Prevent timer callbacks after orchestration completion

[andystaples]

Summary

Fixes #275.

Guard timerFired callbacks once the orchestration context is terminal. A later timer event can no longer schedule another long-timer chunk, dispatch an activity/sub-orchestration retry, or resume suspended orchestrator code after completion, failure, termination, or continue-as-new.

The guard is local to the timer callback, after task lookup and timer tracing. It does not break the history loop, change external-event handling, clear pending actions, or remove legitimate work scheduled before the terminal state. No Go coroutine-unwind/defer machinery is introduced; #274, #276, and #277 are intentionally out of scope.

Core and Azure Functions use chunked long timers. Azure Managed disables chunking because DTS supports native long timers; it benefits from the same terminal guard for retry dispatch and generator resumption, not a default chunk-continuation fix. All three Unreleased changelogs reflect this distinction. No versions or dependency minimums are changed; provider release coordination waits for the corrected core release.

Regression coverage

• Completion, unhandled exceptions, output-serialization failure, termination, and continue-as-new, including a terminal transition in replayed history.
• Trailing carryover events remain processed after the ignored timer callback.
• Short/final timer callbacks cannot resume terminal orchestrators.
• Activity and sub-orchestration retry callbacks with chunked and native timers; active retries still progress.
• Ordinary multi-chunk progression, late cancellation, cancellation of a newly scheduled chunk, and preservation of pre-terminal timer/send-event actions.
• Azure Functions worker response through both native and compatibility APIs.

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused guard matches the requested lifecycle behavior and is supported by comprehensive regression coverage.

Review effort: Balanced
Findings: None

What changed in this PR

Prevents timer callbacks from producing work after an orchestration reaches a terminal state.

Changes:

• Adds a terminal-state guard to timer processing.
• Adds regression coverage for timers, retries, and both Functions APIs.
• Documents behavior across all affected packages.

File	Description
durabletask/​worker.py	Skips terminal timer callbacks.
tests/​durabletask/​test_orchestration_executor.py	Covers terminal timer and retry scenarios.
tests/​azure-functions-durable/​test_worker_compat.py	Tests native and compatibility APIs.
CHANGELOG.md	Documents the core fix.
durabletask-azuremanaged/​CHANGELOG.md	Documents Azure Managed impact.
azure-functions-durable/​CHANGELOG.md	Documents Functions provider impact.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Bernd Verst (berndverst)]

Looks good. The timer-local terminal guard prevents post-terminal work while preserving legitimate earlier actions and subsequent history processing. Active timer behavior is retained, and the provider changelogs accurately distinguish native and chunked timers. No unjustified breaking change or new Python-version requirement.