Re: Psycopg 2.2.x issue with PgBouncer

On 8/17/10, Federico Di Gregorio <fog-NGVKUo/i/[email protected]> wrote:
> On 04/08/2010 13:20, Daniele Varrazzo wrote:
>  > 2010/8/4 Devrim G=DCND=DCZ <[email protected]>:
>  >> >
>  >> > Where are we on this? I will move to new servers next week with 2.2=
.X
>  >> > series. I can rollback to 2.0.14, but I'd prefer to stick to 2.2.X
>  >> > series. Was this fixes somewhere?
>  > If you ensure to have a commit/rollback before close/GC of the
>  > connection, PgBouncer should work fine with all the psycopg versions.
>  > If you can make this change and run the program with the current 2.2.2
>  > a feedback would be very appreciated.
>  >
>  > I haven't heard from fog if he prefers to have 2.2.3 reproducing the
>  > behavior of pre 2.2, where a rollback was automatically issued on
>  > close(), or if drop it for the reasons discussed in this thread.
>
>
> Hi *,
>
>  before going on vacancy I didn't do the release because there was a
>  discussion going on. From what I read it seems to me that rolling back
>  the changes on master and documenting the behaviour is the best way to
>  fix that. Right?

I'm not familiar with changes in master.  The behaviour I was complaining
about - unexpectedly heavy .close()/.__del__() - is seen in 2.0.x.

If you are planning to document that as official feature,
then I give -1 to that.

I'd suggest pushing forward and remove any expensive operations from
.close() and document _that_ as feature.

-- =

marko