After #12 it is possible to have a highly-available system that will automatically rebalance partitions if one node goes down.
However this is a new experimental feature that is disabled by default. In order to have it as a solid new feature:
Some proper tests on a good cluster need to be done to determine, for example, whether some kind of throttling may be needed or not.
Some kind of metric needs to be taken into account not to make a DNode run out of disk space. For example, making the DNodes report their available disk space to Hazelcast once every few minutes (that could probably be enough). Then, the QNode could decide not to rebalance if the destination DNode has few disk space available.
After #12 it is possible to have a highly-available system that will automatically rebalance partitions if one node goes down.
However this is a new experimental feature that is disabled by default. In order to have it as a solid new feature: