On Thu, Jan 20, 2005 at 12:02:28PM -0800, Andrew Gillham wrote:
> Does anyone have time to do a binary search on cvs for the days
> between known good and known bad?
> 
> If someone has a test case that is fairly easy to populate the
> database and leave the script running, I can try to do the search this
> weekend on an i386 box.
> 
> I would need to know the latest possible dates for last known working
> kernels and the earliest possible known broken.
> 
> -Andrew

Hi,
I sent in a PR a little while ago.

    pkg/28729: PostgreSQL consumes 100% CPU after brief heavy loading

    http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=28729

It seemed all too easy to reproduce.  Just 

    repeat 20 sh -c "MyCommand >/dev/null &"

where MyCommand is a shell script that calls the PostgreSQL "psql"
command to do an SQL select over a join of some tables.

Normally a "repeat 50 ..." will mostly finish before I can type "ps t"
but not with the buggy version.

Reverting to 1st Nov 2004 version of postgresql74 works fine, ie

    cd /usr/pkgsrc/databases
    cvs up -D 2004-11-1

I'm using these versions of postgresql74 on a -current system from
arround Nov 24th and all works fine.  Slightly later pgsql versions
may work too.  The first priority was just to get things working again.

It must have happened between 2004-11-1 and 2004-12-21.
I can try to get a more precise date.

Bob