Re: Log full with gigabyes of CurTransactionContex

Michael Monnerie <michael(dot)monnerie(at)is(dot)it-management(dot)at> writes:
> On Montag 15 Juni 2009 Tom Lane wrote:
>> These two settings are probably the cause. With shared_buffers at
>> 2GB, you do not have anywhere near 1GB to play around with in a
>> 32-bit environment. Try something like 200M and 500M.

> Wouldn't a recommendation to switch to 64bit PostgreSQL be a good idea
> also? Maybe his servers can't do that, I don't know.

If he wants numbers that high, he'd need to go to a 64bit build.
It's not clear from here whether any actual benefit would ensue, though.

> Also, I've heard that using shared_buffers = 500MB is enough (even in
> 64bit), and the rest will be cached by Linux cache alone. Is that true?

Yeah, the existence of kernel disk cache tends to negate the benefit of
very large shared_buffer settings. See the pgsql-performance archives
for many many discussions about such matters.

regards, tom lane