slapd locks up with entry_rdwr_rtrylock repeated 154041 times

["Scott Stanley" <stanleys@shaw.ca>]

We are using OpenLDAP 2.0.21 with the Sleepycat 3.2.9 ldbm backend.  We are
experiencing random occurrences where the slapd process takes 99% of the
server CPU and the server becomes unresponsive for 10-30 seconds.  It
doesn't match a pattern of heavy load as we can have 50 users logged in
without any errors and later in the day with only one or two users we will
encounter the problem.

After turning on ldap debug messages and capturing the problem we found huge
number of repeated messages from entry_rdwr_rtrylock from the same thread
and ID:

Nov 16 21:56:49 assignmentbox slapd[5406]: entry_rdwr_rtrylock: ID: 30
Nov 16 21:56:59 assignmentbox last message repeated 154041 times
...
Nov 16 21:56:59 assignmentbox slapd[5406]: entry_rdwr_rtrylock: ID: 30
Nov 16 21:57:05 assignmentbox last message repeated 108970 times
...
Nov 16 21:57:05 assignmentbox slapd[5406]: entry_rdwr_rtrylock: ID: 30
Nov 16 21:57:07 assignmentbox last message repeated 25407 times

Once it gets the lock it completes the request and continues after a large
delay.

Anyone seen this before?  Any suggestions on how to solve this problem?

Thanks,

Scott