[Fwd: Re: misc/72498: Libc timestamp code on jailed SMP machinegenerates incorrect results]

[Justin Clift]

The following reply was made to PR misc/72498; it has been noted by GNATS.

From: Justin Clift <jc at telstra.net>
To: pgsql-bugs at postgresql.org
Cc: Uwe Doering <gemini at geminix.org>,
	freebsd-gnats-submit at FreeBSD.org
Subject: [Fwd: Re: misc/72498: Libc timestamp code on jailed SMP machine generates
   incorrect results]
Date: Tue, 12 Oct 2004 10:47:30 +1000

 Hi guys,
 
 Following up on the bug I posted the other day about incorrect 
 timestamps, the FreeBSD team are wondering if it might be caused by 
 conflicting PostgreSQL instances in different FreeBSD jails on the same 
 host machine.
 
 However, we've noticed no other problems with this configuration over 
 the last several and I was under the impression this is a fairly common 
 scenario.
 
 Is anyone able to verify or deny that PostgreSQL instances in different 
 jails (each with their own IP address) will not corrupt each other?  I'm 
 aware they allocate shared memory from a "global pool" of it as made 
 available on the host system, but have been under the impression 
 PostgreSQL is coded to not corrupt in this kind of situation.
 
 Regards and best wishes,
 
 Justin Clift
 
 
 -------- Original Message --------
 Subject: Re: misc/72498: Libc timestamp code on jailed SMP machine 
 generates   incorrect results
 Date: Mon, 11 Oct 2004 09:33:38 +0200
 From: Uwe Doering <gemini at geminix.org>
 Organization: Private UNIX Site
 To: Justin Clift <jc at telstra.net>
 CC: freebsd-gnats-submit at FreeBSD.org
 References: <200410110202.i9B228Ye060291 at www.freebsd.org>
 
 Justin Clift wrote:
 > 
 >>Environment:
 > 
 > FreeBSD was-dev.telstra.net 4.10-RELEASE FreeBSD 4.10-RELEASE #0: Fri Jun 25 14:23:42 EST 2004     root at verdelho.telstra.net:/usr/src/sys/compile/1GB_SHARED_V3  i386
 > 
 >>Description:
 > 
 > We're using PostgreSQL 7.4.5 in an SMP jailed environment on FreeBSD 4.10.
 > 
 > Inside this jail PostgreSQL is configured to output timestamp information in it's log file entries.  There appears to be a bug in the timestamp generation code, as for hours above 9 oclock (10am and onwards) the timestamp's being generated are occasionally incorrect:
 > [...]
 
 Do you happen to run more than one instance of PostgreSQL on that
 machine, each in its own jail?  If so, are you aware that jails don't
 have separate SysV shared resources (memory regions etc.), at least not
 in FreeBSD's original 4.x implementation?  In this scenario your problem
 might be caused by clashing PostgreSQL instances, and you're likely to
 be in for more serious problems than just time stamp corruption.
 
 Just an educated guess, of course.
 
     Uwe
 -- 
 Uwe Doering         |  EscapeBox - Managed On-Demand UNIX Servers
 gemini at geminix.org  |  http://www.escapebox.net