Research2026
EventFlowSentry: reproducible fault testing for streaming
Turns event-time failures in streaming pipelines into reproducible, explainable experiments.
Model
Fault definition
Late, reordered, replayed events
Run
Pipeline under test
Check
Metamorphic oracle
Compare related runs
Explain
Replayable trace
Context
Streaming systems can pass ordinary tests and still produce different results under late data, reordering, replay, or recovery. Those are exactly the conditions production traffic creates.
The problem
Random chaos testing finds failures but rarely explains them, and a failure you cannot replay is hard to fix or trust. The research question is how to generate timing faults whose expected behavior stays checkable.
My role
Author. Designed the fault model, the metamorphic oracle, and the evaluation.
Approach
- Represents event-time perturbations as explicit, repeatable experimental inputs.
- Uses metamorphic relations to compare related runs when no single expected output exists.
- Keeps the full execution conditions so every observed failure can be replayed and inspected.
Outcome
- Published as a public preprint with a DOI.
- Connects directly to the CDC and checkpointing work I do upstream.