Reputation: 1114
We are using Hazelcast IMDG as an in memory grid. The number of nodes in our cluster is three, and we have one sync backup and the cluster is partition aware. In that case, I expect the distributed map will be distributed across 3 nodes (more or less) homogeneously. In case of a node break down, the leadership should be transferred to a healthy node(which has the sync backup for the lost data). If there is a write request to this newly assigned leader node, the same partition should be replicated synchronously to one of the alive nodes. Does it mean that in case of node failure, approximately one third of the distributed map should be replicated and during the replication time, all reads are blocked? Availability is hit if one of three node is down in case of one sync backup till approximately one third of the distribution is restored?
Upvotes: 1
Views: 697
Reputation: 591
If a node goes down, the cluster will promote the backup partitions to primary. And the migrations will start to create backups of these new primary partitions. Please check the Data Partitioning section.
During migrations, read operations are not blocked. Only write operations are blocked on the partition that is actively migrating. Since the partitions are migrated one by one, the effect on availability is minimal.
Upvotes: 8