Re: Error during garbage collection after calling isready() on a closed connection

On Sat, Jan 10, 2009 at 07:08:24PM +0100, Federico Di Gregorio wrote:
> Il giorno ven, 09/01/2009 alle 17.13 +0200, Brian Sutherland ha scritto:
> > Hi,
> > 
> > Attached is an example script of a nasty little bug I found. It looks
> > like calling .isready() on a server closed connection causes an error
> > during the next garbage collection. That can cause all sorts of havoc on
> > a python process.
> > 
> > I found it in version 2.5.1 and verified that it's still there in 2.0.8.
> 
> Hi,
> 

I've backported your patch to 2.0.5.1 and it seems to work perfectly.
Thanks!

> cursor.isready() now raise an exception if the internal call to libpq
> reports problems. This is probably now _exactly_ what you want because
> the first call to isready() after the disconnection results in a notice
> (correctly appended to connections notices) and not in an error. An
> error will be raised on any successive call that try to use the now
> dropped connection.

Yes, raising an error on the isready() is _exactly_ what I want :)

That way I can dis-connect and try re-connect.

> This fix is available from bzr repository.
> 
> This seems correct to me, because the cursor is ready and is reading
> some "cached" data (the disconnection message) that was sent by the
> server just before dropping the connection.

In my application whether the exception is raised on the first or the
second isready() calls isn't important. In fact, I can't think of a
situation where it's important that the exception be raised exactly on
the first isready() call after the connection is dropped.

> Technical note: even after consuming the input and checking the
> connection (PQconsumeInput and PQisBusy calls), the connection is still
> CONNECTION_OK so on the first call to isready() after the server restart
> we don't have any way to understand what's happening except for parsing
> the notice (unrealiable IMHO).
> 
> federico
> 
> -- 
> Federico Di Gregorio                         http://people.initd.org/fog
> Debian GNU/Linux Developer                                [email protected]
> INIT.D Developer                                           fog-NGVKUo/i/[email protected]
>  And anyone who yells "fork" deserves to get one stuck in them.
>                                                           -- Dan Winship



-- 
Brian Sutherland