[1543] in Kerberos
Kerberos V4 and large databases
daemon@ATHENA.MIT.EDU (Robert Viduya)
Wed Sep 4 13:09:00 1991
Date: 4 Sep 91 16:15:06 GMT
From: robert@shangri-la.gatech.edu (Robert Viduya)
To: kerberos@shelby.Stanford.EDU
We've had a bit of trouble setting up Kerberos V4 (patchlevel 9) in our
environment. We've got over 20,000 user accounts and we've hacked up a
version of kdb_edit to allow us to add them to the database in bulk, rather
than interactively. Partway through the add, the database gets corrupted.
This happens on both a Sun4 and a DEC RISC. We've tried switching database
routines and, to date, have tried ndbm (as distributed), sdbm, libhash
(from BSD4.4) and, most recently mdbm. All of them, except for mdbm, have
failed, each at different points (sometimes at approximately 5,000 users,
sometimes at approximately 7,000 depending on the database library). We
discovered the corruption by running "kdb_util dump". In some cases
kdb_util goes into an infinite loop, in other cases it core dumps. Closer
examination revealed that the "nextkey" routine was returning bad data.
Mdbm was the only database library that hasn't gotten corrupted (yet).
We've got about 10,000 accounts in it now and I'm a bit concerned about
what may happen when we add the remaining 10,000. I've checked the
code as well as I could and couldn't find anything obvious. One other
bit of evidence which may or may not be significant was that on all the
other databases the "virtual" file size of the database was well over
a gigabyte (1.7) whereas with mdbm it's only 214 megabytes (both with
approximately the same amount of data).
Has anyone else seen this problem? Is there a fix?
robert
--
Robert Viduya robert@shangri-la.gatech.edu
Technical Services / Information Technology
Georgia Institute of Technology (404) 894-6296
Atlanta, Georgia 30332-0715
