Field breakdowns: high-load PHP, architecture and trade-offs, reliability, teams. One fact = one number — everything here is already proven in production projects.
Four problems of a live reverse auction — realtime, races, money, trust — and how they are solved in an open-source project: a state machine with 16 statuses, a bid transaction under a pessimistic lock with a single flush, push to clients via Mercure SSE, crash recovery. Code, numbers, and SLOs from load tests.
Why a repeated request is not a repeated operation, and how the exactly-once contract works in a real payment core: a two-phase hold → capture payment, retries by rules, balancing across gateways, recurring payments, observability, and reconciliations. The РОСШТРАФЫ payments case: −25% erroneous operations, +~10% conversion.
Why synchronous document generation inside an HTTP request breaks UX, infrastructure, and client business processes — and how a hybrid scheme on RabbitMQ solves it: where to draw the boundary, how a task queue with statuses works, what to do about load, integrations, and 15 async pitfalls.