[zerafachris]

Fixes #3181

Root Cause

Since init_producer_id() was updated to drive non-transactional idempotent producers through UNINITIALIZED -> INITIALIZING -> READY, calling reset_producer_id() (when an in-retry batch expires or a pre-KIP-360 sequence error requires a producer ID reset) cleared producer_id_and_epoch to (-1, -1) while leaving _current_state at TransactionState.READY. On subsequent sender iterations, init_producer_id() returned early because _current_state != TransactionState.UNINITIALIZED, leaving the producer without a valid producer ID indefinitely.

Changes

• Allow TransactionState transitions to UNINITIALIZED from UNINITIALIZED and READY in TransactionState.is_transition_valid.
• Transition _current_state back to TransactionState.UNINITIALIZED in TransactionManager.reset_producer_id() so the next init_producer_id() call requests a fresh producer ID and epoch from the broker.
• Add test_idempotent_producer_init_then_reset_then_reinit in test/producer/test_transaction_manager_mock_broker.py verifying the full init -> READY -> reset -> UNINITIALIZED -> re-init -> READY cycle.

Test Evidence

• pytest test/producer/test_transaction_manager.py test/producer/test_sender.py test/producer/test_transaction_manager_mock_broker.py test/producer/test_producer.py (255 passed)

Prepared with AI assistance (OpenCode), reviewed for correctness before submission.

[hly0025]

I took a look at the above. Here is what was done: applied proposed change to 3.0.12 the outage test in #3181 recovered (5 of 5 messages delivered with a new producer ID)

[dpkp]

Thanks!