On Tue, Sep 18, 2012 at 1:08 PM, Daniele Varrazzo
<daniele.varrazzo@gmail.com> wrote:
> On Tue, Sep 18, 2012 at 12:53 PM, Daniele Varrazzo
> <daniele.varrazzo@gmail.com> wrote:
>> On Fri, Aug 3, 2012 at 2:42 PM, Richard Harley <richard@scholarpack.com> wrote:
>>> We had the exact same problem with psycopg2/zope2.10/pg9 but could not find
>>> a satisfactory solution.
>>>
>>> There are no problems with psycopg1 apart from it being old and unsupported
>>> :)
>>
>> Sorry, I missed these messages. The problem is probably in the pool
>> rolling back open transactions since psycopg 2.4.3.
>>
>> I need some help with testing this issue: ok for you to test some
>> patch I will provide?
>
> So, here it is. The patch attached assumes zope always uses
> psycopg.pool.PersistentConnectionPool or its subclasses: is this the
> case?
>
> Can you please test:
>
> 1. what happens with regular pages that got broken as in the provided
> case? Desired result is that everything works as expected.
>
> 2. what happens with pages making an error? Desired result is
> eventually an error in the page but further pages generated correctly.
>
> 3. what happens if the database is restarted? Desired result is
> eventually an error in the first page loaded but further pages
> generated correctly (the pool should restore the connection).
>
> Thank you very much.
From analysis performed by Wolfgang Eibner in ticket #125, I suspect
something broke when ticket #73 was fixed and doesn't involve the pool
(although the pool rolling back the connection would still cause
problems).
If this is the case, the source of the problems is to be found in
ZPsycopgDA/db.py, in DB.getconn.
Any help to fix the issue would be appreciated.
-- Daniele