We have a sql server 2005 sp2 on w2k3 standard server. The servers started crashing every 4/5 days and have done this twice so far. checking the event viewer there are always events(2020) happening before the crash.
I have done some memory leak monitoring using performance monitor/poolmon/process explorer. and here is my findings:
sqlservr and lsass handles count keeps increasing linearly with time, i restarted the sql service yesterday and the count originally was at 900, this morning the count is at 29,000 for lsass and 28,000 for sqlservr.
using process explorer i can see that the majority of these handles are security context tokens.
we have a monitoring agent running on that server(altiris agent) which continousely logon/logoff to SQL to do some metric checks. so basically the access tokens created as a result of that never get released during the logoff.
also i noticed in poolmon that TOKE has the highest Bytes consumption of paged memory(50MBs) the token sizes are small (<1kb). so that kind correlates with the handles count problem mentioned above, since token are stored in kerenl paged memory.
Any ideas on how to troobleshoot this further is appreciated. right now i can only think of settings up a restart scheduel on sqlserver to aleviate the problem. but i need a long term solution which i am trying to work out.