On Wed, Dec 15, 2010 at 4:20 AM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On Sat, 2010-11-13 at 19:16 -0500, Robert Haas wrote:
>
>> 3. The third patch (relax-sync-commit-v1) allows asynchronous commit
>> even when synchronous_commit=on if the transaction has not written
>> WAL. Of course, a read-only transaction won't even have an XID and
>> therefore won't need a commit record, so what this is really doing is
>> allowing transactions that have written only to temp - or unlogged -
>> tables to commit asynchronously.
>
> I like this, great idea.
>
> Avoiding the commit record entirely will break Hot Standby though, since
> we rely on the assumption that all xids that are assigned are also
> logged. The xids would be "known assigned", yet since they never
> actually appear they will clog up the machinery (pun unintended).
Uggh, that's a really, really bad pun.
I made the same observation to Tom somewhere-or-other (must have been
a different thread because I don't see it on this one), along with the
further observation that we actually could suppress the commit record
entirely if wal_level < hot_standby, but I'm not sure there's enough
benefit to doing that to worry about the additional complexity.
Changing it from a foreground flush to a background flush already wins
so much that I don't really see the point of doing anything further.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company