On Wed, 2007-01-31 at 12:14 -0500, Tom Lane wrote:
Is anything happening with that? This code fails regression on my
primary development machine, and my annoyance level is rising rapidly.
I'll probably remove the uuid test from the schedule files as a
band-aid, if no fix is forthcoming soon.
It's difficult to profile a backend server process (using gprof)
because each process overwrites any earlier profile as it exits.
It is especially tricky to nab a useful profile if you happen to have
autovacuum enabled.
This patch reduces the problem by forcing the backend to 'cd' to a new
[EMAIL PROTECTED] wrote:
And the patch is where?
You caught me; guess I'd better make something up fast, huh?
Here it is, thanks.
-- Korry
---
[EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote:
It's
Here's an idea and a patch for full page writes improvement.
Idea:
(1) keep full page writes for ordinary WAL, make them available during
the crash recovery, - recovery from inconsistent pages which can be
made at the crash,
(2) Remove them from the archive log except for those written during
Sorry we forgot the list this incompatibility in the 8.2.0 release
notes. I have applied the attached patch to list this item in the 8.2.0
documentation section. This will appear in the next 8.2.X release.
---
Mike G
[EMAIL PROTECTED] wrote:
The name for the define variable could perhaps be better - feels silly
adding -DLINUX_PROFILE on Freebsd! (maybe just PROFILE or GPROF_PROFILE?).
That wasn't my choice, there is other code elsewhere that depends on
that symbol, I just added a little bit more.
Mark Kirkwood wrote:
[EMAIL PROTECTED] wrote:
The name for the define variable could perhaps be better - feels
silly adding -DLINUX_PROFILE on Freebsd! (maybe just PROFILE or
GPROF_PROFILE?).
That wasn't my choice, there is other code elsewhere that depends on
that symbol, I just added a
Mark Kirkwood [EMAIL PROTECTED] writes:
Right - but LINUX_PROFILE was added to correct Linux specific oddities
with the time counter accumulation, whereas your patch is not Linux
specific at all. So I think a more representative symbol is required.
Yeah, that was my problem with the patch
On Thu, 2007-02-01 at 00:59 -0500, Tom Lane wrote:
Is there a way to detect via #if that profiling is enabled? I wouldn't
expect a really portable answer, but maybe there's something that works
for gcc?
What about a --enable-gprof (or --enable-profiling?) configure flag?
This could add the