Jose, Actually, there is no such node, that is responsible for all data, stored in the cluster. The data is distributed across all nodes instead, so each node is responsible for some particular piece. Cache data is split into partitions (1024 for a partitioned cache and 512 for replicated by default), and every node stores some primary and some backup partitions, if any. And backup partitions are synchronized with primary ones, when they are modified.
There are partitioned and replicated caches. The difference between them is that data of replicated caches is stored on every node entirely. But there are still some backup and some primary partitions on every node. You can think of replicated caches as of partitioned, but with backup factor equal to infinity. When a node leaves topology, data rebalancing happens, and some backup partitions become primary. To be able to survive failure of *n *nodes, you need to have backup factor to be at least *n*. Here is documentation on data sharding: https://apacheignite.readme.io/docs/primary-and-backup-copies Denis вт, 27 февр. 2018 г. в 21:58, joseheitor <[email protected]>: > Thanks, Denis. > > I read the docs and also followed this short video demo/tutorial: > > https://www.youtube.com/watch?v=FKS8A86h-VY > > It worked as described, so I then extended the exercise by adding and > additional (3rd) node, using the persistent-store configuration on all > nodes and activating the cluster before transacting...as per the docs. > > Again it all worked as per the original demo, but the persistence now also > worked. > > My concern and confusion is the following: > > Assuming we start 3 nodes... > > - if we bring down the first node (Primary?) - availability of the data is > lost ... even though there are another two active nodes in the cluster. > Doesn't the system automatically elect a new Primary if there are enough > active terminals to maintain a replicated backup? How then do we achieve > 'fault-tolerance' and 'high-availability'? > > - if we bring down all but the first node (Primary?) data access continues > to be available for review and manipulation. But surely this should now > fail because there is no active 'Secondary' node to backup or replicate any > changes to? Doesn't this expose a risk to data 'consistency', as there is > now no backup of the data changes? > > Or is there a way to configure things so that the system will behave as > expected (above)? > > Jose > ------------------------------ > Sent from the Apache Ignite Users mailing list archive > <http://apache-ignite-users.70518.x6.nabble.com/> at Nabble.com. >
