clean up orphaned ExecutionStage instances in the ArrowFlightEndpoint #90
Description
Right now ExecutionStages are held by stage_key which is (query_id, stage_num, task_num)
They can be discarded if:
- We have executed all partitions in it and do not need it anymore (do_get: use TTL map to store task state #108)
- It gets orphaned because a node higher up in the plan died and never requested the data. Having a timeout clean this up is appropriate (Create TTL map with time wheel architecture #96)
Some prior art here: #70