The Dual-Write Problem: Why Two-Step Operations Risk Incomplete Data States

The dual-write problem arises when a single business operation requires two separate writes — such as debiting one bank account and crediting another — which can each become permanent independently. If a system crashes or fails between the two writes, data is left in an inconsistent intermediate state that violates expected rules, known as invariants. While database transactions solve this internally through atomicity and isolation, the guarantee breaks down the moment a second external system — like another bank's database — is involved. Techniques such as write-ahead logging help databases recover their own internal writes, but they cannot roll back changes already committed in a separate system. The problem, studied since the 1970s, remains a core challenge in distributed systems wherever money transfers, bookings, or cross-service messaging are involved.