A 90‑Minute Spike for Health Checks That Can Fail

说明与用途

90分钟的冲刺确保 /ready 在存储下线时返回 503，恢复后返回 200，防止流量路由到不健康的服务。

问题

Health‑check endpoints often return 200 even when critical dependencies are down, causing load balancers to route traffic to unhealthy services.

结果

After a 90‑minute spike, the service correctly reports 503 on /ready when the store is down and recovers to 200 once the store is available, preventing traffic to unhealthy instances.

工作流

• Set up a store stub that raises DependencyDown when STORE_DOWN=1
• Create a baseline handler that always returns 200 for /health and /ready
• Write kill tests that start the app with STORE_DOWN=1 and verify /ready stays 200
• Run tests for 90 minutes, cycling through constrained agent passes and repair passes
• Stop the spike after 90 minutes and document the ship‑or‑kill decision

证据

Unconstrained coding agents skip checks in health routes. They return 200 because the process started.

来源

Kill if /ready stays 2xx while a required store is down.

来源