Java and backend interview questions

Redis pub/sub is fire-and-forget: if a subscriber is down, the message is gone.

A topic is a named stream, like orders.

Each partition is assigned to at most one consumer in a group, so the group shares work.

At-most-once commits the offset before processing, so a crash can skip messages.

Kafka keeps a durable, replayable log so a new billing service can catch up from old offsets.

The key hashes to a partition, so all events for that user land in one ordered log.

Opening a TCP connection and authenticating to Postgres on every HTTP call is slow and will exhaust database max_connections.

HikariCP is the default JDBC pool.

Each request thread that hits the database needs a connection at the same time.

You load N parents with one query, then lazy-load each collection with one extra query, for N+1 round trips.

Use JOIN FETCH in a dedicated query, an @EntityGraph on that use case, or a DTO projection that joins in SQL.

Error handlers have four arguments (err, req, res, next).

The cluster module or a process manager like PM2 forks workers that share the port.

Read the Authorization Bearer token, verify signature and expiry with the IdP public key or a shared secret, then attach req.user.

Key by IP plus username, count hits in Redis with an INCR and TTL window, and return 429 when the cap is exceeded.

If one team owns the whole domain and traffic is moderate, a modular monolith deploys faster and avoids distributed transactions.

It is a single entry point for auth, TLS, rate limits, and routing to many backend services.

Instances come and go with autoscaling, so callers cannot hardcode IPs.

Use synchronous REST when the user is waiting and you need an immediate consistent answer, like checking wallet balance.

After a dependency fails past a threshold, the breaker opens and fails fast instead of waiting on timeouts.