Udo, thanks for the hint. The property was missing indeed. I've put it into my gemfire.properties file and the cluster waits all nodes to start before proceed to any activity. Eugene
On Tue, May 3, 2016 at 4:28 PM, Udo Kohlmeyer <[email protected]> wrote: > Hi there Eugene, > > Can you check if the enable-network-partition-detection property is set, > as per the documentation. > Handling Network partitioning > <http://geode.docs.pivotal.io/docs/managing/network_partitioning/handling_network_partitioning.html> > > --Udo > > > On 4/05/2016 6:22 am, Eugene Strokin wrote: > > I'm testing my 10 nodes cluster under production load and with production > data. > I was using automated tool which created the nodes (VMs) configured > everything and restarted all of them. > Everything worked, I mean, I was getting the data I expected, but when > I've checked the stats I noticed that I'm running 10 one node clusters. My > nodes didn't see each other, they had a separate duplicated set of data on > each node. > I've stopped all the nodes, cleaned all logs/storage files, and restarted > the nodes again. > Now I had one cluster with 7 nodes and 3 nodes separate. > I've stopped the 3 nodes, cleaned them up, and started them up one by one, > they successfully joined the cluster. At the end I've got all 10 nodes > working as a single cluster. > But I'm afraid that if nodes would get restarted or network would have > some problems, I could end up with split cluster again. > I use API to start Cache with locators, and all locator's IPs are provided > in the config. From the documentation I had impression that Geode would > wait till N/2+1 nodes would start before forming the cluster, since the > number of locators is preset. But looks like it is not the case. > Or should I set some setting to force such behavior? > > Thank you, > Eugene > > >
