[ZODB-Dev] ZEO cache bug? (again)

[Chris Withers]

Jim Fulton wrote:
> 
> I'll note that I have been able to provoke the error message, but, for 
> me at least, the result was non-fatal.  The error occurs during cache 
> verification and causes the connection to fail. The connection thread 
> keeps running and tries again.  

Okay, with the proviso that it's been a couple of months since I last
had to deal with this, that fits what I remember, except the retrying
always used to fail, so I'll just get a large number of the same error
in the event log.

> The error, as provoked by me, arises 
> from a race condition in a situation in which one client is connecting 
> while other clients are writing.  Unless other clients are writing 
> furiously, I don't see why this would prevent a client from connecting 
> eventually.  This makes me wonder if, perhaps, I'm not seeing the same 
> error that you are reporting. OTOH, I haven't been able to think of 
> another situation in which this symptom could occur.

Hmmm, well, the setup here is that one client is used for pumping data
into a ZODB, while the other is used mainly for reads but does some
writing (code modification to page templates, etc) That said, I don't 
know what the setups of the other people experiencing this error were...

> If you are seeing this error a lot, I suggest setting 
> invalidation-queue-size to 1 on your storage server (and restarting of 
> course). It would be interesting to know if this setting makes the error 
> go away.

The trouble is that it's a hard to reproduce error ;-)
Disabling persistent client caching was the big-hammer approach. I'll
try turning that back on for one of the dev servers and seeing if I can
reproduce, when I can, I'll try dropping the queue size to one as you
suggest...

cheers,

Chris

-- 
Simplistix - Content Management, Zope & Python Consulting
            - http://www.simplistix.co.uk