+				CPU:         100,
+				MemoryMB:    200,
+				MemoryMaxMB: -1,
+			},
+		},
 	}
 
 	for i := range cases {

website/content/docs/drivers/raw_exec.mdx

@@ -141,6 +141,18 @@ properly. Nomad will not leak any processes if cgroups are being used to
 manage the process tree. Cgroups are used on Linux when Nomad is being run with
 appropriate privileges, and the cgroup system is mounted.
 
+If the cluster is configured with memory oversubscription enabled, a task using
+the `raw_exec` driver can be configured to have no maximum memory limit by 
+setting `memory_max = -1`.
+
+```hcl
+resources {
+  cpu        = 500
+  memory     = 128
+  memory_max = -1 # no limit
+}
+```
+
 
 [hardening]: /nomad/docs/install/production/requirements#user-permissions
 [plugin-options]: #plugin-options