In a topology consisting of a router network hosted outside of OpenShift (two pairs of client-facing routers, linked to two pairs of outward-facing routers) and a router pair hosted in OpenShift, we had a recent incident where the connection was lost between the outward-facing on-prem routers and the hosted routers (Link to Neighbor Router Lost). Most of the links were immediately reestablished, but one link failed to reestablish between the on-prem and hosted routers. We can see that the link was reattached on the client-facing and outward-facing on-prem routers, but the link never got recreated on the hosted router until the application was restarted 3 hours later.

From the analysis following the incident, this appears to have been cause by a relatively unique set of circumstances in which a primary connection failure that resulted in the initial link losses was followed by a secondary connection failure in the network, resulting in a race condition while reestablishing / attaching the links.