Thanks for the info. JVM_RESTART won’t work for us because this a web application running in Tomcat using Ignite for a distributed cache. JVM_RESTART is documented as only working in a command line application.
The problem we experienced was bad since our load balancer didn’t realize one of the nodes was effectively dead because the health checks didn’t realize Ignite had stopped. The documentation for the SegmentationPolicy mentions that events are sent to a listener. If I understand correctly we need to call ignite.events().localListen() to listen for the shutdown and then cause our health detection to respond with a failure so we can restart the tomcat. Is this correct? Ralph > On Jan 18, 2018, at 7:15 AM, ilya.kasnacheev <ilya.kasnach...@gmail.com> > wrote: > > Hello! > > Maybe it's network problems and not full GC. As you can see from logs > there's failure to acknowledge a discovery message. > > As for "re-discovery" - as far as my understanding goes, that't not how > Ignite works. Client nodes will indeed try to re-discover server, but server > nodes consider themself "segmented". After that they *could* proceed working > solo, but by default they won't. This is governed by segmentation policy in > Ignite configuration. And by default it is "stop", i.e., shutdown node. > > https://ignite.apache.org/releases/latest/javadoc/org/apache/ignite/plugin/segmentation/SegmentationPolicy.html > > There's a possibility of RESTART_JVM and that's the one you should probably > set for the behaviour you desire. > > Regards, > > > > -- > Sent from: http://apache-ignite-users.70518.x6.nabble.com/ >
