In several places in gossip code, the entire crds table is scanned only to filter out nodes' contact infos. Currently on mainnet, crds table is of size ~70k, while there are only ~470 nodes. So the full table scan is inefficient. Instead we may maintain an index of only nodes' contact infos.