An update on the wiki.zope.org/*.zwiki.org server for 2007/3/7:
Uptime troubles and investigations continue, and there have been
significant outages as I haven't always been able to respond
quickly. Thanks for your patience; we are making progress.
- Yesterday, an automated zope restart failed because of an error related
to the (new) permanent zeo cache. I have gone back to temporary zeo
cache.
- Today, zope stopped responding as both threads blocked trying to send
mail. (I was able to see this from the traceback in event.log when I
killed them.) Some sort of deadlock with qmail. This has been happening
before so it's good to pinpoint it. I will switch to MaildropHost asap
which may help.
I have been watching the server processes and tuning various
things. Sascha Welter implemented http conditional-get handling for zwiki,
and also configured httpcaches for several of the (dynamic) front
pages. We have reduced cpu load a lot, and made loading of
previously-visited pages a lot more snappy!
(Note there is an open issue with conditional-get: Firefox and Mozilla
users with default settings will often have to reload after an edit to see
the changes. We have some ideas to prevent this, until then I might have
to revert this feature.)
Edit operations, especially those touching many pages, can still be rather
slow as we are still memory-constrained (~250M quota). The server is
currently running two zope threads with cache-size 3000. When it exceeds
quota (be it through normal growth or memory leak) it is restarted; this
is still happening one or more times per day. Restarting and warming up
the caches is quite slow (can be 5-10 minutes if I'm not mistaken).
I have gone from 7-day to 2-day nightly packs to keep zodb size around the
1G mark (and within the 8G disk quota).
Martijn Faassen is pursuing a possible new, beefier hosting situation.
Thanks also to Sascha Welter, Michael Haubenwallner, Christian Theune and
anyone else I've forgotten, for dropping by #zwiki on irc.freenode.net and
helping to troubleshoot and fix issues.
Cheers!
- Simon
PS I seem to have too many places to post such news:
- zope-web list/gmane.comp.web.zope.websites newsgroup
- http://wiki.zope.org/AboutThisSite
- http://wiki.zope.org/zope3/AboutThisSite
- http://wiki.zope.org/zope2/AboutThisSite
- http://zwiki.org/GeneralDiscussion
I will probably pick just the first 2 or 3 henceforth.
_______________________________________________
Zope-web maillist - Zope-web@zope.org
http://mail.zope.org/mailman/listinfo/zope-web
Uptime troubles and investigations continue, and there have been
significant outages as I haven't always been able to respond
quickly. Thanks for your patience; we are making progress.
- Yesterday, an automated zope restart failed because of an error related
to the (new) permanent zeo cache. I have gone back to temporary zeo
cache.
- Today, zope stopped responding as both threads blocked trying to send
mail. (I was able to see this from the traceback in event.log when I
killed them.) Some sort of deadlock with qmail. This has been happening
before so it's good to pinpoint it. I will switch to MaildropHost asap
which may help.
I have been watching the server processes and tuning various
things. Sascha Welter implemented http conditional-get handling for zwiki,
and also configured httpcaches for several of the (dynamic) front
pages. We have reduced cpu load a lot, and made loading of
previously-visited pages a lot more snappy!
(Note there is an open issue with conditional-get: Firefox and Mozilla
users with default settings will often have to reload after an edit to see
the changes. We have some ideas to prevent this, until then I might have
to revert this feature.)
Edit operations, especially those touching many pages, can still be rather
slow as we are still memory-constrained (~250M quota). The server is
currently running two zope threads with cache-size 3000. When it exceeds
quota (be it through normal growth or memory leak) it is restarted; this
is still happening one or more times per day. Restarting and warming up
the caches is quite slow (can be 5-10 minutes if I'm not mistaken).
I have gone from 7-day to 2-day nightly packs to keep zodb size around the
1G mark (and within the 8G disk quota).
Martijn Faassen is pursuing a possible new, beefier hosting situation.
Thanks also to Sascha Welter, Michael Haubenwallner, Christian Theune and
anyone else I've forgotten, for dropping by #zwiki on irc.freenode.net and
helping to troubleshoot and fix issues.
Cheers!
- Simon
PS I seem to have too many places to post such news:
- zope-web list/gmane.comp.web.zope.websites newsgroup
- http://wiki.zope.org/AboutThisSite
- http://wiki.zope.org/zope3/AboutThisSite
- http://wiki.zope.org/zope2/AboutThisSite
- http://zwiki.org/GeneralDiscussion
I will probably pick just the first 2 or 3 henceforth.
_______________________________________________
Zope-web maillist - Zope-web@zope.org
http://mail.zope.org/mailman/listinfo/zope-web