When shutting down we currently remove routes from the kernel FIB without waiting for our final hellos to propagate through the network. This could cause us to either form loops or blackhole traffic using whichever external routes remain.
Instead let the operator define a suitable delay for their network to ensure complete propagation of retractions and divert traffic before doing anything drastic.
jech
commented
1 year ago
When shutting down we currently remove routes from the kernel FIB without waiting for our final hellos to propagate through the network. This could cause us to either form loops or blackhole traffic using whichever external routes remain.
Instead let the operator define a suitable delay for their network to ensure complete propagation of retractions and divert traffic before doing anything drastic.