The short answer
Ask RabbitMQ candidates about the changes that break real clusters: why classic mirrored queues must be migrated before 4.0, why a queue's type cannot be changed in place, what Khepri changes when a majority of nodes is lost, and how publisher confirms and consumer acknowledgements prove no messages were lost.
Hiring for this?
Conntal places RabbitMQ engineers on contract, contract-to-hire or direct hire, each screened by a practitioner before you see a CV.
AceMQ supports RabbitMQ for regulated enterprises, and these are the questions its practitioners use to decide whether a candidate has run RabbitMQ or only used it.
1. We are on 3.13 with classic mirrored queues. What happens at 4.0?
Classic mirrored queues were deprecated in 3.9 and removed in 4.0 [1]. A candidate who plans an in-place upgrade has not done this migration.
2. Can we just change those queues to quorum queues?
No: because the two types differ down to storage, an existing classic queue cannot be turned into a quorum queue in place [1]. Strong candidates describe a blue-green move to a new cluster, using federation or shovels to carry messages across [1].
3. What must be true before any rolling upgrade?
Listen for feature flags: RabbitMQ's guidance is to enable all feature flags after a rolling upgrade, because they become mandatory in later versions [2]. A cluster with unenabled flags is not ready for its next upgrade.
4. What does Khepri change?
Khepri became the default metadata store in 4.2.0 and is based on Raft [3]. Metadata updates — and sometimes queries — need a quorum of nodes [4], so topology changes can time out on the minority side of a partition.
5. How do you prove no messages were lost in an incident?
The senior answer names publisher confirms on the producing side and consumer acknowledgements on the consuming side, and says what each one proves, rather than guessing from queue depth.
6. Tell me about a RabbitMQ incident you handled.
Listen for the version, the queue type and the first signal that something was wrong. Specifics separate operators from users.
Hiring for this?
Conntal places RabbitMQ engineers on contract, contract-to-hire or direct hire, each screened by a practitioner before you see a CV.
Sources
- [1] RabbitMQ, Migrating from Classic Mirrored Queues to Quorum Queues in 2025. Published 29 July 2025.Primary source
- [2] RabbitMQ (Broadcom), Feature flags. Retrieved 10 October 2026.Primary source
- [3] RabbitMQ (Broadcom), Metadata store. Retrieved 8 October 2026.Primary source
- [4] RabbitMQ (Broadcom), Clustering and Khepri. Retrieved 8 October 2026.Primary source
Go deeper with the engineers who run it
Conntal is the staffing arm of AceMQ. When you need the platform supported rather than a person placed, these are the pages to read.
Related questions
What are good RabbitMQ interview questions?
Ask about migrating classic mirrored queues before 4.0, why queue types cannot be changed in place, feature flags before upgrades, Khepri's quorum requirement, and proving message safety with publisher confirms and consumer acknowledgements.
How can I test RabbitMQ experience without a RabbitMQ expert on the panel?
Use scenario questions with known right answers, like the ones in this guide, or have the screen run by a practitioner who supports RabbitMQ in production.