Merge pull request #104094 from flokli/systemd-unified-cgroup-hierarchy

systemd: switch to unified cgroup hierarchy by default
This commit is contained in:
Florian Klink 2020-11-22 22:35:42 +01:00 committed by GitHub
commit c76891314d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 78 additions and 23 deletions

View File

@ -277,6 +277,19 @@
<literal>unbound-control</literal> without passing a custom configuration location.
</para>
</listitem>
<listitem>
<para>
NixOS now defaults to the unified cgroup hierarchy (cgroupsv2).
See the <link xlink:href="https://www.redhat.com/sysadmin/fedora-31-control-group-v2">Fedora Article for 31</link>
for details on why this is desirable, and how it impacts containers.
</para>
<para>
If you want to run containers with a runtime that does not yet support cgroupsv2,
you can switch back to the old behaviour by setting
<xref linkend="opt-systemd.enableUnifiedCgroupHierarchy"/> = <literal>false</literal>;
and rebooting.
</para>
</listitem>
</itemizedlist>
</section>
</section>

View File

@ -23,5 +23,9 @@ with lib;
boot.specialFileSystems."/proc".options = [ "hidepid=2" "gid=${toString config.ids.gids.proc}" ];
systemd.services.systemd-logind.serviceConfig.SupplementaryGroups = [ "proc" ];
# Disable cgroupsv2, which doesn't work with hidepid.
# https://github.com/NixOS/nixpkgs/pull/104094#issuecomment-729996203
systemd.enableUnifiedCgroupHierarchy = false;
};
}

View File

@ -76,6 +76,10 @@ in
enable = mkDefault true;
};
# TODO: disable this once k3s supports cgroupsv2, either by docker
# supporting it, or their bundled containerd
systemd.enableUnifiedCgroupHierarchy = false;
systemd.services.k3s = {
description = "k3s service";
after = mkIf cfg.docker [ "docker.service" ];

View File

@ -550,6 +550,14 @@ in
'';
};
systemd.enableUnifiedCgroupHierarchy = mkOption {
default = true;
type = types.bool;
description = ''
Whether to enable the unified cgroup hierarchy (cgroupsv2).
'';
};
systemd.coredump.enable = mkOption {
default = true;
type = types.bool;
@ -1178,6 +1186,7 @@ in
boot.kernel.sysctl = mkIf (!cfg.coredump.enable) {
"kernel.core_pattern" = "core";
};
boot.kernelParams = optional (!cfg.enableUnifiedCgroupHierarchy) "systemd.unified_cgroup_hierarchy=0";
};
# FIXME: Remove these eventually.

View File

@ -155,6 +155,9 @@ in
users.groups.docker.gid = config.ids.gids.docker;
systemd.packages = [ cfg.package ];
# TODO: remove once docker 20.10 is released
systemd.enableUnifiedCgroupHierarchy = false;
systemd.services.docker = {
wantedBy = optional cfg.enableOnBoot "multi-user.target";
environment = proxy_env;

View File

@ -34,7 +34,6 @@ import ./make-test-python.nix (
podman.wait_for_unit("sockets.target")
start_all()
with subtest("Run container as root with runc"):
podman.succeed("tar cv --files-from /dev/null | podman import - scratchimg")
podman.succeed(
@ -53,16 +52,14 @@ import ./make-test-python.nix (
podman.succeed("podman stop sleeping")
podman.succeed("podman rm sleeping")
with subtest("Run container rootless with runc"):
podman.succeed(su_cmd("tar cv --files-from /dev/null | podman import - scratchimg"))
with subtest("Run container as root with the default backend"):
podman.succeed("tar cv --files-from /dev/null | podman import - scratchimg")
podman.succeed(
su_cmd(
"podman run --runtime=runc -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
)
"podman run -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
)
podman.succeed(su_cmd("podman ps | grep sleeping"))
podman.succeed(su_cmd("podman stop sleeping"))
podman.succeed(su_cmd("podman rm sleeping"))
podman.succeed("podman ps | grep sleeping")
podman.succeed("podman stop sleeping")
podman.succeed("podman rm sleeping")
with subtest("Run container rootless with crun"):
podman.succeed(su_cmd("tar cv --files-from /dev/null | podman import - scratchimg"))
@ -74,6 +71,18 @@ import ./make-test-python.nix (
podman.succeed(su_cmd("podman ps | grep sleeping"))
podman.succeed(su_cmd("podman stop sleeping"))
podman.succeed(su_cmd("podman rm sleeping"))
# As of 2020-11-20, the runc backend doesn't work with cgroupsv2 yet, so we don't run that test.
with subtest("Run container rootless with the default backend"):
podman.succeed(su_cmd("tar cv --files-from /dev/null | podman import - scratchimg"))
podman.succeed(
su_cmd(
"podman run -d --name=sleeping -v /nix/store:/nix/store -v /run/current-system/sw/bin:/bin scratchimg /bin/sleep 10"
)
)
podman.succeed(su_cmd("podman ps | grep sleeping"))
podman.succeed(su_cmd("podman stop sleeping"))
podman.succeed(su_cmd("podman rm sleeping"))
'';
}
)

View File

@ -82,6 +82,10 @@ import ./make-test-python.nix ({ pkgs, ... }: {
"systemd-run --pty --property=Type=oneshot --property=DynamicUser=yes --property=User=iamatest whoami"
)
with subtest("regression test for https://bugs.freedesktop.org/show_bug.cgi?id=77507"):
retcode, output = machine.execute("systemctl status testservice1.service")
assert retcode in [0, 3] # https://bugs.freedesktop.org/show_bug.cgi?id=77507
# Regression test for https://github.com/NixOS/nixpkgs/issues/35268
with subtest("file system with x-initrd.mount is not unmounted"):
machine.succeed("mountpoint -q /test-x-initrd-mount")
@ -122,17 +126,6 @@ import ./make-test-python.nix ({ pkgs, ... }: {
machine.wait_for_unit("multi-user.target")
assert "fq_codel" in machine.succeed("sysctl net.core.default_qdisc")
# Test cgroup accounting is enabled
with subtest("systemd cgroup accounting is enabled"):
machine.wait_for_unit("multi-user.target")
assert "yes" in machine.succeed(
"systemctl show testservice1.service -p IOAccounting"
)
retcode, output = machine.execute("systemctl status testservice1.service")
assert retcode in [0, 3] # https://bugs.freedesktop.org/show_bug.cgi?id=77507
assert "CPU:" in output
# Test systemd is configured to manage a watchdog
with subtest("systemd manages hardware watchdog"):
machine.wait_for_unit("multi-user.target")
@ -168,5 +161,25 @@ import ./make-test-python.nix ({ pkgs, ... }: {
machine.succeed("systemctl status systemd-cryptsetup@luks1.service")
machine.succeed("mkdir -p /tmp/luks1")
machine.succeed("mount /dev/mapper/luks1 /tmp/luks1")
# Do some IP traffic
output_ping = machine.succeed(
"systemd-run --wait -- /run/wrappers/bin/ping -c 1 127.0.0.1 2>&1"
)
with subtest("systemd reports accounting data on system.slice"):
output = machine.succeed("systemctl status system.slice")
assert "CPU:" in output
assert "Memory:" in output
assert "IP:" in output
assert "0B in, 0B out" not in output
assert "IO:" in output
assert "0B read, 0B written" not in output
with subtest("systemd per-unit accounting works"):
assert "IP traffic received: 84B" in output_ping
assert "IP traffic sent: 84B" in output_ping
'';
})

View File

@ -281,9 +281,9 @@ stdenv.mkDerivation {
"-Dmount-path=${utillinux}/bin/mount"
"-Dumount-path=${utillinux}/bin/umount"
"-Dcreate-log-dirs=false"
# Upstream uses cgroupsv2 by default. To support docker and other
# container managers we still need v1.
"-Ddefault-hierarchy=hybrid"
# Use cgroupsv2. This is already the upstream default, but better be explicit.
"-Ddefault-hierarchy=unified"
# Upstream defaulted to disable manpages since they optimize for the much
# more frequent development builds
"-Dman=true"