Communication: Sync, Async, gRPC

Synchronous communication (REST, Feign, gRPC) blocks the caller until the response arrives; asynchronous communication (Kafka, RabbitMQ) lets the producer publish and move on, so you choose synchronous when the answer is needed now and asynchronous when the work can happen later.

Choose REST when you need an immediate answer from a specific service or serve an external client; choose Kafka when the producer should not wait, the event has many consumers, or you need buffering, replay, and decoupling.

REST is text-based synchronous HTTP for universal request-response; gRPC is binary HTTP/2 synchronous for fast, strongly typed internal calls; Kafka is an asynchronous durable event log for decoupled, high-throughput fan-out.

Fire the downstream calls in parallel with non-blocking WebClient, combine them with Mono.zip, and the client gets one aggregated response in the time of the slowest single call, not the sum.

Run both paths together with the old one authoritative, shadow-validate the Kafka path with idempotent consumers and feature flags, then ramp ownership gradually and decommission only after the event path proves equivalent.

The API Gateway pattern is the single synchronous entry point that routes, authenticates, rate-limits, and aggregates calls to services, supported by service discovery, load balancing, and circuit breakers.

One query to get the list, then one query per item, equals 1+N calls, and across the network that is a latency and failure multiplier, not just a database inefficiency.

A crash means the unacknowledged message is redelivered, so you survive it with manual ack plus retry, idempotent processing, a dead-letter queue, a transactional outbox, and a Saga for multi-step work.