First of all, starting node in local JVM (process) is the fastest and most reliable way to access the cluster.
As far as behavior you see, it can be observed in two cases: 1. Node you start is a data node. So, when you stop the node data can be lost if no backups are configured for the cache (see org.apache.ignite.configuration.CacheConfiguration#setBackups). If you start node for short period of time, no point of making it a data node. You can start it as a client, so it will not carry data. Just call: Ignition.setClientMode(true); Ignition.start(..); 2. The second possible reason is that class definitions of the objects you put into cache are not available on remote nodes and when master node (node initiating put and holding the definitions) exits all other nodes have to cleanup caches. To fix this you can make your classes available on all the nodes - just build a jar and put it to classpath of each jvm involved. Please also turn off peer deployment which is recommended for performance - org.apache.ignite.configuration.IgniteConfiguration#setPeerClassLoadingEnabled -- View this message in context: http://apache-ignite-users.70518.x6.nabble.com/Spring-issue-with-Connect-to-Ignite-Server-Runing-on-Diff-Server-tp366p369.html Sent from the Apache Ignite Users mailing list archive at Nabble.com.
