The mysterious death of kprop when running incremental propagtion

[William Clark]

Here is my setup as of now.  I have a single master KDC, and 9 slave KDC’s.  I have incremental propagation set up at 2m interval, and it works quite well for a little while.  At some indeterminate time, KDC’s start getting really far out of sync and I notice that kprop has died on these servers with a SIG ABRT.  Any attempt to restart kprop does not start it.  The only way I have seen to restart it is to remove principal.ulog file on that mdc and then restart.  It then runs just fine.

Couple of thoughts / contemplative questions:
- Could this potentially be FD related?  I am not running out of FD’s at the time this happens though…
- Could this be load related.  I am required to run 'kdb5_util dump' every 10 mins to gather data that is then audited.  There are about 80k + principals in my DB, but the process takes less than 20 seconds.  During this time I wonder if the principal DB is getting locked, and if this is causing kprop/kadmin to get in a very funny state.  Is this even a viable concern?

Need some help on this before I am forced to go back to old propagation methods.


William Clark