Details
-
Sub-system Version
-
Resolution: Resolved
-
None
-
Obs Mgt & Controls
-
Team_CREAM
-
Taranta
-
2.11.2
-
PATCH
-
2.11.2 Latest
-
Sprint 1
-
-
-
Description
Recent "fix" for nginx in taranta causes a major issue, in particular for the bang clusters. The taranta pod launches worker processes in the node where it is running, one per cpu available. This has two effects: in our nodes with 128 cpus we get 128 processes and when nginx refreshes (every 30 minutes) we briefly get 256 processes and we get kernel warnings about oomkills, this will not show in techops or the itfs where there are less cpus hence less worker processes, (ii) during the nginx refresh (every 30 mn) each of the worker processes can get up to 100% of cpu, it's not long (~1 mn), but enough for pods that have cpu limits set to be evicted from that node.
Added
- <TBD>
Changed
- worker_processes 12; # Limit to 12 to fix the high resource usage
Deprecated
- None
Removed
- None
Fixed
- None
Test Evidence
- <TBD>
Security
- None