On 18/Jun/16 13:10, Randy Bush wrote:
i remembered wrongly
RFC6810
A client SHOULD delete the data from a cache when it has been unable to refresh from that cache for a configurable timer value. The default for that value is twice the polling period for that cache.
I suppose that is alright since, in a redundant scenario, the data from the remaining cache that (hopefully) still has a live RTR session will continue to be valid. In single cache scenarios, waiting for some time after the cache has disappeared is akin to standard BGP session keepalive protocols. However, several vendors have implemented protocol enhancements to immediately drop BGP sessions that have failed, rather than wait for the Hold timer to expire. I see value in that, and perhaps it might make sense for an RPKI implementation to support the same where it is more important for the RPKI data to be as current as possible. Mark.