-
Bug
-
Resolution: Done
-
Major
-
None
-
4.14
-
Yes
-
SDN Sprint 237
-
1
-
Approved
-
False
-
Description of problem:
Total ovnkube-network-controller-manager CPU usage is 293 cores across the cluster (i.e cluster with 120 worker nodes and each worker node with 8 CPUs), effectively using 2 cores in each node This is observed while running cluster-density-v2 workload on OVN-IC 120 node environment with Northd Incremental Processing patches i.e OVN_IMAGE - quay.io/itssurya/dev-images:ic-scale-ovnk-v2 CNO image - quay.io/itssurya/dev-images:ic-cno-hack I have got high latency with northd single thread CNO image ( quay.io/itssurya/dev-images:ic-cno-hack-1thread ) as well OVN-IC without incremental processing is using 19 cores across the cluster. Slides for this testing https://docs.google.com/presentation/d/16NMDoF52gUb-MybbAtSs16_888nLbshQYCU1Zw3WuvE/edit#slide=id.g24a34730293_0_0 Version-Release number of selected component (if applicable): OCP 4.14 OVN_IMAGE - quay.io/itssurya/dev-images:ic-scale-ovnk-v2 CNO image - quay.io/itssurya/dev-images:ic-cno-hack
Version-Release number of selected component (if applicable):
OCP 4.14 OVN_IMAGE - quay.io/itssurya/dev-images:ic-scale-ovnk-v2 CNO image - quay.io/itssurya/dev-images:ic-cno-hack
How reproducible:
Steps to Reproduce:
1. Deploy 120 node OCP environment on AWS 2. Apply OVN builds 3. Run node-denisty (light) and cluster-density-v2 workloads
Actual results:
Expected results:
Total ovnkube-network-controller-manager CPU usage is 19 cores across the cluster on OVN-IC without Incremental Processing
Additional info:
- is related to
-
OCPBUGS-14285 [OCP 4.14] High p99 podready latency in OVN-IC with Incremental Processing
- Closed
- links to