[Users] occasional high loadavg without any noticeable cpu/memory/io load

Kirill Korotaev dev at parallels.com
Tue May 22 08:35:33 EDT 2012


On May 22, 2012, at 16:17 , Rene C. wrote:


On Tue, May 22, 2012 at 6:59 PM, Esmé de Wolf <esme at elements.nl<mailto:esme at elements.nl>> wrote:
I also think that these UBC settings are not consistent. Especially when you have all containers configured with these same UBC settings you will have soon or later problems.

See: http://wiki.openvz.org/UBC_consistency_check and other pages on the WIKI.

Kind Regards,

Esme

I read that UBC page already and used it to set these values.

No, all my containers do not have the same UBC settings, they were set depending on how much resources each container should have.

Please let me know where any of the values in my conf file conflicts with the UBC recommendations.

I do understand that they may need to be fine tuned in each case, but that's basically what this question is about :)

So basically at this time I have two questions I don't understand:

1) how is it possible to have physpages hit the limit when top never shows more than about 75-80% of the memory used?

once again: top shows current (immedeate) values.
maxheld in user_beancounters shows you *maximum* over time.
There is an API for resetting it AFAIR, but no user-space tool in OpenVZ :(((

2) how did dcachesize hit limit when both df -i and df -h shows plenty of resources - and haven't been close to limits?

dcachesize has nothing to do with df.
it's kernel memory used for paths and pinned by opened files and CWD.
You can safely increase it if needed. It's just DoS protection.


Could the values in the beancounter file be old? Is there a way to reset them (without restarting the CT) ?

Best,
Rene

<ATT00001.c>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://openvz.org/pipermail/users/attachments/20120522/5e955ea8/attachment.html


More information about the Users mailing list