[ejabberd] ssl_esock leaking file descriptors?

James Dyer jad at hungover.org
Wed Jan 5 16:52:26 MSK 2005


We're currently running ejabberd on a test server at the moment, and 
appear to have encountered a problem with the use of file descriptors in 
ssl_esock. I was wondering if anyone else has seen anything similar?

Running on Redhat Enterprise 3, the server is not currently loaded at all 
- it's only got about 6 users in it's database.

Ejabberd was rejecting connections on it's SSL port. Having experienced 
similar (in fact, identical) problems in the past with jabberd2, the first 
thing I checked was the number of file descriptors in use 
(/proc/[pid]/fd), which I found was 1024, which is the same as 
`ulimit -n`.

My gut instinct is that ejabberd is not closing connections properly; 
either when someone disconnects from the server, or else the connection is 
terminated improperly (like if someone just turns their PC off, without 
shutting down the jabber client).

Has anyone else seen any behaviour similar to this or, even better, has 
anyone come up with a fix for this sort of thing?



More information about the ejabberd mailing list