Re: AIX: mirrored drives and Oracle corruption

You were right on target! It is apparently a microcode problem. BTW, it was mirrored.

Thanks,
Steve

"certified computer guy..." wrote:

> Steve Perry wrote in message <37AF8150.9DF4E927_at_sprynet.com>...
> >Hi Norman,
> >
> >I didn't state it clearly. One of the drives fail, Oracle chokes, AIX
> doesn't
> >know, shutdown the db, sys admin replaced the drive, we had to recover the
> files
> >on the mirrored drive. The other half of the mirror was fine. Oracle sensed
> the
> >problem, but AIX didn't.
> >I would have thought AIX would have noticed it and an alert would have been
> sent
> >to the SA. They would have notified us and scheduled maintenance to replace
> it.
> >
> >Thanks,
> >Steve
>
> First of all, if you have a drive "fail" and no errors are reported in the
> error report, you've got a serious microcode/device driver problem that
> needs to be addressed. The only time I've ever seen that happen is when the
> microcode on the drive and adapter are incompatibile, therefore the adapter
> "doesn't understand" what the disk is telling it and drops the event all
> together. This is of course a real bad situation.
>
> Second, even if the drive fails and no error is reported, if the LV is
> properly mirrored AIX should be able to successfully read/write the data to
> the copy on the mirrored drive. In that scenario, oracle or any other
> application would never know the error occurred because AIX would be
> returning a successful I/O completion to the calling application. However;
> with a microcode incompatibility issue causing "oracle corruption" it is
> more likely the write to both mirrors are failing to complete successfully.
>
> There is one other possibility here. At AIX 4.3.2 on the 7017/S7x systems a
> problem has been noted such that oracle LV's become corrupted. If your
> system meets this criteria, then you need the workaround APAR.
>
> Otherwise, check your mirroring by running "lslv -m {lvname}" and ensuring
> that no LPs map to two PPs on the same disk. This would be a problem which
> would cause you to lose data in a mirrored environment and in fact if the
> LV's are laid out that way I would tend to ask why mirror at all as it
> servers no usefull purpose. A simple "lslv {lvname}" should show you
> COPIES: 2, and the number in the PPs: entry should be twice the number in
> the "LPs:" attribute. If not, mirroring is not even turned on.
>
> Good luck.
Received on Thu Aug 12 1999 - 21:48:19 CDT