TODO

Each statemachine owns a worker to handle all its tasks. To avoid resource contention (e.g. multiple statemachines start taking snapshots/recovering from snapshots simultaneously), we introduce a centralized task limiter to control the concurrent snapshot-related tasks. When we encounter a new snapshot task, either from user's request or from the periodic snapshot config, we check the task limiter and reject requests if the system is overloaded.

TODO