Re: Native DB replication for PG

On Fri, Apr 30, 2010 at 9:02 PM, Greg Smith <greg(at)2ndquadrant(dot)com> wrote:
> Joshua D. Drake wrote:
>>
>> On Fri, 2010-04-30 at 13:42 -0700, Gauthier, Dave wrote:
>>
>>>>>
>>>>> If I had to plan server deployments for the next year (and I do) I'd
>>>>> be sticking with pg 8.3 and a proven replication engine.  Next summer
>>>>>
>>>>
>>>> Surely you mean 8.4? :-)
>>>>
>>
>> No, I would buy the 8.3 argument as well. Depending on your conservative
>> level. 8.4 is fine and all but 8.3 is about as rock solid as it gets.
>
> Unless you don't vacuum enough on a bigger database, run out of FSM pages,
> and the whole vacuum strategy goes to hell afterwards.  I would say that
> running into that issue is *probable* for an 8.3 install of any significant
> size, whereas the odds of running into a regression in 8.4 relative to 8.3
> is pretty low.  The whole "the older version is always more reliable" mantra
> doesn't make sense when you've got a major known issue in the older release
> that just goes away by using the newer one, and I feel that's the case with
> 8.4 vs. 8.3.

Exactly. I've got a LOT of effort involved in free space map sizing
and monitoring on 8.3. However, for me it's no longer a serious
problem. Free space map is 10 to 20x what it needs to be on my
machines now and works like a charm in 8.3. 8.4 randomly crashed, and
honestly I can't afford to test and help fix it right now. This
summer I can and will either with 9.0 or 8.4. But we're talking db
crashes that were happening once every 2 to 3 weeks for me, so testing
it takes a lot of time for me. And I can't do it with my productions
servers. I tested 8.4 what I thought was fairly hardly last year only
to have 8.4.1 die under the same load that 8.3 handled without a
problem, and reverted to the known working version putting testing
8.4.1 on hold.

So to ME, the choice is a fully functional 8.3 installation that has
NO problems with free space map because of configuration choices, or
an 8.4 with a known (to me) issue of crashing and dying.