Aglet

Learn from repeated upload resume offset failures

Repeated resume failures often share a hidden convention about endpoints, chunk numbers, or upload identity. Learning should preserve the interruption traces that exposed the convention and make the expected checkpoint explicit so future retry changes can be reviewed against arithmetic instead of intuition.

Keep the lesson for the next incident

  1. Cluster the offset cases

    Group incidents by inclusive or exclusive endpoints, stale identities, final-chunk handling, lost acknowledgments, and client display drift. For each group, record the smallest file and interruption point that made the difference observable.

  2. Document the recovery contract

    State what the receiver acknowledges, how the next range is derived, when an upload may restart, and how partial state is named. Include a middle-chunk and final-chunk example with expected offsets that a reviewer can calculate by hand.

  3. Add a change trigger

    Require the fixture when changing chunk sizing, retry timing, upload identifiers, persistence, or progress calculations. Assign an owner for the contract and record whether a change affects byte assembly, recovery decisions, or only presentation.

What to carry forward

The learning record is useful when it preserves concrete offset arithmetic and names the state needed to recover safely. If the workflow cannot promise continuation for some interruptions, document the restart boundary and make that limitation visible before future work treats it as a defect.

Technical background: tus protocol documentation.

Keep the decision with the work.

Use a Work Item in Aglet to record the problem, the evidence you have, and the next decision. Add an owner and priority, then keep updates in the discussion so the next person can follow the reasoning.

Create an account See the product workflow