The symptoms:
-------------
A DEC 3000-400s running Digital Unix 3.2c and Networker Save and
Restore (NSR) version 3.1A was unable to (correctly) recover a
large (150,000 file) file domain after a 9Gb Seagate died. The
problem was manifested by the (now infamous) "too may open files"
message.
The confusion factor:
---------------------
Prior to 3.1A if NSR tried to save a fileset containing more than
4080 zero length files it used up all of the available file
descriptors for the process. That problem was fixed by patch
nsrv31-004 and by version 3.1A, for the save procedure, not the
recover procedure. A similar patch is now being created for the
recover procedure.
The real problem:
-----------------
There was a change made between version 3.0 and 3.1 which will
encounter a problem if the following conditions are met.
a) you need to perform a full fileset restore
b) the last full backup was done using version 3.0
c) the incrementals were done with 3.1
d) the restore is being performed with version 3.1 or 3.1A.
Under those conditions, NSR will incorrectly read the full backup
and lay down the files with zero length and the current date.
(Since the files have the current date, subsequent incrementals
don't seem to want to overwrite them.)
AND
if there are more than 4080 files the "too many files open"
message will be produced (by nwrecover), after which point it
will stop restoring files.
If there are less than 4080 files the process appears to work but
all of the files have zero length and the current date.
The workaround:
---------------
The backup tapes are fine. What I had to do was revert to 3.0 and
recover the full backup. I then upgraded to 3.1A and recovered
the incrementals. The disk has been restored and I am in the
process of doing full backups of all my ADVFS file domains using
3.1A. I did not realize how much had improved in 3.1 until I had
to used 3.0 again.
The fix:
--------
A patch should be available from CSC sometime next week.
Special thanks to:
------------------
Ellis Young - Digital CSC
Pascal Pederiva - Digital UNIX Software Support (Switzerland)
Mark Sprague - MIT
Paul E. Rockwell - Digital UNIX Sales Support Consultant (CT)
Knut Helleboe - Norsk Hydro a.s
Harold Bussey - bussey_at_draper.com
Draper Laboratory
Cambridge, MA
Received on Fri Sep 08 1995 - 17:07:23 NZST