Thread: Server Problem

Server Problem

From
"Dave Page"
Date:
OK, obviously these things are sent to try us :-)

The new server hosting the cvs and developer has developed what we can
only put down to a motherboard fault - basically;

- The system clock is running about twice as fast as it should, no
matter what I do with ntpd or adjtimex! Bizarrely though the hw clock is
fine, but I found some posts on the net suggesting there might be a
phase locked loop problem on the mobo.

- APIC errors have started appearing on boot for each CPU :-(

Currently, the system appears stable (it has never crashed), and the
system clock is being resynced to the hardware clock every 5 minutes so
if you see any odd timestamps in CVS, that's why!

We have two of these boards both of which have had problems (the other
is actually in Germany being repaired at the moment!) - suffice it to
say a replacement will be ordered and fitted ASAP (Intel this time
instead of Gigabyte). Unfortunately we had another one die completely
yesterday (a very old MSI with dodgey electrolytic capacitors) so we
don't have a spare in stock right now.

Anyhoo, just thought I'd give you a heads up.

Regards, Dave.

PS. I think I'm getting the flu so may be offline for a coupla days :-(

Re: Server Problem

From
Andreas Pflug
Date:
Dave Page wrote:

>OK, obviously these things are sent to try us :-)
>
>

Those glory days, when we made calculations using fingers, stones and
shells....

>The new server hosting the cvs and developer has developed what we can
>only put down to a motherboard fault - basically;
>
>- The system clock is running about twice as fast as it should, no
>matter what I do with ntpd or adjtimex! Bizarrely though the hw clock is
>fine, but I found some posts on the net suggesting there might be a
>phase locked loop problem on the mobo.
>
>
>

Maybe the machine is in a hurry, but to do what?!?

>- APIC errors have started appearing on boot for each CPU :-(
>
>Currently, the system appears stable (it has never crashed), and the
>system clock is being resynced to the hardware clock every 5 minutes so
>if you see any odd timestamps in CVS, that's why!
>
>

We should be very careful about that, I've seen some really nasty
problems with version checking systems when timestamps are not precise
(with Visual SourceSafe)

Quite often, I do an update just before committing. THIS SHOULD BE
AVOIDED as long the server's clock might take back steps. Instead, there
should be > 5 minutes of delay after updating (if clock is synced every
5 minutes) . This could prevent some future headache on repository
consistency.

>PS. I think I'm getting the flu so may be offline for a coupla days :-(
>
>
Already got one, still fighting to suppress it. BTW, maybe
dev.pgadmin.org caught the double speed virus?!?

Regards,
Andreas


Re: Server Problem

From
"Dave Page"
Date:

> -----Original Message-----
> From: Andreas Pflug [mailto:pgadmin@pse-consulting.de]
> Sent: 18 December 2003 11:14
> To: Dave Page
> Cc: pgadmin-hackers@postgresql.org
> Subject: Re: [pgadmin-hackers] Server Problem
>
> Dave Page wrote:
>
> >OK, obviously these things are sent to try us :-)
> >
> >
>
> Those glory days, when we made calculations using fingers,
> stones and shells....

Shells? Man you live in the lap of luxury. I'll bet you even had fire in
your cave!

>
> Maybe the machine is in a hurry, but to do what?!?
>

Dunno. It's in a rack with some very powerful boxes though - maybe it's
trying to keep up.

>
> We should be very careful about that, I've seen some really
> nasty problems with version checking systems when timestamps
> are not precise (with Visual SourceSafe)

Yeah, well - there are good reasons why we're in the middle of moving
all our Windows code from vss to CVS.

Anyway, I don't think it will be a problem unless successive commits of
the same file are made within a short time of each other (or an update
then commit) - we just need to keep an eye on things until the new mobo
arrives.

> Quite often, I do an update just before committing. THIS
> SHOULD BE AVOIDED as long the server's clock might take back
> steps. Instead, there should be > 5 minutes of delay after
> updating (if clock is synced every
> 5 minutes) . This could prevent some future headache on
> repository consistency.

I really should read the whole message before I type...

> Already got one, still fighting to suppress it. BTW, maybe
> dev.pgadmin.org caught the double speed virus?!?

Hmm, worth looking into :-)

Regards, Dave