Re: psycopg and gevent with pgbouncer

From: Daniele Varrazzo <daniele(dot)varrazzo(at)gmail(dot)com>
To: Marko Kreen <markokr(at)gmail(dot)com>
Cc: Raoul Duke <knevcher(at)gmail(dot)com>, psycopg(at)postgresql(dot)org
Subject: Re: psycopg and gevent with pgbouncer
Date: 2011-10-06 10:31:04
Message-ID: CA+mi_8ZVDUYhu721LLnPL3S2-LJJN3AgZrNpSBXSd7U3eNpoHA@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: psycopg

On Thu, Oct 6, 2011 at 9:54 AM, Marko Kreen <markokr(at)gmail(dot)com> wrote:
> On Thu, Oct 6, 2011 at 10:45 AM, Raoul Duke <knevcher(at)gmail(dot)com> wrote:

>> 2. I use psycopg simple connection pooler :
>> pool = SimpleConnectionPool(1, 3, connection_string)
>>
>> if I add async option:
>> pool = SimpleConnectionPool(1, 3, connection_string, async=1)
>>
>> I've got an error, than I can't understand:
>>   File "/usr/lib/pymodules/python2.6/sqlalchemy/engine/base.py", line 940,
>> in __create_execution_context
>>     self._handle_dbapi_exception(e, kwargs.get('statement', None),
>> kwargs.get('parameters', None), None, None)
>>   File "/usr/lib/pymodules/python2.6/sqlalchemy/engine/base.py", line 931,
>> in _handle_dbapi_exception
>>     raise exc.DBAPIError.instance(statement, parameters, e,
>> connection_invalidated=is_disconnect)
>> OperationalError: (OperationalError) asynchronous connection attempt
>> underway None [{}]
>>
>> Do I understand that when using the gevent, I have to use async option?
>
> I leave that to psycopg guys to answer.

An async connection it is not usable just after it's been created:
it's in an half baked state where poll() returns POLL_READ or
POLL_WRITE until the client and the server have completed the
asynchronous login handshake. The protocol must be pushed by the
client as described in
<http://initd.org/psycopg/docs/advanced.html#asynchronous-support>.
The wait loop should be somewhere after the connection creation and
before the usage: I don't know if the libraries you are using allow
that.

Note that if you are using gevent and not twisted you may use the
green mode instead
<http://initd.org/psycopg/docs/advanced.html#support-to-coroutine-libraries>,
that provides you async connections whose behaviour is entirely
equivalent to the blocking one (they block a single greenlet and
schedule the other to go). I suspect you should try them instead of
the raw async mode.

-- Daniele

In response to

Responses

Browse psycopg by date

  From Date Subject
Next Message Daniele Varrazzo 2011-10-06 10:37:07 Re: Hstore array implementation.
Previous Message Daniele Varrazzo 2011-10-06 10:16:51 Re: Hstore array implementation.