Some platform services have excessive number of worker processes
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
StarlingX |
Fix Released
|
Low
|
Tee Ngo |
Bug Description
Brief Description
-----------------
Currently most platform services scale their worker processes based on the number of platform cores in standard configuration. However, this number should not exceed 10. It's observed in a lab with 36 core controller, there are 30 keystone-public worker processes, 18 dashboard worker processes, 8 barbican worker processes.
Now that OpenStack is no longer part of the platform, the number of worker processes should follow the following general rule:
- Frequently used services: number of platform cores/2, min=2, max=10 (e.g. keystone)
- Moderately used services: number of platform cores/5, min=2, max=8 (e.g. sysinv, horizon. vim)
- Rarely used service: number of cores/1, min=1, max=3 (e.g. barbican)
Furthermore, in All-in-one configuration, the number of worker processes is currently limited to 3 max. Since the number of workers that can be added in AIODX is not limited, this restriction must be removed and the above rule must also be applied to this configuration as well.
Severity
--------
Major
Steps to Reproduce
------------------
Bring up a standard system with high core controllers
Expected Behavior
------------------
Number of worker processes of platform services should not exceed 10.
Actual Behavior
----------------
Some services have excessive number of worker processes taking up a considerable amount of platform memory and contributing to large cpu spike during swact.
Reproducibility
---------------
100% reproducible
System Configuration
-------
Standard system
Branch/Pull Time/Commit
-------
Latest
Last Pass
---------
N/A
Timestamp/Logs
--------------
N/A
Test Activity
-------------
Evaluation
Changed in starlingx: | |
status: | Triaged → Won't Fix |
stx.3.0 / medium priority - this is a tuning exercise