Details
-
Bug
-
Resolution: Won't Fix
-
Major - P3
-
None
-
None
-
None
-
None
-
ALL
-
Storage - Ra 2020-09-07, Storage - Ra 2020-09-21
-
25
-
0
Description
As I was diagnosing a test failure, I noticed this sequence of events:
- The body of execCommandDatabase throws a SnapshotTooOld error due to trying to read earlier than the oldest.
- When this happens, we try to increase the target snapshot window size to make it less likely for this to happen for subsequent operations.
- We expect this window size to be 0 for in-memory configurations so we trip up an invariant later down the track.
The way I understand it, this scenario makes sense and is ok. The only problem is the way we're handling the exception by widening the snapshot window. We shouldn't do this if we're running an in-memory configuration and haven't enabled test commands.