[ZODB-Dev] [Performance] excessive time in "Connection.afterCompletion"

[Jim Fulton]

Dieter Maurer wrote:
> Hi Tim,
> 
> Tim Peters wrote at 2006-3-31 18:48 -0500:
> 
>>...
>>
>>>What is the motivation to request synchronized transaction completion
>>>by default?
>>
>>Primarily so we could stop seeing endless threads on zodb-dev asking
>>why ZEO clients that didn't happen to modify any objects never saw
>>current data.  From the NEWS file:
>>
>>- (3.4a6) Relatedly, ``Connection`` implements ``ISynchronizer``, and
>> ``Connection``'s ``afterCompletion()`` and ``newTransaction()`` methods now
>> call ``sync()`` on the underlying storage (if the underlying storage has
>> such a method), in addition to processing invalidations.  The practical
>> implication is that storage synchronization will be done automatically now,
>> whenever a transaction is explicitly started, and after top-level
>> transaction commit or abort.  As a result, ``Connection.sync()`` should
>> virtually never be needed anymore, and will eventually be deprecated.
> 
> 
> Based on my observation, I fear it is a too expensive measure.
> 
> If an asyncore mainloop (or something equivalent) is running,
> then we would get the same effect without the sync.
> Maybe we should go this route (and ensure instead that a mainloop is
> always running).

You can do that now. Just start a thread with the main loop.

My plan, sometime in the future, uh, when I have time, is to
implement a separate thread (or pair of threads) to handle ZEO
network communication independent of any mainloop.

Jim

-- 
Jim Fulton           mailto:jim at zope.com       Python Powered!
CTO                  (540) 361-1714            http://www.python.org
Zope Corporation     http://www.zope.com       http://www.zope.org