Bruce Momjian wrote:
> Don Baccus wrote:
> > Folks who have popular web sites with a world-wide audience don't have
> > the traditional early-morning "quiet periods", etc that local databases
> How do you get around vacuum downtime?
I'll attempt to field that one, as I am helping a little with the port
of this same toolkit, and have been using PostgreSQL in moderate
intranet/light internet production for two and a half years (since 6.1.1
-- scary thought).
I vacuum nightly, at semi-random times around my quietest times, which
are around 3-4 AM EDT. While 6.[1234] were pretty hokey around those
times, like locking out readers during vacuum... but 6.5.x drastically
improved the situation, to where I have not seen any error returns or
noticeable delays during vacuum times -- but, then again, I don't have
very many accesses during that time.
Now if a continuous vacuuming storage manager could be built... I can
see conceptually how one would go about it, I am nowhere near
confortable trying to do it myself. However, the list of 7.1 things
todo already is staggering -- several major projects all at once. IMHO,
those major projects should be tackled before relatively minor ones
are. In particular, once the fmgr redesign is done, the separate Alpha
patches may get to be retired. The WAL stuff is essential for good
recoverability, large tuples have been on nearly everyone's wish list
for a very long time, and lack of outer joins are a hindrance,
particularly when porting a web toolkit from Oracle :-). Although,
CONNECT BY would be nice for Oracle porting :-)
In any case, the PostgreSQL team's progress from 6.1.1 to now is more
than impressive.
--
Lamar Owen
WGCR Internet Radio
1 Peter 4:11
