On 06/01/07 17:31, Andrew Sullivan wrote:
> On Sat, Jun 02, 2007 at 12:23:44AM +0200, Alexander Staubo wrote:
>> Could you not (I ask naively) detect the first DDL statement is
>> submitted in a transaction
>
> Maybe.
>
>> on the master, then start a transaction on
>> each slave, then funnel this and all subsequent statements
>> synchronously to every nodes, then prepare and commit everyone?
>
> You could if 2PC was ubiquitous, which is certainly wasn't when the
> code was designed (remember, it was originally compatible all the way
> back to 7.3). Some people suggested using 2PC "if it's there", but
> that just seems to me to be asking for really painful problems. It
> also entails that all DDL has to happen on every node at the same
> time, which imposes a bottleneck not actually currently in the
> system.
Since DDL is infrequent, is that bottleneck an acceptable trade-off?
> It is probably the case, however, that version 2 of the system will
> break some of these backwards compatibility attempts in order to
> depend on some new back end features -- putting this entirely in user
> space turns out to be awful. It's how we got the monstrous catalog
> corruption hack.
>
> This is getting pretty Slony specific, though, so if we're to
> continue this thread, I suggest we do it on the Slony list.
--
Ron Johnson, Jr.
Jefferson LA USA
Give a man a fish, and he eats for a day.
Hit him with a fish, and he goes away for good!