Red Hat Bugzilla – Bug 453392
virt-manager.py is taking all the memory!
Last modified: 2009-12-14 16:18:30 EST
Description of problem:
Run virt-manager.py one day or a couple of hours (connect it to a host):
It take more and more memory and may overload the system.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
2.Connect it to a host
3.Lets run a couple of hours
[root@xenhost1 ~]# date
lun jun 30 14:22:19 RET 2008
[root@xenhost1 ~]# ps -auxw --sort=rss | grep virt-manager
Warning: bad syntax, perhaps a bogus '-'? See /usr/share/doc/procps-3.2.7/FAQ
root 10770 0.0 0.1 3952 732 pts/2 S+ 14:22 0:00 grep virt-manager
root 3493 1.1 77.0 668692 557352 ? Ss Jun29 18:36 python
It may crash the host server
Not leak memory
Hi, this is a previously reported issue but we are tracking progress in a
private bug. I think we have a working patch though, so I'll keep you informed.
One of customer in China also have the same problem!
The output of ps command.
root 16256 5.8 75.4 6768480 6042652 ? Ss Jul23 1017:25 python /usr/share/virt-manager/virt-manager.py
any patch or workaround method?
Is the the bug which causes me to see ridiculous %CPU values from procps with fedora rawhide (procps-3.2.7-20.fc9.i386), or is that a different bug?
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
3168 root 20 0 96852 20m 7748 S 424.4 1.1 363:21.90 Xorg
3769 jik 20 0 25976 8888 7532 S 284.5 0.4 37:36.81 multiload-apple
3830 jik 27 7 74204 30m 12m S 141.9 1.6 60:44.16 beagled
2804 haldaemo 20 0 7128 4600 3932 S 41.2 0.2 5:24.25 hald
4839 jik 20 0 2560 1092 828 R 1.3 0.1 0:00.10 top
Oh, Damn, never mind, I put that comment in the wrong ticket.
We see this as well when we leave Virtual Machine Manager open on a machine for an extended time. The machine runs out of memory, and problems start cropping up. Interestingly, it's another virtualization item that then complains via email (due to a cron job failing). Only the subject line of the email and the contents are included here:
Subject: Cron <root@briar04> python /usr/share/rhn/virtualization/poller.py
Traceback (most recent call last):
File "/usr/share/rhn/virtualization/poller.py", line 213, in ?
debug = options and options.debug)
File "/usr/share/rhn/virtualization/poller_state_cache.py", line 50, in __init__
File "/usr/share/rhn/virtualization/poller_state_cache.py", line 123, in _load_state
except PickleError, pe:
NameError: global name 'PickleError' is not defined
I've had to reboot servers that have gotten into this state. Haven't figured out what services to kick to avoid reboot.
Created attachment 315545 [details]
Add support for bonding and vlan devices (also plugs memory leak)
The attached patch fixes the memory leak. It is being tracked by a private bug to add support for bonding and vlan devices for bridges, and also happens to fix the leak :)
This will be in 5.3, but here is the patch in the interim.
FYI, this fix has been committed and built. I'm just going to move this bug to ASSIGNED and leave it open until 5.3 is out, at which point I'll close it. For anyone with the proper access, the private bug we are using to track this is 443604.
Ah sorry, the actual bug is 443680.
Okay, fix is built and pending release for 5.3, so I am closing this bug.
So, am I missing something, or was this left out of RH EL 5.3? I have rhn-virtualization-host-1.0.1-55 installed. It appears that this is the latest version and that it is from January 2008, before this bug was opened.
And, we're seeing the exact same messages as Daniel Senie reported in comment #5, or I wouldn't have come looking for an answer. :-)
So, the patch in the private BZ was committed to RHEL-5.3, so this issue should be fixed. What version of virt-manager do you have installed, exactly?
Cole, do you have anything else to add here?
The traceback in comment #5 looks like an RHN bug, so Dave (Comment #10) should file a bug with them. I can pretty much guarantee that the original bug (memory leak) was fixed in RHEL5.3.
Okay, thanks. I'll try to track down what went wrong and where.