vpp-agent strange behaviour on pod killed about vpp-agent HOT 2 CLOSED

Hi, I would need to understand the differences between the case 1 and the cases 2,3. I see that in 2, 3 the (supposedly) killed pod has the configuration first removed and then re-created with the new namespace reference. In the case 1, however, the obsolete configuration is not deleted first - could that step be added? Otherwise, the only reliable solution to fix practically any (outstanding) error caused by external forces (e.g. pod kill/crash) is to run resync - downstream resync would be sufficient in this case (not exposed via GRPC, but could be added). Without resync or the delete, the agent will not find out that the VPP-side of the TAP has disappeared and will continue assuming that it still exists (since it didn't get any errors from its CREATE). Another solution could be to add dependency link between the VPP-side of TAP and the namespace of the Linux side, but for that you would need to label your pods with microservice labels (add env. variable MICROSERVICE_LABEL=pod_ID).

from vpp-agent.

@milanlenco thanks for the comment. Resync is helped 👍

from vpp-agent.