This bug was fixed in the package systemd - 204-5ubuntu20.28
---
systemd (204-5ubuntu20.28) trusty; urgency=medium
* logind: fix memleaks in session's free path and cgmanager glue code
(LP: #1750013)
-- gpicc...@canonical.com (Guilherme G. Piccoli) Tue, 03 Apr 2018
13:38:08
As mentioned by Dan in the above comments, some failures in autopkgtest
like:
autopkgtest [14:44:48]: test ubuntu-regression-suite: [---
Source Package Version: 4.4.0-1017.17
Running Kernel Version: 3.13.0-145.194
ERROR: running version does not match source package
Are
> caused by bug 1723272
sorry, i meant bug 1723223
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1750013
Title:
systemd-logind: memory leaks on session's connections
Note, the 'linux-*' autopkgtest failures in pending-sru for this pkg are
caused by bug 1723272, a known bug that can be ignored as it's unrelated
to this sru.
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
Hi Łukasz, thanks for the heads-up.
I just tested systemd from -proposed, version 204-5ubuntu20.28 and it
fixes the issue reported in this LP.
The test consists in doing multiple SSHs in a loop (better explained in
the description). I've run the test with and without cgmanager
installed, in
Hello Guilherme, or anyone else affected,
Accepted systemd into trusty-proposed. The package will build now and be
available at
https://launchpad.net/ubuntu/+source/systemd/204-5ubuntu20.28 in a few
hours, and then in the -proposed repository.
Please help us by testing this new package. See
Dan, this is the v3 of the patch. I bumped the version to 20.28 since my
proposed 20.27 caused the regression aforementioned.
For this version, I removed the dependency of cgmanager, along with the
code that added closing sessions to garbage collector. Happens that a
similar code is present on
Thank you Mauro! One thing that worth to take a look is that you're
using kernel 3.13, and this could be related to the high CPU utilization
issue you're observing.
We're changing the approach of this fix to not rely on cgmanager
anymore. The CPU utilization issue is however another bug that
...and for systemd-login
(cgmanager and systemd-login were using both around 49-50% of CPU)
** Attachment added: "trace-syslog.txt"
https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1750013/+attachment/5098335/+files/trace-syslog.txt
--
You received this bug notification because you
my trace for cgmanager
Linux www 3.13.0-144-generic #193-Ubuntu SMP Thu Mar 15 17:02:07 UTC
2018 i686 athlon i686 GNU/Linux
** Attachment added: "trace-cgman.txt"
https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1750013/+attachment/5098334/+files/trace-cgman.txt
--
You received this
I am also experiencing cpu overload for cgmanager and (after disabling it in
grub's default with "cgroup_disable=cpu") systemd-logind as described in #16.
I see nothing strange in the logs, but if I can be of any help by providing any
information please just ask me.
Cheers,
mauf
--
You
Thanks Stéphane, the 2nd point is a reasonable consideration.
Regardless if I could provide a fix to cgmanager for the regression,
certainly it won't be added again as a dependency.
Cheers,
Guilherme
--
You received this bug notification because you are a member of Ubuntu
Touch seeded
Removed this SRU from -proposed.
Two reasons for this:
1) The mentioned regression
2) We should NEVER introduce a new dependency to a package through an SRU.
Doing so will prevent any user that doesn't dist-upgrade (uses
upgrade/safe-upgrade instead) from getting the new dependency,
We had a potential regression reported after some users installed this package
from -proposed.
The reports starts on comment #17 in LP: #1303649 .
Summary: users are reporting high CPU loads from both systemd-logind and
cgmanager processes, as well as delays in logins. I'm investigating to
Thanks Dan and Brian.
Just tested: after 1h and more than 5500 SSH sessions, no leaks at all were
observed.
(As a comparison, testing the "old" version 20.26 I got 1.8 MB of leak in half
an hour!)
Cheers,
Guilherme
** Tags removed: sts-sru-needed verification-needed
Hello Guilherme, or anyone else affected,
Accepted systemd into trusty-proposed. The package will build now and be
available at
https://launchpad.net/ubuntu/+source/systemd/204-5ubuntu20.27 in a few
hours, and then in the -proposed repository.
Please help us by testing this new package. See
Excellent, thanks - that is perfect justification to change it to
Depends:, I'll get the upload going soon.
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1750013
Title:
Thanks Dan.
About cgmanager: if we don't have it installed, it does not affect
memory of systemd-logind tool per se. What happens, IMHO, is even more
severe: we don't free/de-allocate sysfs cgroup paths for sessions. So,
in my tests _without_ cgmanager, after 8000 SSH sessions to my target
That looks good.
My only comment now is, I see you've switched cgmanager from Suggests:
to Depends:...is that *really* needed to fix a memleak here? i.e., if
you uninstall cgmanager and re-run your memleak tests, does mem still
leak? While changing it to a Depends: isn't impossible, I'd prefer
** Patch added: "debdiff v2"
https://bugs.launchpad.net/ubuntu/artful/+source/systemd/+bug/1750013/+attachment/5067376/+files/lp1750013-trusty_v2.debdiff
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
Dan, please let me know if the debdiff v2 above addresses your concern
and if it's properly following your suggestion. I tested the
modification and it's working fine.
Thanks,
Guilherme
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is
Dan, really well-observed! Thanks for the suggestion, it makes total
sense. I'll update the approach: instead of using session_remove_fifo()
to add the session on GC, I'll use the DBus handler itself.
After testing, I'll submit a debdiff_v2 here.
Thanks,
Guilherme
--
You received this bug
> when SSH ends, for example, a Release event is sent
> through dbus and systemd-logind captures it, in the function
> manager_message_handler().
>
> From there, the function session_remove_fifo() is called. That point is our
> "bootstrap"
> to add the closing session on gc
ok i see that path,
Dan, thanks for looking at the patch and for your question!
I tried to explain this a bit in comment #1, not too much detailed though:
"[...]closing session should be somehow added to gc [...] and then (the gc)
performs
the clean-up.
To include a session in gc, the function
This does not look right to me:
@@ -951,6 +952,7 @@
free(s->fifo_path);
s->fifo_path = NULL;
}
+session_add_to_gc_queue(s);
}
int session_check_gc(Session *s, bool drop_not_started) {
that puts the session on the manager's gc queue, however
** Description changed:
- It was observed that systemd-logind tool is leaking memory at each session
- connected. The issue happens in systemd from Trusty (14.04), which latest
- version currently (Feb/2018) is 204-5ubuntu20.26 (and still reproduces
- the bug).
-
- The basic test-case is to run
** Changed in: systemd (Ubuntu Trusty)
Importance: Undecided => Medium
** Changed in: systemd (Ubuntu Trusty)
Status: New => In Progress
** Changed in: systemd (Ubuntu Trusty)
Assignee: (unassigned) => Guilherme G. Piccoli (gpiccoli)
** Changed in: systemd (Ubuntu Xenial)
** Also affects: systemd (Ubuntu Artful)
Importance: Undecided
Status: New
** Also affects: systemd (Ubuntu Bionic)
Importance: Medium
Assignee: Guilherme G. Piccoli (gpiccoli)
Status: In Progress
** Also affects: systemd (Ubuntu Trusty)
Importance: Undecided
** Changed in: systemd (Ubuntu)
Importance: Undecided => Medium
** Changed in: systemd (Ubuntu)
Status: New => In Progress
--
You received this bug notification because you are a member of Ubuntu
Touch seeded packages, which is subscribed to systemd in Ubuntu.
As already briefly mentioned, Trusty systemd's package has
additional patches to customize systemd - specially, we have the addition
of glue code to make Trusty's systemd work with cgmanager. This is
necessary due to the choice of upstart as init manager - it prevents
regular cgroup handling from
It was noticed that during the logind manager logic for creating a session
based on dbus event, the sessions's initialization procedure will allocate
2 structures related to cgroup management of a session: controllers and
reset_controllers.
Both these structs are filled with cgroup controllers
After some debug, it was observed that the only path in which a session
is freed (other than in user/seat free or manager free) is in garbage
collecting, specifically in the function manager_gc().
It means a closing session should be somehow added to gc in order gc has
a chance to validate if the
32 matches
Mail list logo
