Process-local replay protection fails once a second replica is added
A signed request with a nonce is verified and its nonce recorded in one replica's local memory, so the same request can be accepted again if it is routed to a second replica. Scaling a Deployment or Service does not synchronize that security state, and routing affinity only lowers the odds of a replay without enforcing it. Preventing this requires a shared or safely replicated authority that atomically checks and records each nonce for the full replay window, scoped by tenant or source. The author also recommends deciding in advance how the service should behave when that authority is unavailable, with a retryable refusal often safer than skipping the check. Testing should cover replays across different replicas, replica replacement, scaling, and authority outages, not just Pod restarts.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in