I checked out AMQ-1739. I think my situation is different. My stomp sockets all show as ESTABLISHED, not CLOSE_WAIT. The interesting thing is that the message producer clients don't show the sockets as existing anymore. I think what may be happening is that an intermediary firewall drops the connection from its table after an inactivity timeout. Later, a clients try to send a message and get a failure and closes the original socket, opening another. On the activemq side, no traffic arrives again on the original socket and it leaks. This happens about 25 times a day until I hit the ulimit is hit and ActiveMQ won't accept new connections. (BTW, it seems a better behavior when all sockets are full would be to close sockets from an LRU list.)
What's strange about this, is my linux kernel should be closing quiet connections: # sysctl -a | grep keepalive net.ipv4.tcp_keepalive_intvl = 75 net.ipv4.tcp_keepalive_probes = 9 net.ipv4.tcp_keepalive_time = 900 Which makes me think ActiveMQ somehow forces them to be kept open, as if maxInactivityDuration is unlimited. rajdavies wrote: > > Its worth looking at https://issues.apache.org/activemq/browse/AMQ-1739 > > In particular the last comment from vik dhawan > > cheers, > > Rob > -- View this message in context: http://www.nabble.com/Setting-stomp-wire-format-parameters-tp17661413p17843902.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
