Could you provide some information on split-brain scenario, please?
1. What network problems can cause split-brain? How fast does segmentation happen?
2. Can node segmentation be prevented? Can reconnectCount and failureDetectionTimeout in TcpDiscoverySpi be used to defer segmentation?
3. Is there an alternative to node restart to make a segmented node come back to the original cluster and for rebalancing to happen?
4. What topology does ignite.cluster() return when called on a segmented node? Does segmentation result in disjoint clusters or will there only be groups of nodes inside the original cluster that cannot see the whole?
5. What would be a safe criterion to check in a SegmentationResolver? If a TcpDiscoverySharedFsIpFinder is used would it then be sufficient to check if a node cannot see the shared network directory to recognize it as segmented?
6. Why is it recommended to run TopologyValidator and SegmentationResolver in parallel? Will segmented or non-segmented nodes be passed to the TopologyValidator method?
