Replayed and out-of-order reports
Prevent an old observation from silently replacing a newer state.
Research note · November 14, 2026Research question
How should a collector handle duplicates, delayed reports and reports from a restarted worker?
Method
- Generate a sequence of fixture reports with explicit identifiers, observation times and a worker-session identifier. Deliver them in order, then replay selected reports.
- Deliver an older report after a newer one and inspect both the event history and latest-state selection. Define the ordering rule before implementing it, including how uncertain timestamps are handled.
- Restart the worker with a new session identifier. Test whether sequence resets are distinguished from replay, and retain rejected reports for bounded diagnostic review.
Records to retain
- Report and session identifiers
- Event and arrival order
- Duplicate-rejection decisions
- Latest-state selection policy
Source basis
Event time and observation time can differ, as the OpenTelemetry data model makes explicit. Freshness and replay concerns also appear in remote-attestation architectures, but application reports need their own defined acceptance rules.
Interpretation & limits
Sequence numbers alone do not authenticate a report. A compromised source can fabricate new identifiers. This protocol does not claim replay protection is implemented in Contour.
Next experiments
- Test duplicated delivery without double-counting a machine.
- Evaluate clock rollback independently from network reordering.
References
Technical reference
Implementation details supporting this note, drawn directly from the maintained documentation. These are technical references, not additional papers or new experimental results. Expand a topic to inspect its method, fields and limits.