[Twisted-Python] trac's reliability

James Y Knight foom at fuhm.net
Thu Nov 30 22:43:50 MST 2006


On Nov 29, 2006, at 6:56 PM, Magnus Lycka wrote:
> James Y Knight wrote:
>>
>> 2) Horrible horrible locking (?) issues on the /trac/browser page.  
>> Hitting it with concurrency > 2 makes it handle about one request  
>> every 2 seconds, probably database lock contention or something.  
>> But also, it starts giving error pages like this. This is an  
>> traceback when trying to generate an error page. Yay.
> Is this due to sqlite? I haven't looked at the Trac code, and I  
> couldn't see it
> explicitly in the traceback, but it certainly sounds a bit like an  
> sqlite vs
> postgresql  problem, and  it fits the description of the Trac  
> developers. The
> lock problems in sqlite are typcially related to writes though. Is  
> something
> written to the database here? What sqlite version is this? Sqlite 3  
> has a
> better locking mechanism than sqlite 2.

The traceback looks like it's from a bug in the error page generator,  
so the real error is obscured. It's using sqlite3. I'm no trac  
expert, so I can't actually answer the rest of your questions.

In other news: the master svn repository has been upgraded to 1.4.2  
fsfs (from debian testing), and thus trac now gets an up-to-date view  
of the repository. (currently synched via rsync every 10 minutes  
because that was most expedient).

So besides being horrendously slow, trac now seems to be in pretty  
good shape.

I think next it'd be nice if someone would install/enable/whatever  
the anti-spam plugin and re-enable editing the wiki for all  
registered users.

James




More information about the Twisted-Python mailing list