[1.21] utils/RunUnderSystemdScope: fix wrt channel deadlock #5914

kolyshkin · 2022-05-27T01:55:52Z

As seen in [1], sometimes coreos/go-systemd/dbus package deadlocks: the
jobCompete is stuck trying to send job result string to the channel
while holding the jobListener lock, while startJob (called by
StartTransientUnit) waits for the same lock.

Alas, it is not clear why the channel is not being read, nor was I able
to reproduce it locally.

Make the job result channel buffered, so jobJistener won't block on
channel send and thus StartTransientUnit won't be stuck either.

While at it,

move the error wrapping out of mgr.RetryOnDisconnect function,
and use fmt.Errorf with %w instead of obsoleted errors.Wrap;
improve error messages, printing the systemd unit name (so we can
check it in systemd log);
do check the job result string -- in case it is not "done",
return an error back to the caller, which should help avoid other
issues down the line.

[1] https://bugzilla.redhat.com/show_bug.cgi?id=2082344

Avoid a deadlock while moving conmon to a different system scope (RHBZ#2082344)

As seen in [1], sometimes coreos/go-systemd/dbus package deadlocks: the jobCompete is stuck trying to send job result string to the channel while holding the jobListener lock, while startJob (called by StartTransientUnit) waits for the same lock. Alas, it is not clear why the channel is not being read, nor was I able to reproduce it locally. Make the job result channel buffered, so jobJistener won't block on channel send and thus StartTransientUnit won't be stuck either. While at it, - move the error wrapping out of mgr.RetryOnDisconnect function, and use fmt.Errorf with %w instead of obsoleted errors.Wrap; - improve error messages, printing the systemd unit name (so we can check it in systemd log); - do check the job result string -- in case it is not "done", return an error back to the caller, which should help avoid other issues down the line. [1] https://bugzilla.redhat.com/show_bug.cgi?id=2082344 Signed-off-by: Kir Kolyshkin <[email protected]>

codecov · 2022-05-27T02:37:38Z

Codecov Report

Merging #5914 (343bcdd) into release-1.21 (8ed37a9) will decrease coverage by 0.00%.
The diff coverage is 28.57%.

@@               Coverage Diff                @@
##           release-1.21    #5914      +/-   ##
================================================
- Coverage         44.97%   44.97%   -0.01%     
================================================
  Files               109      109              
  Lines             11063    11065       +2     
================================================
  Hits               4976     4976              
- Misses             5608     5610       +2     
  Partials            479      479

haircommander · 2022-05-27T13:29:56Z

utils/utils.go

+		if s != "done" {
+			return fmt.Errorf("error moving conmon with pid %d to systemd unit %s: got %s", pid, unitName, s)
+		}
 	case <-time.After(time.Minute * 6):


I find myself wondering if we still need this case. Ideally, the buffered channel would fix the deadlock long term. From the sounds of it, having the request timeout just caused a less obvious deadlock

I believe we do. The buffered channel fixes the deadlock, but if we haven't received a reply from systemd we should still say so and return an error.

kolyshkin · 2022-05-31T17:00:43Z

lint job fails since it uses an old version of golangci-lint but a new version of golang (IOW release-1.21 branch needs some love).

haircommander · 2022-05-31T17:08:45Z

/retest
/approve

LGTM, @cri-o/cri-o-maintainers PTAL

openshift-ci · 2022-05-31T17:09:05Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: haircommander, kolyshkin

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

~~OWNERS~~ [haircommander,kolyshkin]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

kolyshkin · 2022-05-31T18:04:39Z

lint job fails since it uses an old version of golangci-lint but a new version of golang (IOW release-1.21 branch needs some love).

Being fixed in #5918

rphillips · 2022-05-31T18:58:43Z

@kolyshkin I think we should get this into 'main' first.

Can we give the customer a test build to try out?

kolyshkin · 2022-05-31T19:11:33Z

@kolyshkin I think we should get this into 'main' first.

Does it matter if we do the backport or forward port?

Can we give the customer a test build to try out?

Yes, that would be a good thing to do, despite the lack of clear repro.

haircommander · 2022-05-31T20:29:23Z

Does it matter if we do the backport or forward port?

let's start with main and pull back sequentially, so we order when they merge by inverse relative stability (1.21 should be very stable, 1.24 should be stable)

kolyshkin · 2022-05-31T23:39:11Z

Forward-port to main branch: #5922

kolyshkin · 2022-05-31T23:40:12Z

let's start with main and pull back sequentially, so we order when they merge by inverse relative stability (1.21 should be very stable, 1.24 should be stable)

Meaning I can still do a forward-port, but the order of merging is important, right?

Here's one for main: #5922

TomSweeneyRedHat · 2022-06-01T18:47:39Z

LGTM
But lint doesn't look happy

nee1esh · 2022-06-08T01:06:27Z

/retest

github-actions · 2022-07-09T00:02:35Z

A friendly reminder that this PR had no activity for 30 days.

EmmanuelKasper · 2022-07-22T08:00:58Z

/retest

haircommander · 2022-07-22T18:52:02Z

/lgtm

openshift-bot · 2022-07-22T19:03:22Z

/retest-required