This one sounds like the problem they may be hitting. As far as I remember,
this bug is connected with control file corruption on normal shutdowns of
the database which was running non-stop more than 248 days or so. Solaris
has a ticker counting number of seconds since last restart, this ticker overflows
and becomes negative and Oracle goes nuts. The workaround was to reboot
the system once in a while and if it was up more than said amount of time,
shutdown abort the oracle instance and reboot the system. They also said that
if controlfile corruption is already in effect, there is no other way to get the
system back than to reinstall and restore from last known good backup. By the
way, the bug was not platform-specific, it was confirmed on other unixes too,
though time to failure was different for some of them (where ticker counted
hundredths of a second the failure was coming earlier).
Correct me if I am wrong.
--
Vladimir Zakharychev (bob@dpsp-yes.com) http://www.dpsp-yes.com
Dynamic PSP(tm) - the first true RAD toolkit for Oracle-based internet applications.
All opinions are mine and do not necessarily go in line with those of my employer.
"Burt Peltier" <burtpelt_at_bellsouth.net> wrote in message
news:GREM7.49186$q36.3417177_at_e3500-atl2.usenetserver.com...
> For Solaris, there is a bug that is related to how long the machine has been
> up. I thought it was going to be fixed in 8.1.7 , but it causes things to
> hang and usually a reboot is required. I think the number of days before
> the hang starts is dependent on some Solaris setting and for most machines
> is around 244 days, but could be a lot lower . Doing this from memory... I
> don't have my work account handy, but the bug is documented in Metalink.
>
> Also, sometimes putting sql_trace = true in the init.ora will cause Oracle
> to give more messages, especially useful during shutdown or startup . This
> generates trace files for EVERY process , so I wouldn't let users in while
> doing this .
>
> "Sean Leach" <sleachNOSPAM_at_netlojix.comnospam> wrote in message
> news:20011126.163450.1295166342.813_at_netlojix.comnospam...
> > Hi,
> >
> > Sorry, but here is a very vague trouble description:
> >
> > Platform: Sun Solaris 8 (Sparc), latest patch clusters
> > DB: Oracle 8.1.7.0.0
> >
> > For some reason, the database will just hang, no errors in any of the
> > alert or trace logs, and sometimes, it will correct itself, and other, we
> > can do a "shutdown abort" to stop the instance, but all of the child SID
> > processes are still there, and eventually go to a "<defunct>" state,
> > such that we have to reboot the machine to fix it.
> >
> > Sometimes, we can do a SELECT COUNT(*) FROM table1, and it will succeed,
> > but then a SELECT COUNT(*) FROM table2 will fail. Sounded like a table
> > lock issue, but not sure...
> >
> > Sometimes, we can restart the listener, and that fixes it, other times,
> > it doesn't.
> >
> > The db data is mounted on a NetApp, but we ran for over 200 days with no
> > problems before, and then this problems just suddenly started happening.
> > We have disk_async_io set to false, as we had a problem with asynchronous
> > IO before, and that fixed THAT problem, but now we have this no error
> > hanging. It sucks as I have no where to start looking. Any tips?
> >
> > Thanks,
> > Sean
>
>
>
Received on Tue Nov 27 2001 - 11:16:59 CST
