Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-22091

[release-4.12] High CPU usage in the ovs-vswitchd daemon

XMLWordPrintable

    • No
    • SDN Sprint 244, SDN Sprint 245, SDN Sprint 246
    • 3
    • Rejected
    • False
    • Hide

      None

      Show
      None
    • Service Delivery Architecture Overview

      Description of problem:

      ovs-vswitchd process on worker nodes service HCP components have very high CPU usage.

      Version-Release number of selected component (if applicable):

      4.12.

      How reproducible:

      This has only been seen on HyperShift worker nodes.

Root cause is believed to be Egress policies which define exceptions via CIDR ranges or port ranges, ie:
spec:
  podSelector:
    matchLabels:
      role: db
  policyTypes:
    - Ingress
    - Egress
  ingress:
    - from:
        - ipBlock:
            cidr: 172.17.0.0/16
            except:
              - 172.17.1.0/24
OR 

spec:
  podSelector:
    matchLabels:
      role: db
  policyTypes:
    - Egress
  egress:
    - to:
        - ipBlock:
            cidr: 10.0.0.0/24
      ports:
        - protocol: TCP
          port: 32000
          endPort: 32768

      Steps to Reproduce:

      1.
2.
3.

      Actual results:

       

      Expected results:

       

      Additional info:

      This was seen during an incident mentioned in the releated OHSS ticket

              sseethar Surya Seetharaman
              pebrown.openshift Pete Brown
              Anurag Saxena Anurag Saxena
              Votes:
              0 Vote for this issue
              Watchers:
              11 Start watching this issue

                Created:
                Updated:
                Resolved: