|From:||ITAGAKI Takahiro <itagaki(dot)takahiro(at)oss(dot)ntt(dot)co(dot)jp>|
|Cc:||Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, Alvaro Herrera <alvherre(at)commandprompt(dot)com>, Greg Smith <gsmith(at)gregsmith(dot)com>|
|Subject:||Re: pgbench duration option|
|Views:||Raw Message | Whole Thread | Download mbox | Resend email|
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> My understanding is that it's basically "cheap PC hardware" (with clock
> interfaces based on old ISA bus specs) that has the issue in a
> significant way. I wouldn't expect you to see it on a serious database
> server. But lots of people still do development on cheap PC hardware,
> which is why I think this is worth worrying about.
Ok, I rewrote the patch to use SIGALRM instead of gettimeofday.
The only thing I worried about is portability issue. POSIX functions
like alarm() or setitimer() are not available at least on Windows.
I expect alarm() is available on all platforms except Win32 and
used CreateTimerQueue() instead on Win32 in the new patch.
(We have own implementation of setitimer() in the server core, but pgbench
cannot use the function because it is a client application.)
Comments welcome and let me know if there are still some problems.
NTT Open Source Software Center
|Next Message||Alvaro Herrera||2008-08-19 02:39:32||Re: pgbench duration option|
|Previous Message||Tom Lane||2008-08-19 02:15:38||Improving non-joinable EXISTS subqueries|