Open vincentbernat opened 6 years ago
This is a well known issue with stateful data and using time series storage (Kafka). Some folks are trying to use influx or other TSDB to store BGP data, but they of course have the same problem when data is expired because it wasn’t refreshed.
Recommendation is definitely not to restart the collector, router, or peers. That would directly effect the network.
The recommendation is to have a fault tolerant and correctly insync consumer that is always running. This consumer then can sync other consumers with the rib table for which ever peers are of interest to that consumer. The new consumer can get the offset from the root consumer if it wants to consume live updates after sync.
The MySQL consumer can be used for this.
Hey!
The documentation doesn't really deal with how expiration should be handled in Kafka. For example, if all topics are set to expire in 24 hours, new consumers don't have most information to be able to bootstrap. A simple solution is to restart the collector every 24 hour. Maybe a openbmpd could force a rolling disconnect instead? Or maybe I am missing another solution?