Bug 2179165

Summary: In cgroupv1 mode, delegated slices are instantly SIGKILLed on shutdown
Product: Red Hat Enterprise Linux 9 Reporter: Colin Walters <walters>
Component: systemdAssignee: systemd-maint
Status: CLOSED ERRATA QA Contact: Frantisek Sumsal <fsumsal>
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: 9.2CC: dtardon, jamacku, jwboyer, kdudka, mpatel, msekleta, plumber, pvlasin, rphillips, sdodson, systemd-maint-list, systemd-maint, wking
Target Milestone: rcKeywords: Triaged
Target Release: 9.2Flags: pm-rhel: mirror+
Hardware: Unspecified   
OS: Unspecified   
Whiteboard:
Fixed In Version: systemd-252-13.el9_2 Doc Type: If docs needed, set a value
Doc Text:
Story Points: ---
Clone Of:
: 2180120 (view as bug list) Environment:
Last Closed: 2023-05-09 08:22:35 UTC Type: Bug
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On:    
Bug Blocks: 2180120    

Description Colin Walters 2023-03-16 20:32:57 UTC
See https://issues.redhat.com/browse/OCPBUGS-10353 for more

OCP uses delegated slices for Kubernetes static pods and daemonsets.  OCP 4.13 will also use cgroupv1 by default.

Upstream commit https://github.com/systemd/systemd/commit/e9db43d5910717a1084924c512bf85e2b8265375
caused the processes in this scenario to be instantly SIGKILLed, which causes disruption.

Reproducing is just:

$ rpm-ostree kargs --append=systemd.unified_cgroup_hierarchy=0 --append="systemd.legacy_systemd_cgroup_controller=1" 
$ systemctl reboot
...
$ systemd-run -p Delegate=yes -p TimeoutStopSec=30 -u test-scope --scope /bin/sh -c 'trap "echo got SIGTERM; sleep 5; exit 0" TERM; sleep infinity'  &
$ systemctl stop test-scope.scope
$ journalctl -b --grep=test-scope
Mar 16 20:08:15 cosa-devsh systemd[1]: test-scope.scope: Deactivated successfully.
Mar 16 20:08:22 cosa-devsh systemd[1]: test-scope.scope: Killing process 1315 (sh) with signal SIGKILL.
Mar 16 20:08:22 cosa-devsh systemd[1]: test-scope.scope: Killing process 1316 (sleep) with signal SIGKILL.
Mar 16 20:08:22 cosa-devsh systemd[1]: test-scope.scope: Deactivated successfully.
$

Comment 1 Colin Walters 2023-03-16 20:38:43 UTC
Tested MR in https://gitlab.com/redhat/centos-stream/rpms/systemd/-/merge_requests/72

Comment 2 David Tardon 2023-03-17 07:14:46 UTC
(In reply to Colin Walters from comment #1)
> Tested MR in
> https://gitlab.com/redhat/centos-stream/rpms/systemd/-/merge_requests/72

Could you create a PR against https://github.com/redhat-plumbers/systemd-rhel9 instead? Also, please add "RHEL-only" and "Resolves: #2179165" to the commit message.

Comment 3 Colin Walters 2023-03-17 11:59:51 UTC
> Could you create a PR against https://github.com/redhat-plumbers/systemd-rhel9 instead? Also, please add "RHEL-only" and "Resolves: #2179165" to the commit message.

I'd hoped to help speed up the process by doing a MR, but at this point given the urgency it's probably best if someone who knows and can own shepherding this patch through takes over to avoid round-trips.  Can one of the systemd team do that please?

We'd like this in RHEL 9.2 if at all possible, so that would also involve e.g. going through the exception? process here etc.

Comment 11 Colin Walters 2023-03-17 14:15:21 UTC
As far as managing risk, I'll note that this code path *only* matters in legacy hierarchy setups.  I suspect this will be a pretty small set of RHEL9 users outside of OCP 4.13.

Comment 12 Michal Sekletar 2023-03-17 15:01:12 UTC
This bug affects layered product so it should be a blocker instead of expection. Dropping exception?.

Comment 15 Colin Walters 2023-03-20 14:44:45 UTC
One question that came up is "Why didn't this appear in RHEL8, it has the same code?".  No, it doesn't have the same code as far as I can tell.  Compare:

rhel8: https://github.com/redhat-plumbers/systemd-rhel8/blob/05a06e34ac8a38c1cff2a08ba071386141e0b78d/src/core/unit.c#L4615
rhel9: https://github.com/redhat-plumbers/systemd-rhel9/blob/a810aef1f95727ec3e044441bf02e0261b2e09ec/src/core/unit.c#L4592

Comment 16 Frantisek Sumsal 2023-03-20 14:49:19 UTC
(In reply to Colin Walters from comment #15)
> One question that came up is "Why didn't this appear in RHEL8, it has the
> same code?".  No, it doesn't have the same code as far as I can tell. 
> Compare:
> 
> rhel8:
> https://github.com/redhat-plumbers/systemd-rhel8/blob/
> 05a06e34ac8a38c1cff2a08ba071386141e0b78d/src/core/unit.c#L4615
> rhel9:
> https://github.com/redhat-plumbers/systemd-rhel9/blob/
> a810aef1f95727ec3e044441bf02e0261b2e09ec/src/core/unit.c#L4592

In RHEL-8 the "offending" patch is reverted by a RHEL-only patch:

[0] https://github.com/redhat-plumbers/systemd-rhel8/commit/9d0046ceca10911361137d6496987cb15ffff132

Comment 26 errata-xmlrpc 2023-05-09 08:22:35 UTC
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (systemd bug fix and enhancement update), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2023:2531