Issue Addressed

N/A

Proposed Changes

On receiving unaggregated attestations from the network, lighthouse sets the should_process variable on attestations depending on if we have a connected validator having aggregation duties for the same slot .
For should_process == false, we only validate and forward the attestation, otherwise, we add it to the naive_aggregation_pool so that we can aggregate them.

The should_process calculation happens in the subnet service where it checks if the duty slot and subnet exists in a delay map.

The bug is that during insertion into the delay map, we use the default expiration delay of a single slot duration(12s).

Since we implemented #4806 to reduce subscription spam, we don't send subscriptions every slot before the duty, so we get a situation where an entry into the delay map gets evicted before it has been read.

Hence, the should_process_attestation function almost always returns false unless import-all-attestations flag is turned on, leading to Lighthouse producing bad quality aggregate attestations.