On Mon, 9 Jun 2003, Eric Siegerman wrote:

> Well, "SCSI parity error" seems pretty clear -- data being
> received with bad parity by one of your SCSI devices (either the
> tape drive or the SCSI adapter; I'm being carefully agnostic as
> to which way the data's flowing when the errors are detected).
> Most likely the corruption is occurring on the SCSI bus.  The
> only other possibility I can think of is a firmware bug or other
> malfunction in one of the two devices.

seems reasonable.
 
> > can you think of any other non-amanda test i can do that will exercise
> > the tape drive as amanda does?
> 
> Only what I've already suggested -- use real disk files instead
> of /dev/urandom as your data source.  Based on what I'm about to
> say, it'd be wise to make sure those files come from the same
> drive your Amanda holding disk is on.  To get as close as
> possible to what Amanda's doing, use the holding-disk files
> themselves.

ok, i did this (made a big disc file from /dev/urandom, then dd'ed that to
the tape), and it still works fine.  i emphasise that my discs aren't on
the same SCSI bus as the tape drive - the tape drive has its own
controller, which no other device uses - but still no errors came.  i
dd'ed around 2GB of data to the tape, and had no problems at all.
 
> If, as I suspect, the problem is noise on the SCSI bus, one ugly
> possibility is that your SCSI cable is picking up noise from the
> rest of the system; for example, from the IDE cable that your
> disk is connected to.  No disk I/O => no noise being transmitted
> => no SCSI errors.

hopefully, this test has disproved that theory.

> The obvious first stab at an answer would be to just move the
> cables around inside the case, hoping to get the SCSI cable far
> enough away from the noise source.  But *please* figure out how
> to duplicate the problem first, otherwise you'll never know if
> you fixed it!

at the moment, duplicating it is *really* easy.  any time the amanda taper
tries to write to the tape, nothing works (write errors in the first
gigabyte).  the rest of the time, i can send data to the tape just fine.
 
> A better solution (if this is indeed the problem!) would be
> better-quality SCSI cables.

so noted.
 
> This is the sort of problem I *hate*!  Best of luck with it.

<grin> thanks for your ideas!  does anyone else have any suggestions, at 
all, for a tape drive which seems to be fine until amanda uses it, and 
which then goes south fairly rapidly?

does anyone know anything about decoding kernel SCSI error messages?


-- 

  Tom Yates  -  madhatter AT teaparty DOT net  -  http://www.teaparty.net