rewind

Rewind failed work

This sample repairs a failed activity and rewinds its orchestration after the dependency becomes available.

Run

Use a DTS service that supports rewind:

export DTS_CONNECTION_STRING='Endpoint=http://127.0.0.1:8080;TaskHub=default;Authentication=None'
go run ./samples/rewind

What it proves

• The original execution reaches FAILED before rewind is requested.
• RewindInstance enqueues recovery rather than waiting for it to finish.
• Recovery creates a new execution ID and produces the expected "kept:recovered" output.
• The successful activity runs once, while only the failed activity runs again.
• Cleanup purges only the sample-owned instance.

WaitForOrchestrationCompletion can initially return the previous failed execution, so the sample polls metadata until it observes the replacement execution complete. SAMPLE_OK rewind is printed only after the recovery assertions, exact-ID cleanup, and worker shutdown succeed.

Rewind follows the pinned Python SDK algorithm. Activity retry-policy histories that contain retry timers are not supported because they can replay nondeterministically. Rewind also cannot repair application code or a dependency that is still failing.