We have some Solaris 10 clients ldapcliented to a Directory Proxy Server. After 15 minutes, the Solaris name service cache will fail to communicate to the proxy instance and the proxy instance's readconnectionsrefused attribute will start incrementing.
At first it seemed we would need to increase the worker-threads and num-bind-limit, but those do not fix the problem.
At the same time the name-service-cache starts failing, I am still able to query and search the proxy. I have set up a Jmeter test which continues to run and they never fail.
It seems very consistent that the problem with the name-service-cache occurs every 15 minutes and I am able to reproduce this at the client's site and in my lab. Restarting either the proxy or the name-service-cache clears the problem.
I may worth having a look at the proxy access log:
The problem might be due to inactive connections dropped by the proxy (connection idle timeout) or other connection-related problems
My 2 cents
Thanks for the suggestion. It is odd that if I client directly to the Directory Server and not the Directory Proxy, then the name-service-cache problems don't exist. The Proxy and Directory Server are on the same server.
The fact readconnectionsrefused increases seems to indicate that the problem is between the proxy and the backend server.
For some reasons, DPS cannot grab a valid connection to forward the traffic to the directory server.
DPS uses connection pooling, so it may be negatively impacted when the directory server is configured to drop idle connections after a while.
Could you please check whether the directory server is configured to drop inactive connection after 15mn ? If this is the case, this would explain the issue and then we could
devise the best solution to address it.
See idle-timeout property in http://docs.oracle.com/cd/E19424-01/820-4813/idle-timeout-5dsconf/index.html