01The lookup every transaction makes
Before a card payment is routed, the gateway looks up its BIN: which issuer, which country, which scheme, which card type. Routing, fraud screening and reporting all depend on the answer. That lookup sits on the hot path, so anything that slows it slows every transaction.
BIN data changes constantly, and a full refresh is more than a million records. Loading that into the same table the live lookups read from is how a routine data update becomes an incident.
02Separate the write path from the read path
The design that works is to make sure the load and the lookup never contend. New data is ingested in chunks into a structure that live transactions do not read. It is validated there: record counts, range overlaps, required fields. Indexes are built there as well.
Only when the new set is complete and checked does the lookup switch to it, in a single quick step. Transactions see the old data, then the new data, and never a half-loaded table or a lock.
03The same idea, applied elsewhere
Bulk merchant exports have the same shape in reverse: a large read that must not disturb live writes. Those belong on queues, processed in the background and delivered when ready, so a merchant downloading a million rows never competes with a customer trying to pay.