Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
Impala 3.1.0
-
None
-
ghx-label-3
Description
We're currently logging a lot of things per query on executors that is redundant or unnecessary, e.g.:
- ExecQueryFInstances() and StartQueryFInstances() are both logged but are redundant
- Some random low-level details are logged, e.g. "Max row batch queue size for scan node '0' in fragment instance '940cdcd50664a51:368ab6690000001e': 180"
I think we should cut back the logging at -v=1, which is the default log level most deployments run at:
- Do log key finstance lifecycle events that are useful to understand what finstances are running and if/when failures occurred
- Don't log things redundantly
- Don't log lower-level events that occur in the normal flow of query execution (e.g. memory limits, queue lengths). These things have limited diagnostic value and should be in the profile, if anywhere