Home › Blog › RabbitMQ Interview Questions (2026): 30 Questions With Answers, From Exchanges to Quorum Queues

RabbitMQ Interview Questions (2026): 30 Questions With Answers, From Exchanges to Quorum Queues

RabbitMQ interview questions test whether you can trace a message from a publisher through an exchange and its bindings into a queue, and explain what stops it being lost on the way: acknowledgements, publisher confirms, durable queues and replication. Answer for the 4.x world.

RabbitMQ interview questions test whether you can trace a message from a publisher through an exchange and its bindings into a queue, and explain what stops it being lost on the way: acknowledgements, publisher confirms, durable queues and replication. Answer for the 4.x world. As of November 17, 2026, rabbitmq.com lists 4.4 as the current release series (4.4.1, released November 16, 2026), classic queue mirroring has been gone since 4.0, and Khepri is the only supported metadata store from 4.3.

Below are 30 questions by level with concise answers, five troubleshooting scenarios and a table of how interviews test this. For the log-based alternative, see Kafka interview questions; for the architecture around both, see microservices interview questions.

RabbitMQ interview questions: fundamentals (1–10)

  1. What is RabbitMQ and when would you use it? A message broker: publishers send messages, the broker routes and stores them in queues, and consumers receive them. Use it to decouple services, absorb load spikes, spread background jobs across workers and route messages flexibly. It supports AMQP 0-9-1 and, natively since 4.0, AMQP 1.0.
  2. Explain publisher, exchange, binding, queue and consumer. Publishers never write to a queue directly. They publish to an exchange with a routing key; bindings are the rules that connect an exchange to queues; queues hold messages; consumers subscribe to queues and acknowledge what they finish.
  3. What exchange types are there? direct (exact routing-key match), fanout (every bound queue, key ignored), topic (patterns on dot-separated words, * for one word and # for zero or more, as in orders.*.created) and headers (match on header values, with x-match set to all or any).
  4. What is the default exchange? A nameless direct exchange that every queue is bound to by its own name. Publishing to exchange "" with routing key invoices lands in queue invoices. Fine for demos; named exchanges let you change routing without touching publishers.
  5. What is the difference between a connection and a channel? A connection is one TCP (usually TLS) link; channels are lightweight sessions multiplexed over it. Keep one long-lived connection per process and use separate channels per thread. Opening a connection per message is a classic performance bug.
  6. What do consumer acknowledgements do? With manual acks the broker keeps a delivered message until the consumer sends basic.ack. If the channel closes first, the message is requeued and redelivered. Automatic acks are faster but lose whatever a crashing consumer had in memory.
  7. What is prefetch? basic.qos caps how many unacknowledged deliveries a consumer can hold. Without it the broker pushes as fast as it can and one consumer can hoard the backlog. A low prefetch gives fair dispatch for slow, uneven jobs; a higher one raises throughput for fast ones. Tune it by measuring.
  8. Durable queue versus persistent message? Durability is a queue property (the queue definition survives a restart); persistence is a per-message flag. You need both, plus publisher confirms, to survive a broker restart. Quorum queues are always durable.
  9. When is a message dead-lettered? The docs list four cases: a consumer rejects or nacks it with requeue=false, it expires by TTL, it is dropped because the queue exceeded a length limit, or it is returned to a quorum queue more times than the delivery limit. If a whole queue expires, its messages are not dead-lettered.
  10. How do you configure a dead-letter exchange? Preferably with a policy (dead-letter-exchange, optionally dead-letter-routing-key). The docs recommend policies over hardcoded x-arguments because policies can change at any time without redeploying, and arguments override policies when both exist.

Intermediate RabbitMQ questions (11–20)

  1. What are publisher confirms? The broker acknowledges a published message once it has taken responsibility for it, for example after a quorum queue has replicated it. Enable confirm mode on the channel, track outstanding messages and republish on a nack or timeout. Republishing can create duplicates, so consumers must be idempotent.
  2. What delivery guarantees does RabbitMQ give? At-most-once with automatic acks and no confirms; at-least-once with confirms, durable queues, persistent messages and manual acks after processing. There is no end-to-end exactly-once: you get the effect with idempotent consumers, a deduplication key or an inbox table.
  3. Classic queues, quorum queues or streams? Classic queues are non-replicated since 4.0 and suit transient or high-churn work. Quorum queues replicate with Raft and need a majority of members available; they are the safe default for important messages. Streams are replicated append-only logs that consumers read non-destructively from any offset, for replay and large fan-out.
  4. What replaced classic mirrored queues? Classic queue mirroring was removed in RabbitMQ 4.0. Quorum queues are the replicated queue type, with streams for log-style workloads. A 3.x upgrade therefore includes migrating mirrored queues, often by declaring quorum queues in a new virtual host and moving traffic across.
  5. How do quorum queues handle poison messages? They count failed deliveries (the x-delivery-count header) and, since 4.0, enforce a default delivery limit of 20. A message past the limit is dropped, or dead-lettered if a DLX is set, so always configure one. Setting x-delivery-limit=-1 disables the limit.
  6. Does RabbitMQ guarantee ordering? A queue is FIFO, but competing consumers, prefetch and redelivery after a nack can reorder processing. For per-key order, use a single active consumer, or partition by key across queues or a super stream so each key has one consumer.
  7. What is single active consumer? A queue setting (x-single-active-consumer) where only one registered consumer receives messages; the others wait and one takes over if it disconnects. It gives ordered processing with failover.
  8. How do priorities work? Classic queues need x-max-priority at declaration. Since 4.3, quorum queues always support strict priorities with 32 levels (0–31), and a message with no priority is treated as 4. Priority only reorders what has not yet been delivered within the prefetch window.
  9. What is the consumer delivery timeout? If a consumer holds an unacknowledged delivery longer than the timeout (30 minutes by default), the message is returned for redelivery. On AMQP 0-9-1 the consumer is cancelled if the client supports cancel notifications, otherwise the channel closes with PRECONDITION_FAILED. Since 4.3 the timeout applies only to quorum queues.
  10. How do you build retries with backoff? Nack failures to a dead-letter exchange that routes to a wait queue with a message TTL; the wait queue dead-letters back to the work queue when the TTL expires. Use one wait queue per delay tier, count attempts and park the message in an error queue after the last one.

Experienced and senior RabbitMQ questions (21–30)

  1. How does clustering work? Nodes share metadata (users, virtual hosts, exchanges, bindings, queue definitions). Khepri, a Raft-based store, is the only metadata store from 4.3, replacing Mnesia, so a majority of nodes must be up for metadata changes. Message data lives where the queue type puts it: one node for classic queues, replicas for quorum queues and streams. Clusters of 3 or 5 nodes are typical.
  2. What happens during a network partition? Raft-based components (quorum queues, streams, Khepri) keep working on the majority side; the minority side cannot elect leaders or accept writes for them. Clients on the minority should reconnect to another node, so give them several addresses or a load balancer. Older 3.x clusters on Mnesia used partition-handling modes such as pause_minority; know which one you run.
  3. What are memory and disk alarms? When a node crosses its memory high watermark or free disk drops below its limit, the broker blocks publishing connections across the cluster until it recovers, while consumers keep draining. Clients see connection.blocked. The fix is more consumers, smaller backlogs or more resources, not just raising the limit.
  4. RabbitMQ or Kafka? RabbitMQ is a smart broker: flexible routing, per-message acks, consumed messages removed, priorities, TTL and dead lettering, low-latency task distribution. Kafka is a partitioned log: consumers track offsets, retention is long and replay is native. Pick by consumption model; RabbitMQ streams cover some replay needs.
  5. How do you scale consumers past one queue? Competing consumers scale until the queue itself becomes the bottleneck, because each queue has one leader. Then shard work across several queues by key and spread their leaders across nodes, keeping prefetch tuned and handlers idempotent.
  6. How do quorum queue replicas work? One leader accepts publishes and deliveries and replicates to followers; an entry commits once a majority has it. With 3 replicas you tolerate one node down, with 5 you tolerate two. Spread leaders so one node does not carry every queue.
  7. How do you do request-reply over RabbitMQ? The client publishes with reply_to (a reply queue or the direct reply-to pseudo-queue amq.rabbitmq.reply-to) and a correlation_id; the server replies to that address; the client matches on the ID and applies a timeout. It couples services synchronously, so justify it over HTTP or gRPC.
  8. What happens to an unroutable message? If no binding matches, it is dropped by default. Publish with mandatory=true to have it returned to the publisher, or set an alternate exchange on the exchange so the broker catches it server-side.
  9. How do you make the whole path loss-proof? Quorum queue, persistent messages, publisher confirms with retry, mandatory or an alternate exchange, manual acks only after the work commits, idempotent handlers, a dead-letter exchange, and alerts on queue depth and unacked counts.
  10. What do you monitor in production? Ready and unacked message counts and their growth, publish, deliver and ack rates, redelivery rate, consumer count and utilisation, connection and channel churn, memory and disk alarms, file descriptors and quorum queue leader placement, usually through the Prometheus plugin.

RabbitMQ scenario and troubleshooting questions

A queue keeps growing during a sale. What do you check? Compare publish and ack rates first. Then look at consumer count, unacked per consumer (prefetch too high or too low), a slow dependency such as the database, and a high redelivery rate that suggests a poison message looping. Fixes: add consumers, tune prefetch, park the bad message, shard the queue.

Messages are vanishing. Where do they go? Usual suspects: unroutable messages with no mandatory flag or alternate exchange, automatic acks with a crashing consumer, TTL or length limits dropping messages, non-persistent messages lost in a restart, or a quorum queue delivery limit dropping messages because no DLX is set.

The downstream database has duplicate orders. Expected under at-least-once: a consumer crashed after writing but before acking, or a publisher retried after a missing confirm. Add an idempotency key with a unique constraint or an inbox table, and ack only after the transaction commits.

Publishers suddenly hang. Check for a memory or disk alarm and connection.blocked notifications. Drain the backlog with more consumers, find the queue that is growing, and only then revisit limits. Also check for publishers opening a connection per message.

A long job's channel closes with PRECONDITION_FAILED. The delivery passed the consumer timeout (30 minutes by default). Raise consumer_timeout for that workload, through configuration or the consumer-timeout policy key, or split the job so the message is acked quickly and progress is tracked elsewhere.

How RabbitMQ interviews test this

FormatWhat it usually looks likeHow to prepare
Technical Q&ARapid questions on exchange types, acks versus confirms, prefetch, durability and dead letteringExplain each in two sentences with one example; the fundamentals above
System designDesign order processing or notifications with retries; draw exchanges, queues and the DLXPractice one design end to end, including failure paths and idempotency
Debugging story"Tell me about a messaging incident": a growing queue, duplicates, a blocked publisherPrepare one real incident with numbers you can defend
Senior or platformCluster sizing, quorum queues versus streams, 3.x to 4.x migrationKnow the 4.0 and 4.3 changes and the Raft majority rule

Candidates commonly report RabbitMQ coming up inside backend and microservices loops rather than as a round of its own. Related ground is in backend engineer interview questions.

Where CoPilot Interview fits

CoPilot Interview is an AI interview assistant. Its free AI mock interview builds RabbitMQ and backend questions from your résumé and the job description, reads them aloud and scores each spoken answer out of 10, which is a fast way to find the answers above that you know but cannot yet explain cleanly under pressure. The browser demo has 145 practice questions in 10 tracks, scores typed answers out of 10 and needs no signup. The free plan gives 15 minutes a day of live help in sessions of up to 15 minutes, with AI mock interviews included. Paid plans are Standard at $24.99 a month and Pro at $39.99 a month, in the desktop app for Windows 10/11 and macOS 12 or later.

Use it to practice. Online assessments, proctored tests, take-home tasks with rules and AI-led rounds forbid outside help, so prepare for those instead of bringing a tool. For any live interview, follow your employer's rules.

Disclosure: CoPilot Interview is our product, so weigh what we say about it accordingly.

Frequently asked questions

What are the most common RabbitMQ interview questions?

Exchange types and routing keys, acknowledgements versus publisher confirms, prefetch, durable queues and persistent messages, dead-letter exchanges, quorum queues versus classic queues and streams, and RabbitMQ versus Kafka.

What is the difference between RabbitMQ and Kafka?

RabbitMQ routes messages to queues and deletes them once acknowledged, which suits task distribution and flexible routing. Kafka appends messages to a partitioned log that consumers read by offset, which suits high-throughput streaming and replay.

Which RabbitMQ version should I know for a 2026 interview?

The 4.x series. As of November 17, 2026, 4.4 is current. Know that 4.0 removed classic queue mirroring and added a default quorum delivery limit of 20, and that 4.3 made Khepri the only metadata store and added strict priorities to quorum queues.

Is RabbitMQ exactly-once?

No. With confirms, durable queues and manual acks it is at-least-once, so duplicates are possible. Make consumers idempotent to get an exactly-once effect.

How do I prepare for a RabbitMQ interview quickly?

Run a local broker with the management plugin, write a publisher with confirms and a consumer with manual acks and a DLX, kill the consumer mid-message, and explain out loud what happened and why.

Practice this interview before the real one

The AI mock interview builds questions from your resume and the job description, reads them aloud and scores each spoken answer out of 10. Included on the free plan.

Start a mock interview