Terracotta Discussion Forums (LEGACY READ-ONLY ARCHIVE)

Dear all,

I'm learning ehcache library recently and embedded it with some testing java codes. In my testing network environment, I only can use unicast instead of multicast. Hence, I used and configured the ManualRMICacheManagerPeerProvider. The replication of ehcache worked fine for all caches.

I have conducted some further testing of ehcache on its stability by plugging out the connection cable between two replicated servers. The data would remain synchronized after I put back the cable.

However, I noticed one difficulty of ehcache when the connection was lost between two servers before I start my testing program in one of the servers. In this scenario, my testing program would take very long time to start up. Through tracing logs, I found each cache would cost around 19 to 25 seconds to detect RemoteException when tried to bootstrap.

I have reviews the source code of ehcache, under the manual mode, ehcache didn't follow the formula MulticastKeepaliveHeartbeatSender.getHeartBeatInterval() * 2 + SHORT_DELAY like what multicast does to wait cache to join cluster. Instead, Its getTimeForClusterToForm() time is 0. When the manual mode tried to listRemoteCachePeers() during bootstrap, it actually called lookupRemoteCachePeer() to look up RMI URLs one by one. Thus, in order to speed up the timeout of bootstrap, I have followed Oracle's document to reduce sun.rmi.transport.tcp.handshakeTimeout (Default is 15 seconds) and sun.rmi.transport.tcp.handshakeTimeout (Default is 1 minutes) two system properties. What a pity, it didn't work and affect at all to echache behavior.

I'd like to ask any expert here, is there any other solution to achieve better timeout timing in my scenario ? Thank you for your time and reading ! Cheers !