Number of connections increases rapidly.
After upgrading CentOS (RedHat) OpenSSL to version 0.9.8e-12.el5_4.6.i686 (backported) my relay stopped working (see weeek 13 in attached month graph) with no signs of problem in the log.
Then after help from #tor, one authority server administrator saw a problem in my connection attempts. I downgraded my OpenSSL to version 0.9.8e-12.el5_4.1.i686. As a result I got the descriptor published and it seemed to work ok.
Since then the connections has grown and grown to the limits of the relays virtual server capacity leaving it unresponsive (see week 14). Then after restart it regained a too large number of connections very rapid (see attached day graph).
After consulting #tor again (thanks for great help and kindness!) I become aware of that I have lost the stable flag and should try to shut down the relay for a day and then restart configured with 'MaxAdvertisedBandwith to 20KBytes' for a while to regain the stable-flag.
(Tor-relay version 0.2.1.25 (torproject rpm) on Cent-OS 5 in OpenVZ virtual container. Configured bandwitdth 100KB and 200KB burst.)
I will report back if or if not this solves the problem.
[Automatically added by flyspray2trac: Operating System: Redhat/CentOS Linux]