On Wed, Feb 3, 2010 at 2:05 AM, Jeff Davis <pgsql@j-davis.com> wrote:
>> Thanks, very well spotted... Actually the same is true for LISTEN... I
>> have reworked the patch to do the changes to listenChannels only in
>> the post-commit functions.
>
> I'm worried that this creates the opposite problem: that a LISTEN
> transaction might commit before a NOTIFY transaction, and yet miss the
> notification.
See the following comment and let me know if you agree...
! /*
! * Exec_ListenBeforeCommit --- subroutine for AtCommit_NotifyBeforeCommit
! *
! * Note that we do only set our pointer here and do not yet add the channel to
! * listenChannels. Since our transaction could still roll back we do this only
! * after commit. We know that our tail pointer won't move between here and
! * directly after commit, so we won't miss a notification.
! */
However this introduces a new problem when an initial LISTEN aborts:
Then we are not listening to anything but for other backends it looks
like we were. This is tracked by the boolean variable
backendExecutesInitialListen and gets cleaned up in AtAbort_Notify().
> It seems safest to me to add a backend (LISTEN) to the list before
> commit, and remove a backend (UNLISTEN) after commit. That way we are
> sure to only receive spurious notifications, and can't miss any.
If a LISTEN aborted we would not only receive a few spurious
notifications from it but would receive notifications on this channel
forever even though we have never executed LISTEN on it successfully.
Joachim
