open-nomad/drivers/exec/driver_test.go

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

878 lines
24 KiB
Go
Raw Normal View History

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: MPL-2.0
2018-10-16 02:37:58 +00:00
package exec
import (
"bytes"
"context"
"errors"
2018-10-16 02:37:58 +00:00
"fmt"
"os"
"path/filepath"
"runtime"
2019-11-15 14:31:34 +00:00
"strconv"
2018-10-16 02:37:58 +00:00
"strings"
"sync"
2019-11-15 14:31:34 +00:00
"syscall"
2018-10-16 02:37:58 +00:00
"testing"
"time"
"github.com/hashicorp/nomad/ci"
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
"github.com/hashicorp/nomad/client/lib/cgutil"
ctestutils "github.com/hashicorp/nomad/client/testutil"
"github.com/hashicorp/nomad/drivers/shared/executor"
"github.com/hashicorp/nomad/helper/pluginutils/hclutils"
2018-10-16 02:37:58 +00:00
"github.com/hashicorp/nomad/helper/testlog"
"github.com/hashicorp/nomad/helper/testtask"
"github.com/hashicorp/nomad/helper/uuid"
"github.com/hashicorp/nomad/nomad/structs"
basePlug "github.com/hashicorp/nomad/plugins/base"
2018-10-16 02:37:58 +00:00
"github.com/hashicorp/nomad/plugins/drivers"
dtestutil "github.com/hashicorp/nomad/plugins/drivers/testutils"
2018-10-16 02:37:58 +00:00
"github.com/hashicorp/nomad/testutil"
"github.com/stretchr/testify/require"
)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
const (
cgroupParent = "testing.slice"
)
2018-10-16 02:37:58 +00:00
func TestMain(m *testing.M) {
if !testtask.Run() {
os.Exit(m.Run())
}
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
func testResources(allocID, task string) *drivers.Resources {
if allocID == "" || task == "" {
panic("must be set")
}
r := &drivers.Resources{
NomadResources: &structs.AllocatedTaskResources{
Memory: structs.AllocatedMemoryResources{
MemoryMB: 128,
},
Cpu: structs.AllocatedCpuResources{
CpuShares: 100,
},
2018-12-14 00:21:41 +00:00
},
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
LinuxResources: &drivers.LinuxResources{
MemoryLimitBytes: 134217728,
CPUShares: 100,
2018-12-14 00:21:41 +00:00
},
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
}
if cgutil.UseV2 {
r.LinuxResources.CpusetCgroupPath = filepath.Join(cgutil.CgroupRoot, cgroupParent, cgutil.CgroupScope(allocID, task))
}
return r
}
func TestExecDriver_Fingerprint_NonLinux(t *testing.T) {
ci.Parallel(t)
require := require.New(t)
if runtime.GOOS == "linux" {
t.Skip("Test only available not on Linux")
}
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
fingerCh, err := harness.Fingerprint(context.Background())
require.NoError(err)
select {
case finger := <-fingerCh:
require.Equal(drivers.HealthStateUndetected, finger.Health)
case <-time.After(time.Duration(testutil.TestMultiplier()*5) * time.Second):
require.Fail("timeout receiving fingerprint")
}
}
2018-10-16 02:37:58 +00:00
func TestExecDriver_Fingerprint(t *testing.T) {
ci.Parallel(t)
2018-10-16 02:37:58 +00:00
require := require.New(t)
ctestutils.ExecCompatible(t)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
2018-10-16 02:37:58 +00:00
fingerCh, err := harness.Fingerprint(context.Background())
require.NoError(err)
select {
case finger := <-fingerCh:
require.Equal(drivers.HealthStateHealthy, finger.Health)
require.True(finger.Attributes["driver.exec"].GetBool())
2018-10-16 02:37:58 +00:00
case <-time.After(time.Duration(testutil.TestMultiplier()*5) * time.Second):
require.Fail("timeout receiving fingerprint")
}
}
func TestExecDriver_StartWait(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
logger := testlog.HCLogger(t)
d := NewExecDriver(ctx, logger)
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "test",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
2018-10-16 02:37:58 +00:00
}
tc := &TaskConfig{
Command: "cat",
Args: []string{"/proc/self/cgroup"},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2018-10-16 02:37:58 +00:00
ch, err := harness.WaitTask(context.Background(), handle.Config.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2018-10-16 02:37:58 +00:00
result := <-ch
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Zero(t, result.ExitCode)
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
func TestExecDriver_StartWaitStopKill(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
2018-12-08 14:58:23 +00:00
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "test",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
}
tc := &TaskConfig{
Command: "/bin/bash",
Args: []string{"-c", "echo hi; sleep 600"},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
defer harness.DestroyTask(task.ID, true)
ch, err := harness.WaitTask(context.Background(), handle.Config.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.WaitUntilStarted(task.ID, 1*time.Second))
2018-10-16 02:37:58 +00:00
go func() {
harness.StopTask(task.ID, 2*time.Second, "SIGINT")
2018-10-16 02:37:58 +00:00
}()
select {
case result := <-ch:
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.False(t, result.Successful())
case <-time.After(10 * time.Second):
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Fail(t, "timeout waiting for task to shutdown")
2018-10-16 02:37:58 +00:00
}
// Ensure that the task is marked as dead, but account
// for WaitTask() closing channel before internal state is updated
testutil.WaitForResult(func() (bool, error) {
status, err := harness.InspectTask(task.ID)
if err != nil {
return false, fmt.Errorf("inspecting task failed: %v", err)
}
if status.State != drivers.TaskStateExited {
return false, fmt.Errorf("task hasn't exited yet; status: %v", status.State)
}
return true, nil
}, func(err error) {
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
})
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
func TestExecDriver_StartWaitRecover(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
dCtx, dCancel := context.WithCancel(context.Background())
defer dCancel()
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
d := NewExecDriver(dCtx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "test",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
2018-10-16 02:37:58 +00:00
}
tc := &TaskConfig{
Command: "/bin/sleep",
Args: []string{"5"},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2018-10-16 02:37:58 +00:00
ctx, cancel := context.WithCancel(context.Background())
ch, err := harness.WaitTask(ctx, handle.Config.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2018-10-16 02:37:58 +00:00
var wg sync.WaitGroup
wg.Add(1)
go func() {
defer wg.Done()
result := <-ch
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Error(t, result.Err)
2018-10-16 02:37:58 +00:00
}()
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.WaitUntilStarted(task.ID, 1*time.Second))
2018-10-16 02:37:58 +00:00
cancel()
waitCh := make(chan struct{})
go func() {
defer close(waitCh)
wg.Wait()
}()
select {
case <-waitCh:
status, err := harness.InspectTask(task.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Equal(t, drivers.TaskStateRunning, status.State)
2018-10-16 02:37:58 +00:00
case <-time.After(1 * time.Second):
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Fail(t, "timeout waiting for task wait to cancel")
2018-10-16 02:37:58 +00:00
}
// Loose task
d.(*Driver).tasks.Delete(task.ID)
2018-10-16 02:37:58 +00:00
_, err = harness.InspectTask(task.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Error(t, err)
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.RecoverTask(handle))
2018-10-16 02:37:58 +00:00
status, err := harness.InspectTask(task.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Equal(t, drivers.TaskStateRunning, status.State)
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.StopTask(task.ID, 0, ""))
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
// TestExecDriver_NoOrphans asserts that when the main
// task dies, the orphans in the PID namespaces are killed by the kernel
func TestExecDriver_NoOrphans(t *testing.T) {
ci.Parallel(t)
2019-11-15 14:31:34 +00:00
ctestutils.ExecCompatible(t)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
2019-11-15 14:31:34 +00:00
harness := dtestutil.NewDriverHarness(t, d)
defer harness.Kill()
config := &Config{
NoPivotRoot: false,
2021-02-08 16:58:44 +00:00
DefaultModePID: executor.IsolationModePrivate,
DefaultModeIPC: executor.IsolationModePrivate,
}
var data []byte
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, basePlug.MsgPackEncode(&data, config))
baseConfig := &basePlug.Config{PluginConfig: data}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.SetConfig(baseConfig))
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2019-11-15 14:31:34 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "test",
}
if cgutil.UseV2 {
task.Resources = testResources(allocID, "test")
2019-11-15 14:31:34 +00:00
}
cleanup := harness.MkAllocDir(task, true)
defer cleanup()
taskConfig := map[string]interface{}{}
taskConfig["command"] = "/bin/sh"
// print the child PID in the task PID namespace, then sleep for 5 seconds to give us a chance to examine processes
taskConfig["args"] = []string{"-c", fmt.Sprintf(`sleep 3600 & sleep 20`)}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&taskConfig))
2019-11-15 14:31:34 +00:00
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2019-11-15 14:31:34 +00:00
defer harness.DestroyTask(task.ID, true)
waitCh, err := harness.WaitTask(context.Background(), handle.Config.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.WaitUntilStarted(task.ID, 1*time.Second))
var childPids []int
taskState := TaskState{}
testutil.WaitForResult(func() (bool, error) {
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, handle.GetDriverState(&taskState))
if taskState.Pid == 0 {
return false, fmt.Errorf("task PID is zero")
}
children, err := os.ReadFile(fmt.Sprintf("/proc/%d/task/%d/children", taskState.Pid, taskState.Pid))
if err != nil {
return false, fmt.Errorf("error reading /proc for children: %v", err)
}
pids := strings.Fields(string(children))
if len(pids) < 2 {
return false, fmt.Errorf("error waiting for two children, currently %d", len(pids))
}
for _, cpid := range pids {
p, err := strconv.Atoi(cpid)
if err != nil {
return false, fmt.Errorf("error parsing child pids from /proc: %s", cpid)
}
childPids = append(childPids, p)
}
return true, nil
}, func(err error) {
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
})
2019-11-15 14:31:34 +00:00
select {
case result := <-waitCh:
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.True(t, result.Successful(), "command failed: %#v", result)
case <-time.After(30 * time.Second):
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Fail(t, "timeout waiting for task to shutdown")
2019-11-15 14:31:34 +00:00
}
// isProcessRunning returns an error if process is not running
isProcessRunning := func(pid int) error {
process, err := os.FindProcess(pid)
if err != nil {
return fmt.Errorf("failed to find process: %s", err)
}
err = process.Signal(syscall.Signal(0))
if err != nil {
return fmt.Errorf("failed to signal process: %s", err)
}
return nil
}
// task should be dead
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Error(t, isProcessRunning(taskState.Pid))
2019-11-15 14:31:34 +00:00
// all children should eventually be killed by OS
2019-11-15 14:31:34 +00:00
testutil.WaitForResult(func() (bool, error) {
for _, cpid := range childPids {
err := isProcessRunning(cpid)
if err == nil {
return false, fmt.Errorf("child process %d is still running", cpid)
}
if !strings.Contains(err.Error(), "failed to signal process") {
return false, fmt.Errorf("unexpected error: %v", err)
}
2019-11-15 14:31:34 +00:00
}
return true, nil
}, func(err error) {
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2019-11-15 14:31:34 +00:00
})
}
2018-10-16 02:37:58 +00:00
func TestExecDriver_Stats(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
dctx, dcancel := context.WithCancel(context.Background())
defer dcancel()
d := NewExecDriver(dctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "test",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
2018-10-16 02:37:58 +00:00
}
tc := &TaskConfig{
Command: "/bin/sleep",
Args: []string{"5"},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.NotNil(t, handle)
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.WaitUntilStarted(task.ID, 1*time.Second))
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
statsCh, err := harness.TaskStats(ctx, task.ID, time.Second*10)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
select {
case stats := <-statsCh:
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NotEmpty(t, stats.ResourceUsage.MemoryStats.Measured)
require.NotZero(t, stats.Timestamp)
require.WithinDuration(t, time.Now(), time.Unix(0, stats.Timestamp), time.Second)
case <-time.After(time.Second):
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Fail(t, "timeout receiving from channel")
}
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
func TestExecDriver_Start_Wait_AllocDir(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "sleep",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
2018-10-16 02:37:58 +00:00
}
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
exp := []byte{'w', 'i', 'n'}
file := "output.txt"
tc := &TaskConfig{
Command: "/bin/bash",
Args: []string{
"-c",
fmt.Sprintf(`sleep 1; echo -n %s > /alloc/%s`, string(exp), file),
},
2018-10-16 02:37:58 +00:00
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.NotNil(t, handle)
2018-10-16 02:37:58 +00:00
// Task should terminate quickly
waitCh, err := harness.WaitTask(context.Background(), task.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
2018-10-16 02:37:58 +00:00
select {
case res := <-waitCh:
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.True(t, res.Successful(), "task should have exited successfully: %v", res)
2018-10-16 02:37:58 +00:00
case <-time.After(time.Duration(testutil.TestMultiplier()*5) * time.Second):
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Fail(t, "timeout waiting for task")
2018-10-16 02:37:58 +00:00
}
// Check that data was written to the shared alloc directory.
outputFile := filepath.Join(task.TaskDir().SharedAllocDir, file)
act, err := os.ReadFile(outputFile)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Exactly(t, exp, act)
2018-10-16 02:37:58 +00:00
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
func TestExecDriver_User(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "sleep",
User: "alice",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "sleep"),
2018-10-16 02:37:58 +00:00
}
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
tc := &TaskConfig{
Command: "/bin/sleep",
Args: []string{"100"},
2018-10-16 02:37:58 +00:00
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Error(t, err)
require.Nil(t, handle)
2018-10-16 02:37:58 +00:00
msg := "user alice"
if !strings.Contains(err.Error(), msg) {
t.Fatalf("Expecting '%v' in '%v'", msg, err)
}
}
// TestExecDriver_HandlerExec ensures the exec driver's handle properly
// executes commands inside the container.
func TestExecDriver_HandlerExec(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
2018-10-16 02:37:58 +00:00
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
2018-10-16 02:37:58 +00:00
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "sleep",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "sleep"),
2018-10-16 02:37:58 +00:00
}
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
tc := &TaskConfig{
Command: "/bin/sleep",
Args: []string{"9000"},
2018-10-16 02:37:58 +00:00
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
2018-10-16 02:37:58 +00:00
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.NotNil(t, handle)
2018-10-16 02:37:58 +00:00
// Assert cgroup membership
res, err := harness.ExecTask(task.ID, []string{"/bin/cat", "/proc/self/cgroup"}, time.Second)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.True(t, res.ExitResult.Successful())
stdout := strings.TrimSpace(string(res.Stdout))
if !cgutil.UseV2 {
for _, line := range strings.Split(stdout, "\n") {
// skip empty lines
if line == "" {
continue
}
// skip rdma & misc subsystems
if strings.Contains(line, ":rdma:") || strings.Contains(line, ":misc:") || strings.Contains(line, "::") {
continue
}
// assert we are in a nomad cgroup
if !strings.Contains(line, ":/nomad/") {
t.Fatalf("not a member of the allocs nomad cgroup: %q", line)
}
2018-10-16 02:37:58 +00:00
}
} else {
require.True(t, strings.HasSuffix(stdout, ".scope"), "actual stdout %q", stdout)
2018-10-16 02:37:58 +00:00
}
// Exec a command that should fail
res, err = harness.ExecTask(task.ID, []string{"/usr/bin/stat", "lkjhdsaflkjshowaisxmcvnlia"}, time.Second)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.False(t, res.ExitResult.Successful())
2018-10-16 02:37:58 +00:00
if expected := "No such file or directory"; !bytes.Contains(res.Stdout, []byte(expected)) {
t.Fatalf("expected output to contain %q but found: %q", expected, res.Stdout)
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.DestroyTask(task.ID, true))
2018-10-16 02:37:58 +00:00
}
func TestExecDriver_DevicesAndMounts(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
tmpDir := t.TempDir()
err := os.WriteFile(filepath.Join(tmpDir, "testfile"), []byte("from-host"), 600)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
task := &drivers.TaskConfig{
ID: uuid.Generate(),
Name: "test",
User: "root", // need permission to read mounts paths
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "test"),
StdoutPath: filepath.Join(tmpDir, "task-stdout"),
StderrPath: filepath.Join(tmpDir, "task-stderr"),
Devices: []*drivers.DeviceConfig{
{
TaskPath: "/dev/inserted-random",
HostPath: "/dev/random",
Permissions: "rw",
},
},
Mounts: []*drivers.MountConfig{
{
TaskPath: "/tmp/task-path-rw",
HostPath: tmpDir,
Readonly: false,
},
{
TaskPath: "/tmp/task-path-ro",
HostPath: tmpDir,
Readonly: true,
},
},
}
require.NoError(t, os.WriteFile(task.StdoutPath, []byte{}, 660))
require.NoError(t, os.WriteFile(task.StderrPath, []byte{}, 660))
tc := &TaskConfig{
Command: "/bin/bash",
Args: []string{"-c", `
export LANG=en.UTF-8
echo "mounted device /inserted-random: $(stat -c '%t:%T' /dev/inserted-random)"
echo "reading from ro path: $(cat /tmp/task-path-ro/testfile)"
echo "reading from rw path: $(cat /tmp/task-path-rw/testfile)"
touch /tmp/task-path-rw/testfile && echo 'overwriting file in rw succeeded'
touch /tmp/task-path-rw/testfile-from-rw && echo from-exec > /tmp/task-path-rw/testfile-from-rw && echo 'writing new file in rw succeeded'
touch /tmp/task-path-ro/testfile && echo 'overwriting file in ro succeeded'
touch /tmp/task-path-ro/testfile-from-ro && echo from-exec > /tmp/task-path-ro/testfile-from-ro && echo 'writing new file in ro succeeded'
exit 0
`},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
ch, err := harness.WaitTask(context.Background(), handle.Config.ID)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
result := <-ch
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.DestroyTask(task.ID, true))
stdout, err := os.ReadFile(task.StdoutPath)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Equal(t, `mounted device /inserted-random: 1:8
reading from ro path: from-host
reading from rw path: from-host
overwriting file in rw succeeded
writing new file in rw succeeded`, strings.TrimSpace(string(stdout)))
stderr, err := os.ReadFile(task.StderrPath)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Equal(t, `touch: cannot touch '/tmp/task-path-ro/testfile': Read-only file system
touch: cannot touch '/tmp/task-path-ro/testfile-from-ro': Read-only file system`, strings.TrimSpace(string(stderr)))
// testing exit code last so we can inspect output first
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.Zero(t, result.ExitCode)
fromRWContent, err := os.ReadFile(filepath.Join(tmpDir, "testfile-from-rw"))
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.Equal(t, "from-exec", strings.TrimSpace(string(fromRWContent)))
}
func TestConfig_ParseAllHCL(t *testing.T) {
ci.Parallel(t)
cfgStr := `
config {
command = "/bin/bash"
args = ["-c", "echo hello"]
}`
expected := &TaskConfig{
Command: "/bin/bash",
Args: []string{"-c", "echo hello"},
}
var tc *TaskConfig
hclutils.NewConfigParser(taskConfigSpec).ParseHCL(t, cfgStr, &tc)
require.EqualValues(t, expected, tc)
}
func TestExecDriver_NoPivotRoot(t *testing.T) {
ci.Parallel(t)
ctestutils.ExecCompatible(t)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
d := NewExecDriver(ctx, testlog.HCLogger(t))
harness := dtestutil.NewDriverHarness(t, d)
config := &Config{
NoPivotRoot: true,
2021-02-08 16:58:44 +00:00
DefaultModePID: executor.IsolationModePrivate,
DefaultModeIPC: executor.IsolationModePrivate,
}
var data []byte
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, basePlug.MsgPackEncode(&data, config))
bconfig := &basePlug.Config{PluginConfig: data}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, harness.SetConfig(bconfig))
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
allocID := uuid.Generate()
task := &drivers.TaskConfig{
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
AllocID: allocID,
ID: uuid.Generate(),
Name: "sleep",
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
Resources: testResources(allocID, "sleep"),
}
cleanup := harness.MkAllocDir(task, false)
defer cleanup()
tc := &TaskConfig{
Command: "/bin/sleep",
Args: []string{"100"},
}
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, task.EncodeConcreteDriverConfig(&tc))
handle, _, err := harness.StartTask(task)
client: enable support for cgroups v2 This PR introduces support for using Nomad on systems with cgroups v2 [1] enabled as the cgroups controller mounted on /sys/fs/cgroups. Newer Linux distros like Ubuntu 21.10 are shipping with cgroups v2 only, causing problems for Nomad users. Nomad mostly "just works" with cgroups v2 due to the indirection via libcontainer, but not so for managing cpuset cgroups. Before, Nomad has been making use of a feature in v1 where a PID could be a member of more than one cgroup. In v2 this is no longer possible, and so the logic around computing cpuset values must be modified. When Nomad detects v2, it manages cpuset values in-process, rather than making use of cgroup heirarchy inheritence via shared/reserved parents. Nomad will only activate the v2 logic when it detects cgroups2 is mounted at /sys/fs/cgroups. This means on systems running in hybrid mode with cgroups2 mounted at /sys/fs/cgroups/unified (as is typical) Nomad will continue to use the v1 logic, and should operate as before. Systems that do not support cgroups v2 are also not affected. When v2 is activated, Nomad will create a parent called nomad.slice (unless otherwise configured in Client conifg), and create cgroups for tasks using naming convention <allocID>-<task>.scope. These follow the naming convention set by systemd and also used by Docker when cgroups v2 is detected. Client nodes now export a new fingerprint attribute, unique.cgroups.version which will be set to 'v1' or 'v2' to indicate the cgroups regime in use by Nomad. The new cpuset management strategy fixes #11705, where docker tasks that spawned processes on startup would "leak". In cgroups v2, the PIDs are started in the cgroup they will always live in, and thus the cause of the leak is eliminated. [1] https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html Closes #11289 Fixes #11705 #11773 #11933
2022-02-28 22:24:01 +00:00
require.NoError(t, err)
require.NotNil(t, handle)
require.NoError(t, harness.DestroyTask(task.ID, true))
}
func TestDriver_Config_validate(t *testing.T) {
ci.Parallel(t)
t.Run("pid/ipc", func(t *testing.T) {
for _, tc := range []struct {
pidMode, ipcMode string
exp error
}{
{pidMode: "host", ipcMode: "host", exp: nil},
{pidMode: "private", ipcMode: "host", exp: nil},
{pidMode: "host", ipcMode: "private", exp: nil},
{pidMode: "private", ipcMode: "private", exp: nil},
{pidMode: "other", ipcMode: "private", exp: errors.New(`default_pid_mode must be "private" or "host", got "other"`)},
{pidMode: "private", ipcMode: "other", exp: errors.New(`default_ipc_mode must be "private" or "host", got "other"`)},
} {
require.Equal(t, tc.exp, (&Config{
DefaultModePID: tc.pidMode,
DefaultModeIPC: tc.ipcMode,
}).validate())
}
})
t.Run("allow_caps", func(t *testing.T) {
for _, tc := range []struct {
ac []string
exp error
}{
{ac: []string{}, exp: nil},
{ac: []string{"all"}, exp: nil},
{ac: []string{"chown", "sys_time"}, exp: nil},
{ac: []string{"CAP_CHOWN", "cap_sys_time"}, exp: nil},
{ac: []string{"chown", "not_valid", "sys_time"}, exp: errors.New("allow_caps configured with capabilities not supported by system: not_valid")},
} {
require.Equal(t, tc.exp, (&Config{
DefaultModePID: "private",
DefaultModeIPC: "private",
AllowCaps: tc.ac,
}).validate())
}
})
}
func TestDriver_TaskConfig_validate(t *testing.T) {
ci.Parallel(t)
t.Run("pid/ipc", func(t *testing.T) {
for _, tc := range []struct {
pidMode, ipcMode string
exp error
}{
{pidMode: "host", ipcMode: "host", exp: nil},
{pidMode: "host", ipcMode: "private", exp: nil},
{pidMode: "host", ipcMode: "", exp: nil},
{pidMode: "host", ipcMode: "other", exp: errors.New(`ipc_mode must be "private" or "host", got "other"`)},
{pidMode: "host", ipcMode: "host", exp: nil},
{pidMode: "private", ipcMode: "host", exp: nil},
{pidMode: "", ipcMode: "host", exp: nil},
{pidMode: "other", ipcMode: "host", exp: errors.New(`pid_mode must be "private" or "host", got "other"`)},
} {
require.Equal(t, tc.exp, (&TaskConfig{
ModePID: tc.pidMode,
ModeIPC: tc.ipcMode,
}).validate())
}
})
t.Run("cap_add", func(t *testing.T) {
for _, tc := range []struct {
adds []string
exp error
}{
{adds: nil, exp: nil},
{adds: []string{"chown"}, exp: nil},
{adds: []string{"CAP_CHOWN"}, exp: nil},
{adds: []string{"chown", "sys_time"}, exp: nil},
{adds: []string{"chown", "not_valid", "sys_time"}, exp: errors.New("cap_add configured with capabilities not supported by system: not_valid")},
} {
require.Equal(t, tc.exp, (&TaskConfig{
CapAdd: tc.adds,
}).validate())
}
})
t.Run("cap_drop", func(t *testing.T) {
for _, tc := range []struct {
drops []string
exp error
}{
{drops: nil, exp: nil},
{drops: []string{"chown"}, exp: nil},
{drops: []string{"CAP_CHOWN"}, exp: nil},
{drops: []string{"chown", "sys_time"}, exp: nil},
{drops: []string{"chown", "not_valid", "sys_time"}, exp: errors.New("cap_drop configured with capabilities not supported by system: not_valid")},
} {
require.Equal(t, tc.exp, (&TaskConfig{
CapDrop: tc.drops,
}).validate())
}
})
}