APIs, Messaging Queues, Files and Event Streaming

Application programming interfaces, known as APIs, are the standard way for one system to ask another system to do something and receive an immediate answer. A channel system calls a payment API with a JSON request, and the API returns a confirmation or a rejection within seconds. The pain.001 is a different thing: it is an XML payment initiation message, normally carried as a file or as a message between systems, not as an API payload. A payment API accepts a JSON request that describes the payment, while the pain.001 XML is the format the instruction takes when it travels as a message or a file. REST APIs use standard web protocols and JSON messages, which makes them easy to build and test. Good APIs are versioned, so that version one continues to work while version two adds new fields, and they are documented with examples, error codes, and limits. A bank that publishes an API without versioning forces every consumer to change whenever the API changes, which quickly becomes unmanageable. Messaging queues provide reliable delivery when the sender and the receiver do not need to be available at the same time. A payment hub places a pacs.008 message on a queue, and the receiving system collects it when it is ready, even if it was down for an hour. Queues guarantee that a message is delivered at least once, which means a receiver must handle duplicate deliveries. Ordering is not guaranteed by a queue in general. Only a FIFO or ordered queue, or a single ordered partition in an event stream, preserves order, and that ordering guarantee is what a designer relies on when a debit must be processed before a credit. A standard queue can deliver messages out of order, so a system that needs ordering must choose an ordered queue or partition, not assume one. The cost is complexity, because the consumer must handle duplicate deliveries and must acknowledge each message after processing it. Queues are ideal for high-volume flows, such as the 20,000 direct debits that arrive in a morning batch. Files remain the right choice for bulk transfers, because they are simple, auditable, and easy to handle in large volumes. A corporate customer sends a file with 5,000 supplier payments totalling EUR 1.2 million, and the bank processes it as a single unit with a single control total. Files are also the standard format for scheme settlement reports and for regulatory submissions, where a complete and signed record matters more than speed. The weakness of files is latency, because nothing is processed until the file arrives, and error handling must be designed per record so that one bad payment does not reject 4,999 good ones. Modern file integration adds a checksum and a record count to catch corruption in transit. Event streaming treats every change as an event that any interested system can consume. When a payment status changes from pending to settled, the hub publishes an event, and the notification service, the data warehouse, and the customer app all receive it at the same time. This decouples the systems, because the hub does not need to know who is listening, and new consumers can be added without changing the hub. Event streaming is powerful for real-time reporting and for customer notifications, but it needs careful design around ordering, replay, and retention. An event that arrives out of order can show a payment as settled before it shows it as sent, which confuses both customers and auditors. Most banks use all four styles together, and the skill lies in choosing the right one for each connection. A real-time customer payment uses an API for the request and event streaming for the status updates. A bulk salary file uses file transfer for the instructions and a messaging queue for the internal processing. A scheme connection uses a queue for guaranteed delivery and a file for the end-of-day settlement report. The integration architect documents each choice in an integration map that names the transport, the format, the frequency, and the owner for every connection. Without that map, the landscape becomes a tangle that nobody can change safely.