Avoid unsafe automatic replay after partial request transmission
The proxy retries a POST after its connection breaks, potentially duplicating a write.
- Focused work estimate
- 3h + prerequisites
- Priority in the scenario
- High
- Engineering practice
- HTTP retries · Consistency
Estimated field mix
- Networking40%
- API design30%
- Distributed systems30%
Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence.
Review it, then add it to your workspace.
The board opens an editable draft; nothing is saved until you confirm it. Sign-in and workspace permissions apply, and Demo boards remain ephemeral.
Project context
A fictional reporting API sits behind a reverse proxy. Clients see intermittent disconnects and incorrect origins after changes to keep-alive and forwarding headers.
Setup prerequisites
- Create two loopback HTTP services and a local proxy with synthetic requests; no public scanning or external traffic.
Preceding work
Complete these dependencies, or supply their agreed outputs before taking this ticket.
- BPROXY-101 · Document the trusted proxy chain and public origin
- BPROXY-102 · Normalize forwarded headers without accepting client spoofing
- BPROXY-103 · Strip hop-by-hop headers before forwarding requests
- BPROXY-104 · Align keep-alive lifetimes between proxy and upstream
- BPROXY-105 · Limit upstream connections per service without unbounded waiting
- BPROXY-106 · Propagate downstream disconnects through streamed responses
Acceptance criteria
- Distinguish retryable methods and declared idempotent operations.
- Track whether request transmission may have occurred.
- Return unknown outcome when safe replay cannot be established.
Implementation constraints
- Do not infer idempotency from an empty response.
Verification to include
- Retry a safe read after a stale connection.
- Break a write after transmission and prevent blind replay.
Deliverables
- Proxy retry policy.
Rollout and recovery
Disable automatic write retries until application identity semantics are explicit.
Value of the work
For the engineer: Practice HTTP intermediaries, connection lifetimes, and network failure isolation.
For the team: Provide a proxy contract that preserves request identity and releases resources predictably.
Evidence boundaries
Outcome Evidence: Tests, patches, and runbooks are requested deliverables. They become Outcome Evidence only through a qualified Mission and immutable Evidence IDs.
Ownership Evidence: Independent adaptation must be observed under a declared verification policy and cite immutable Evidence IDs. Completing a planning ticket establishes no Ownership Evidence.