Обсуждение: RE: [HACKERS] VACUUM ANALYZE problem on linux
What user ID are you using? Are you "su"ing over the the postgres user ID. Also, make sure all of the environment variables are exported in the script. PGPORT, PGUSER, etc... D. -----Original Message----- From: Tom Lane [mailto:tgl@sss.pgh.pa.us] Sent: Wednesday, February 24, 1999 1:35 PM To: phd2@earthling.net Cc: PostgreSQL-development Subject: Re: [HACKERS] VACUUM ANALYZE problem on linux Oleg Broytmann <phd@sun.med.ru> writes: > 3. Run postmaster -b -D/usr/local/pgsql/data -o -Fe -S (to detach it) > and run VACUUM ANALYZE - worked > (I took these parameters from script /etc/init.d/postgres) > 4. Run /etc/init.d/postgres start > and run VACUUM ANALYZE - failed, no core file. So there is something different about the environment of your postmaster when it's started by init.d versus when it's started by hand. Now you just have to figure out what. I thought of environment variables, ulimit settings, ownership/permission settings ... but it's not clear why any of these would affect VACUUM in particular yet leave you able to do other stuff successfully. Puzzling. regards, tom lane
Hello! On Wed, 24 Feb 1999, Dan Gowin wrote: > What user ID are you using? Are you "su"ing over > the the postgres user ID. But of course! > Also, make sure all of the environment variables > are exported in the script. PGPORT, PGUSER, etc... I want to remind you and all who watch the thread: I have the problem on 3 different glibc2 linucies - heavily modified Debain 2.0, newly installed Debain 2.0 and modified RedHat 5.1. I have no problem with sparc-solaris, and I have a report from Tom Lane he has no problem with HP-UX. IWB interesting to test libc5-linux... > D. > > -----Original Message----- > From: Tom Lane [mailto:tgl@sss.pgh.pa.us] > Sent: Wednesday, February 24, 1999 1:35 PM > To: phd2@earthling.net > Cc: PostgreSQL-development > Subject: Re: [HACKERS] VACUUM ANALYZE problem on linux > > > Oleg Broytmann <phd@sun.med.ru> writes: > > 3. Run postmaster -b -D/usr/local/pgsql/data -o -Fe -S (to detach it) > > and run VACUUM ANALYZE - worked > > (I took these parameters from script /etc/init.d/postgres) > > 4. Run /etc/init.d/postgres start > > and run VACUUM ANALYZE - failed, no core file. > > So there is something different about the environment of your postmaster > when it's started by init.d versus when it's started by hand. Now you > just have to figure out what. > > I thought of environment variables, ulimit settings, > ownership/permission settings ... but it's not clear why any of these > would affect VACUUM in particular yet leave you able to do other stuff > successfully. Puzzling. > > regards, tom lane > Oleg. ---- Oleg Broytmann http://members.xoom.com/phd2/ phd2@earthling.net Programmers don't die, they justGOSUB without RETURN.
Say, is it possible that your VACUUM problem is a locale-related bug? If init runs with a different locale setting than hand-started processes, then that would affect index ordering ... which could perhaps cause fatal problems while vacuuming indexes. I could believe that VACUUM is not able to cope with indexes that appear to be out of order according to the sort operators it's using. This line of thought leads to the idea that indexes had better be marked explicitly with the locale that they're for. Or else we need to change Postgres so that the locale setting is hard-wired at compile time and not dependent on environment variables. regards, tom lane
Hi! On Thu, 25 Feb 1999, Tom Lane wrote: > Say, is it possible that your VACUUM problem is a locale-related bug? Hmm... May be. > If init runs with a different locale setting than hand-started > processes, then that would affect index ordering ... which could > perhaps cause fatal problems while vacuuming indexes. I could > believe that VACUUM is not able to cope with indexes that appear > to be out of order according to the sort operators it's using. I have no indexes in this database. Looks like I need toreproduce the same env in command line. I'll try. Oleg. ---- Oleg Broytmann http://members.xoom.com/phd2/ phd2@earthling.net Programmers don't die, they justGOSUB without RETURN.
On Thu, 25 Feb 1999, Oleg Broytmann wrote: > > I have no indexes in this database. > Looks like I need toreproduce the same env in command line. I'll try. why don't you create a sh script to start postgres out off, ans set -x and of cource a setenv etc.. Call the _SAME_ file from your init and from your command line and diff the results. -- Incredible Networks LTD Angelos Karageorgiou 20 Karea st, +30.1.92.12.312 (voice) 116 36 Athens, Greece. +30.1.92.12.314 (fax) http://www.incredible.com angelos@incredible.com (e-mail)
> Say, is it possible that your VACUUM problem is a locale-related bug? > > If init runs with a different locale setting than hand-started > processes, then that would affect index ordering ... which could > perhaps cause fatal problems while vacuuming indexes. I could > believe that VACUUM is not able to cope with indexes that appear > to be out of order according to the sort operators it's using. > > This line of thought leads to the idea that indexes had better be > marked explicitly with the locale that they're for. Or else we > need to change Postgres so that the locale setting is hard-wired > at compile time and not dependent on environment variables. That is an excellent point. I never considered that. In fact, initdb would also need identical locale, wouldn't it. -- Bruce Momjian | http://www.op.net/~candle maillist@candle.pha.pa.us | (610) 853-3000+ If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania19026
Oleg Broytmann wrote: > I have no indexes in this database. > Looks like I need toreproduce the same env in command line. I'll try. This (locale of indices) still might be it - the system tables have indices, which I believe get updated when you VACUUM (in fact, that's kind of the point) Ross -- Ross J. Reedstrom, Ph.D., <reedstrm@rice.edu> NSBRI Research Scientist/Programmer Computer and Information Technology Institute Rice University, 6100 S. Main St., Houston, TX 77005
Hi! On Thu, 25 Feb 1999, Ross J. Reedstrom wrote: > This (locale of indices) still might be it - the system tables have I recompiled postgres without locale, run initdb, load my dump - VACUUM ANALYZE worked :( I wonder, what may be wrong with locale support, that trigger only on glibc2? Oleg. ---- Oleg Broytmann http://members.xoom.com/phd2/ phd2@earthling.net Programmers don't die, they justGOSUB without RETURN.