THIS IS A TEST INSTANCE. ALL YOUR CHANGES WILL BE LOST!!!!
...
- Its internal hashmap-like structure and behavior during checkpoint is pretty similar to the existing CopyOnWriteStateMap used by HashMapStateBackend.
- It supports concurrent checkpoints.
- Each key/value pair will be deep-copied at most once per checkpoint triggering.
- The time complexity of operations done in the synchronous checkpoint stage is O(number_of_hashmap_buckets) and it involves only shallow-copy. So the extra overhead should be lightweight (similar to HashMapStateBackend's sync state).
Analysis
Performance overhead for read/write operations
...