Details
-
Bug
-
Status: Stalled (View Workflow)
-
Trivial
-
Resolution: Unresolved
-
1.2.0
-
None
-
2021-4, 2021-5, 2021-6, 2021-7, 2021-8, 2021-9, 2021-10, 2021-11, 2021-12, 2021-13, 2021-14, 2021-15, 2021-16, 2021-17, 2022-22, 2022-23, 2023-4, 2023-5, 2023-6, 2023-7, 2023-8, 2023-10, 2023-11
Description
It has been proven by multiple experiments that Columnstore does not scale with increasing concurrency, when queries with aggregates are involved. Every additional query stream adds average latency to all other sessions regardless of the available resources and secret columnstore settings.
Total CPU consumed by PrimProcs is likewise not linear (2 concurrent sessions spend more than 2x total CPU in PrimProc as compared to just one), indicating some sort of spin lock.
The issue was escalated from a prospect (Quinstreet) who discovered it.
Attachments
Issue Links
- causes
-
MCOL-4679 Simplify configuration of EM to PM connections
- Closed
- includes
-
MCOL-4692 Remove global mutex lock in OAMCache facility.
- Closed
-
MCOL-4694 Distribute primitive jobs messages b/w TCP connections provided by DEC
- Closed
- is blocked by
-
MCOL-5043 Reduce a number of pre-spawned ExeMgr threads
- Stalled
-
MCOL-5044 Improve PP thread pool with a fair scheduler
- Closed
- relates to
-
MCOL-4691 Major Regression: Selects with aggregates 2x slower in 5.x than in 1.2 (due to collation support)
- Closed
-
MCOL-4770 Distribute queries processing b/w ExeMgrs/primprocs in the cluster.
- Closed
-
MCOL-4849 Optimize ExeMgr to reduce a number of context switches
- Closed