Summary

Targets upstream issue dapr/python-sdk#980.

A misbehaving runtime can replay the same unexpected timerFired event many times across activity retries, producing 1000+ identical warning lines for a single workflow request. Fix dedupes the warning per (instance_id, timer_id) pair using a bounded LRU cache on _OrchestrationExecutor.

Changes

• dapr/ext/workflow/_durabletask/worker.py
  • Add _UNEXPECTED_TIMER_WARN_CAPACITY = 1024 module constant.
  • Add _warned_unexpected_timers: OrderedDict[tuple[str, int], None] on _OrchestrationExecutor.__init__.
  • Add _warn_unexpected_timer_once(instance_id, timer_id) helper that emits the warning at most once per pair, evicting the oldest entry when capacity is exceeded.
  • Replace inline warning in the timerFired branch of process_event with the helper.
• tests/ext/workflow/durabletask/test_orchestration_executor.py — 4 new tests:
  • same key warns once across 50 repeats
  • distinct keys each warn once
  • LRU evicts oldest when capacity reached (verified via monkeypatch on _UNEXPECTED_TIMER_WARN_CAPACITY)
  • end-to-end via process_event with 100 repeated timerFired events emits a single warning

Validation

• uv run ruff check — clean
• uv run ruff format — clean
• uv run mypy — Success: no issues found in 175 source files
• uv run pytest -m "not e2e" ./tests/ext/workflow/ — 391 passed, 33 deselected

Safety / scope notes

• Helper is intentionally scoped to the timerFired warning (matches issue title style: limit durabletask warnings for timers). Identical patterns exist for taskCompleted, taskFailed, childWorkflowInstanceCompleted, and childWorkflowInstanceFailed — they can be refactored to the same helper in a follow-up.
• Capacity 1024 caps memory at ~tens of KB per worker. Entries are not cleared on orchestration completion (intentional — runs are typically short and the LRU bound makes manual cleanup unnecessary).
• DCO sign-off included.