Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
We need to move performance measurements to the next level and implement whatever is needed to make us able to find easier answers to problems like “query is slow”. The problem is that we keep digging into logs + watching metrics that are provided by something in the environment (that can be anything that the actual vendor implements outside of hive).
Let's try to localize the environment problems to the interval of the slow query and make it exposed through counters.
Also, let's keep in mind that performance measurements ideally should never cause performance problems itself: heavyweight measurements should be disabled by default.
Attachments
Issue Links
- is related to
-
-
- Open
-
1.
|
Counter for query concurrency |
|
Resolved | László Bodor |
2.
|
Counters for metastore usage |
|
Open | Unassigned |
3.
|
FileSystem counters summary to include durations |
|
Open | Unassigned |
4.
|
LLAP perf counter: all nodes, used nodes |
|
Resolved | László Bodor |