Add reqests start timing out. Truss output shows threads lwp_mutex_lock (ITS#3148)

[puranik@lucent.com]

Full_Name: Ravi Puranik
Version: 2.1.29
OS: 
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (192.11.226.116)


+ OpenLDAP 2.1.29 is being used by our product which internally uses Sleepycat
4.2.52.
+ We have four databases which are maintained using the OpenLdap product.
+ There is a process which continuously reads billing data deposited on the
machine
add performs inserts to these databases.
+ Every evening there is another billing process which starts up at 7:00PM and
generates 
a billing file and  performs several thousand inserts in to these databases
during its run. 
+ The inserts are achieved by sending a LDAP adds to the Open LDAP which in
turn
performs 
the inserts into the databases.
+ It has been observed that the Open LDAP frequently gets into a state which
causes all the 
add requests to time out. The only recovery mechanism identified so far when in
this state
is to stop the "slapd" (OpenLDAP) process, perform a db_recovery and start
"slapd"
+ The truss output shows a bunch of threads in slapd stuck on lwp_mutex_lock  

I have truss and pstack output of the slapd process when it is in this state.
Please
let me know if there is a place I can ftp it to or an email address I can send
it.
(I tried ftp'ing the info to ftp.openldap.org/incoming but I get "No space left
on device"  error)

Thanks,
Ravi