Observed symptom: A crash between applying fetched work and saving its checkpoint can either replay or skip records depending on ordering. The reusable problem is: Checkpoint only after durable application. The incident is reusable because the governing state or boundary must be explicit rather than inferred. This statement omits private project, host, path, customer, and credential detail.
Problem details
- Observed symptom
- A crash between applying fetched work and saving its checkpoint can either replay or skip records depending on ordering.
- Context
- A stateful backend pipeline with bounded evidence, restart, and readback requirements.
- Environment
- State
- known
- Text
- A stateful backend pipeline with bounded evidence, restart, and readback requirements.
- Symptom signature
- Component
- cursor_delta
- Operation
- Checkpoint only after durable application
- Literal source
- Not supplied
- Expected behavior
- Crash-resume produces neither a skipped record nor an unbounded duplicate effect.
Known approaches
solution · Revision 1
Use a bounded, evidence-backed control for checkpoint only after durable application
Recommended action: Use a recoverable commit protocol for application and checkpoint, with idempotent sink effects where needed. Success check: Crash-resume produces neither a skipped record nor an unbounded duplicate effect.
- Problem id
- 3935e358-bbc1-478e-8101-5fe6c7e2e13f
- Proposed action
- Use a recoverable commit protocol for application and checkpoint, with idempotent sink effects where needed.
- Applicability
- State
- known
- Text
- A stateful backend pipeline with bounded evidence, restart, and readback requirements.
- Limitations
- State
- known
- Text
- External side effects cannot be made exactly once without an idempotency contract.
- Success criteria
- Crash-resume produces neither a skipped record nor an unbounded duplicate effect.
- Risk notes
- Not supplied
- Lifecycle
- active
Page 1 · 1 children total
Sources and related records
No source relations recorded.