Uploaded image for project: 'OpenShift Request For Enhancement'
  1. OpenShift Request For Enhancement
  2. RFE-4975

"ICMP fragmentation needed" to backend pods when landed to a random node

XMLWordPrintable

    • Icon: Feature Request Feature Request
    • Resolution: Unresolved
    • Icon: Undefined Undefined
    • None
    • None
    • Node
    • False
    • None
    • False
    • Not Selected

      1. Proposed title of this feature request
        2. What is the nature and description of the request?
        3. Why does the customer need this? (List the business requirements here)
        4. List any affected packages or components.

       

      There was originally a bug https://issues.redhat.com/browse/OCPBUGS-9079 which is not resolved in 4.14.3. 

      Metallb lb svc with externalTrafficPolicy: Local, with multiple host interfaces on the nodes and RoutingViaHost=true.

      I have three backend pods p0,p1,p2 in three nodes m0,m1,m2, the original syn packet (src: green client to lb ip) is being sent by the gateway to m2 node reaching p2 pod. The pod established connection and start sending jumpo frames traffic. When traffic reaches the gateway(router) an icmp packet generated because upstream interface has 1500B MTU and is being sent (from gateway ip to lb ip) to m3 node (depends on the ECMP hash policy but we can assume is not the correct).

      That icmp packet does not arrive in the p0 pod and connections halts for ever.

       

        1. image.png
          184 kB
          Konstantinos Karampogias

              gausingh@redhat.com Gaurav Singh
              kkarampo@redhat.com Konstantinos Karampogias
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated: