Andrew Sullivan <ajs at crankycanuck.ca> writes:
> On Tue, Jul 11, 2006 at 05:13:14PM -0400, Christopher Browne wrote:
>> One wild-eyed idea came up in the discussions of the last week...
>> 
>> Is it possible that there might be some small value to the notion of
>> running certain operations via 2PC?
>
> What worries me about this is the way the locking will work.  Won't
> this cause some rather nasty locks on all the nodes for (e.g.) DDL? 
> ALTER TABLE takes an ACCESS EXCLUSIVE lock according to the
> documents.  AFAIK, that lock has to be maintained until COMMIT, and
> across node crashes.

Consider all that stipulated...

Yes, this mandates a pretty ferocious degree of locking, rather
"meaner and nastier" than what is already the case.

At times like the moment of switching origin to a new node, or
applying some "ferociously dangerous" DDL script, it's possible that
it could be warranted.

The idea was not brought up as a proposal to switch everything (or
necessarily anything) over to 2PC; just to suggest that this is an
increasingly-available feature in modern PG releases, and it seemed
worth asking if there could be anything to be gained by using it.

When I mentioned it, it got some quizzical "I'll have to think about
that" responses.  If someone sees some massive unperceived benefit,
they can always point it out.  It's not realistic to seriously
consider TPC until we're thinking about dropping 7.4 as a supported PG
release, so there's plenty of time :-).
-- 
(format nil "~S@~S" "cbbrowne" "ntlug.org")
http://www.ntlug.org/~cbbrowne/spreadsheets.html
"Intel engineering seem to have misheard Intel marketing strategy. The
phrase was ``Divide and conquer'' not ``Divide and cock up''"
-- <iialan at www.linux.org.uk> Alan Cox