Re: DBCC CHCECKTABLE faults go away after reboot?
From: Paul S Randal [MS] (prandal_at_online.microsoft.com)
Date: 04/29/04
- Next message: Peter Feakins: "Auto Create Stats & Auto Update stats"
- Previous message: Wilhelm: "Decimal column multiplication is rounding"
- In reply to: Robert Carnegie: "DBCC CHCECKTABLE faults go away after reboot?"
- Messages sorted by: [ date ] [ thread ]
Date: Thu, 29 Apr 2004 13:21:05 -0700
Hi Robert,
My suggestion is that you call Product Support to help sort this out - its
too complex to do over a newsgroup.
Various comments on your problem description:
The database with corrupt sysindexes cannot be repaired and will need to be
restored from a backup.
CHECKDB should never crash - if by crash you mean AV/assert that leaves a
sqlxxxxx.dmp dump file in your log directory - if not, what do you mean?
You could see it 'hang' when processing on a very busy system - it's not
hanging but processing log records - this is by design and one of the
documented reasons not to run CHECKDB on a very busy system.
A corruption in one database cannot affect another database's structural
consistency.
Trace flag 2528 is the documented way to prevent parallel processing during
DBCC CHECKDB.
Thanks
-- Paul Randal Dev Lead, Microsoft SQL Server Storage Engine This posting is provided "AS IS" with no warranties, and confers no rights. "Robert Carnegie" <rja.carnegie@excite.com> wrote in message news:3D647628-D83A-4669-B313-F978AFEEB1C3@microsoft.com... > One of our clustered servers lost one short of a critical number of disks at once - same problem as the Hubble Space Telescope, the spin units don't all spin right any more. I can't be sure we didn't have data corrupted on the disks that continued working. I think it did go down. > > Our tech people put the files of the user databases - a hundred or more - onto another, existing server. The new server is SQL Server 2000 Service Pack 3 and the old one was SP2, and the new one isn't a cluster, but I believe that isn't supposed to matter, if you don't have replication running. > > And since then, we appear to have, > - one database with sysindexes table failing DBCC CHECKDB, which can't fix it > - spawning CHECKDB fails across other databases, two or three a day, some of which disappear upon reboot or service restart > - various crashes when running DBCC CHECKDB, perhaps particularly if while the serveis in use, around times 0800-1800 > > The new hardware ran SQL happily for a while in another role before now, so should be good, but who knows. > > What I mostly want to ask is whether any of this is normal. As I understand startup recovery, it merely ensures that transactions are properly applied to databases or are discarded, and if a database is un-OK by CHECKDB immediately before shutdown then it would be un-OK afterwards. I also don't like DBCC crashing. On one shot, it left four processor cores spinning at 100% usage indefinitely (with strange double dips to 0%) - since then I've found "DBCC TRACEON (2528)" as a claimed, implied, preventative for /that./ > > I am considering the idea that the one unfixable database is leaking confusion into the others, both in RAM and sometimes written to disk. Fixing it, by exporting all data to a new database with no system data faults and dropping the bad one, is therefore high on the agenda, but is tricky, because there are a hundred or so tables, in complex relationships. > > Other comments, advice, or war stories, are invited. > > Yes, we do also take backups, and no, I don't know... look, I imagine if my colleagues had noticed one bad database (our databases are for school administration, and the bad one represents one "primary" [elementary] school) amongst those, they'd have used its backup instead, so either it checked out OK or they didn't check. I'm not sure which would worry me more.
- Next message: Peter Feakins: "Auto Create Stats & Auto Update stats"
- Previous message: Wilhelm: "Decimal column multiplication is rounding"
- In reply to: Robert Carnegie: "DBCC CHCECKTABLE faults go away after reboot?"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
