The endpoint that 'feels slow' gets a number by Wednesday

Wednesday, a customer email: "checkout feels slow lately." Feelings aren't measurements, so you go get one.

Your API is exec scripts, which means it arrived instrumented. GET /api/v1/exec/monitor/scripts lists every endpoint with its request activity; POST /api/v1/exec/monitor/script-performance returns the duration distribution for default/1/checkout.ts. The p95 doubled on Monday. You pull that script's logs, find the vendor call you added Monday afternoon sitting in the request path, and move it out.

Thursday, the endpoint reads like its neighbors again. You reply to the customer with a before-and-after chart, not an apology.

Nobody installed an APM, nobody wired a tracing SDK. The measurement was standing next to the endpoint the whole time.

The measurement was standing next to the endpoint the whole time.