Poison pill messages cause deterministic partition processing failure

Specific message payload triggers heavy processing (database query, GC pause) causing consumer to exceed max.poll.interval.ms. Consumer leaves group, rejoins, CooperativeStickyAssignor assigns same partitions, consumer re-consumes same message and fails again. With static membership, this creates guaranteed loop where failing node reclaims the poison pill every time, preventing offset advancement on that partition.

Implement DeserializationExceptionHandler or custom Interceptor with local LRU cache tracking message IDs or hashes. If message ID seen >3 times within 10 minutes (indicating re-delivery due to crash), log payload to Dead Letter Queue (DLQ), return null/sentinel value, and commit the offset. Prevent single bad message from blocking entire partition. Alternatively, offload heavy processing to separate worker thread pool outside poll() loop.