Event-Driven & Messaging Testing

Async systems fail quietly: messages get lost, duplicated, reordered or poisoned, and the effect shows up minutes later somewhere else.

Properties to test

Tooling

docker compose up -d kafka schema-registry            # or Testcontainers
kcat -b localhost:9092 -t orders -P <<< '{"id":1}'     # produce
kcat -b localhost:9092 -t orders -C -o beginning -e    # consume
kafka-consumer-groups --bootstrap-server :9092 --describe --group billing   # lag
rabbitmqadmin list queues name messages_ready messages_unacknowledged
awslocal sqs receive-message --queue-url $Q --max-number-of-messages 10     # LocalStack

// Testcontainers Kafka in JUnit 5
@Container static KafkaContainer k = new KafkaContainer(DockerImageName.parse("confluentinc/cp-kafka:7.6.0"));

# wait for effect — never sleep
from tenacity import retry, stop_after_delay, wait_fixed
@retry(stop=stop_after_delay(10), wait=wait_fixed(0.2))
def order_is_shipped(): assert db.get(order_id).status == "SHIPPED"

Contract & schema

Register schemas; CI check curl -X POST .../compatibility/subjects/orders-value/versions/latest; consumer-driven contracts for messages (Pact message pacts, contract-testing); never remove/rename fields, only add optional ones.

Webhooks

Signature verification (HMAC), replay window/timestamps, retries with 5xx, out-of-order & duplicate delivery, 2xx-fast + async processing, endpoint idempotency, delivery logs.

Failure injection

Broker restart · partition rebalance during processing · network partition (toxiproxy) · consumer crash after side-effect before ack · duplicate producers · clock skew — see chaos-resilience-testing.

Related

microservices-testing, contract-testing, integration-testing, observability-testing, concurrency-race-condition-testing