Repair HTTP/2 GOAWAY Retry Bugs

The problem

HTTP/2 connection rotation and graceful drains often trigger race conditions that cause dropped requests or dangerous duplicate retries. Standard retry logic frequently fails to distinguish between streams the server definitely ignored and those with unknown processing outcomes during a GOAWAY event.

What it does

• Identifies and repairs repository-level HTTP/2 transport wrappers to handle GOAWAY frames correctly.
• Prevents connection pools from assigning new streams to draining connections.
• Implements stream-aware retry classification based on the peer's last-processed stream ID.
• Enforces body replayability checks and idempotency safety during connection failovers.
• Generates deterministic local HTTP/2 test fixtures to validate drain semantics without production traffic.

Frameworks & tools

Works with HTTP/2 client libraries, custom transports, and connection pool implementations. Requires standard shell tools for local testing and file editing.

Why this beats prompting it yourself

Generic LLM prompts often suggest blindly retrying failed requests, which leads to double-processing and data corruption. This skill follows a strict protocol to isolate transport-level retryability from application-level operation safety, ensuring your fix handles edge cases like concurrent stream herds and unknown outcomes.

Use cases

• Fixing request failures during rolling deployments or load balancer connection rotation.
• Stopping retry storms triggered by frequent GOAWAY events in high-concurrency environments.
• Adding observability to track connection drain state and stream-level retry success rates.
• Validating client behavior against specific last-stream ID boundaries in local CI.

Known limitations

Cannot capture production traffic or invent provider-side idempotency guarantees. Local tests may not perfectly replicate the behavior of specific production intermediaries or hardware proxies.