Details
-
Sub-task
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
Impala 3.1.0
-
None
-
ghx-label-3
Description
Currently we respond with CANCELLED only when hitting EOS. It seems a bit more robust to always respond with CANCELLED whenever query execution has terminated. That way, if the cancel RPC from the coordinator to a backend fails, the backend will still cancel if it can send status back to the coordinator later on.
Without this fix, the query can hang and/or finstances can continue running (until the query is closed, at which point the response to this RPC will be an error).