Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-27821

pods that are HostNetworked on nodes using routingViaHost:true ipForwarding: global cannot route to default kubernetes service IP

XMLWordPrintable

    • Important
    • No
    • SDN Sprint 250, SDN Sprint 254, SDN Sprint 255
    • 3
    • Rejected
    • False
    • Hide

      None

      Show
      None
    • Requires 2nd def gateway with routingViaHost=true

      Description of problem:

      • We're seeing that on two baremetal nodes where `routingViaHost=true` is enabled (with ipForwarding set properly as Global) the following problem:
      • They have set NodeIP Hint to force OVN to bind to a secondary interface at `bond1.2039 `
      • we're seeing that specific pods that are hostNetworked can't reach the default kubernetes service IP address; and are failing to initialize as a result (CLBO).

      ~~~
      F0120 03:20:42.221327 879146 driver.go:131] failed to get node "wb02.pdns-edtnabtf-arch01.nfvdev.tlabs.ca" information: Get "https://192.168.0.1:443/api/v1/nodes/wb02.pdns-edtnabtf-arch01.nfvdev.tlabs.ca": dial tcp 192.168.0.1:443: i/o timeout
      ~~~

      other pods on affected node with above config can hit the target service however, pods that are hostNetworked appear to be failing:

      $ oc get pod csi-rbdplugin-kpz7n -o yaml | grep hostNetwork
      hostNetwork: true

       

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

      4.14

      • bare-metal  

      How reproducible:

      • new cluster, every time

      Steps to Reproduce:

      We have redeployed the cluster. and have
      routingViaHost and ipForwarding both enabled.

      We also pushed out a NODEIP_HINT configuraiton to all the nodes to make sure SDN is overlayed on the correct interface.

      Default gateway has been moved to bond1.2039on the 2 x baremetal worker nodes.

      wb01

      wb02

      observe that hostNetworked pods crashloop backoff
       

      Actual results:

      • hostnetworked pods cannot call the default kube service address  

      Expected results:

      • hostnetworked pods should be able to do so.
         

      Additional info:

      See the first comment for data samples + must-gathers + sosreports

            trozet@redhat.com Tim Rozet
            rhn-support-wrussell Will Russell
            Ross Brattain Ross Brattain
            Votes:
            1 Vote for this issue
            Watchers:
            22 Start watching this issue

              Created:
              Updated:
              Resolved: