I found the issue. It's permgen and its the watch list query. Memory gets back to
"normal" after the error and it does not necessarly fully bring down the wiki
but it might load it seriously and kill it in some cases
I've disabled te watch list for now. This has to do with the scalability of the watch
list with wikis with many pages (it could be auto watch also)
Ludovic
Envoyé de mon iPhone
Le 20 janv. 2013 à 23:45, Sergiu Dumitriu <sergiu(a)xwiki.com> a écrit :
On 01/20/2013 05:37 PM, Ludovic Dubost wrote:
Hi,
I've modified l10n locally to support custom mapping and updated all the
HQL queries which required it. Quick testing shows that it is working.
I'll need a test server to do some performance testing, as well as migrate
to innoDB and see if everything is working fine. While looking at the l10n
server, there seems to be regular out of memory which I don't understand
where they come from as the server is setup with 1GB. It could come from
the L10NGroovy caches but it might be necessary to do some testing using a
memory analyzer.
Is it heap or PermGen? I noticed that Groovy scripts tend to leak classes.
Once the migration is ready, we'll probably
need to take l10n offline for a
day to perform the migration on the latest data.
During this process I would like to publish the l10n application on
xwiki-contrib. Could somebody create the "application-l10n" repository ?
Done:
https://github.com/xwiki-contrib/application-l10n
--
Sergiu Dumitriu
http://purl.org/net/sergiu
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs