rsyslogd gives 100%CPU

Bug #523468 reported by plun
This bug report is a duplicate of:  Bug #523610: rsyslogd spins CPU on some kernels. Edit Remove
76
This bug affects 15 people
Affects Status Importance Assigned to Milestone
rsyslog (Ubuntu)
Triaged
High
Canonical Foundations Team
Lucid
Triaged
High
Canonical Foundations Team

Bug Description

Binary package hint: rsyslog

Checked with "top" and it was Ok after package removal.

After boot ubuntuone-client gave high CPU.

ProblemType: Bug
Architecture: i386
Date: Wed Feb 17 22:51:54 2010
DistroRelease: Ubuntu 10.04
Package: rsyslog (not installed)
ProcEnviron:
 PATH=(custom, user)
 LANG=C
 SHELL=/bin/bash
SourcePackage: rsyslog
Uname: Linux 2.6.33-020633rc8-generic i686

Revision history for this message
Pavel Rojtberg (rojtberg) wrote :

the problem is your custom kernel, which does not work with the updated rsyslogd yet. see https://bugs.edge.launchpad.net/ubuntu/+source/rsyslog/+bug/517773

Revision history for this message
Robert Moerland (veel-mail) wrote :

Confirming, after an upgrade of rsyslogd from 4.2.0-2ubuntu5.1 to 4.2.0-2ubuntu7, rsyslog is consuming a lot of CPU cycles on my machine. See also the attached image of htop running. CPU use is more than 80%.

Revision history for this message
Pavel Rojtberg (rojtberg) wrote :

I suggest downgrading to 4.2.0-2ubuntu5.1 until the bug is fixed

Changed in rsyslog (Ubuntu):
status: New → Confirmed
Revision history for this message
Robert Moerland (veel-mail) wrote :

BTW, I'm running a 2.6.31-14 kernel, which is supposedly supported. Will downgrade for the moment though.

Revision history for this message
Susan Cragin (susancragin) wrote :

I have Ubuntu Studio and my kernel's going absolutely crazy. Can I uninstall rsyslog.

Revision history for this message
Martin Pitt (pitti) wrote :

Regression from bug 517773, which was not completely implemented.

rsyslog needs to check if it can read from /proc/kmsg as non-root, and not drop privileges if not. In other works, seteuid(), read() -> on fail, seteuid(0) and keep it that way, on success -> setuid() -> permanently drop privs.

tags: added: regression-potential
Changed in rsyslog (Ubuntu):
assignee: nobody → Canonical Foundations Team (canonical-foundations)
importance: Undecided → High
status: Confirmed → Triaged
Revision history for this message
Dave Martin (dave-martin-arm) wrote :

I have observed the problem on the standard armel imx51 kernel.

linux-image-2.6.31-604-imx51 2.6.31.604.5
rsyslogd 4.2.0-2ubuntu7

rsyslogd seems to be spinning forever around select(), but I hit debugger issues and was unable to diagnose it further.

Revision history for this message
Loïc Minier (lool) wrote :

You mention select(), could it be that userspace expects pselect support but the imx51 kernel doesn't have it? For instance, under qemu-arm we were getting a lot of "unsupported syscall" for pselect, and I think this appeared recently. I suspect eglibc got rebuilt against kernel headers with pselect() support (2.6.32). If this is really related to select, it might be interesting to try with a pselect() backport in the imx51 kernel.

Revision history for this message
Dave Martin (dave-martin-arm) wrote :

Actually, select() may be a misdiagnosis---

I may have been looking at the wrong thread. There seem to be two threads spinning, but the thread in select appears to be blocked and not consuming CPU.

The spinning threads both intermittently call pthread_yield(), which may could indicate some kind of livelock situation... but I really haven't gone into this in detail.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.