Details
-
Sub-task
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
Followup from HIVE-19975.
W.r.t. write IDs and txn IDs, stats validity check currently verifies one of two things - that stats write ID is valid for query write ID list, or that stats txn ID (derived from write ID) is the same as the query txn ID.
I'm not sure the latter check is needed; removing it would allow us to make a bunch of APIs a little bit simpler.
ekoifman do you have any feedback? Can any stats reader (e.g. compile) observe stats written by the same txn; but in such manner that it doesn't have the write ID of the same-txn stats writer, in its valid write ID list? I'm assuming it's not possible, e.g. in multi statement txn each query would have the previous same-txn writer for the same table in its valid write ID list?