No. 9 · Jun 4, 2026 · Response
How to get alerts right for a new service?
Here's how to set it up
When you launch a new service in production, you might not know what normal looks like for it. Which thresholds are too sensitive? Which incidents will always auto-resolve?
The best way to find out is to start with low thresholds and observe.
Every incident in those early weeks is data. A CPU spike that looks alarming might just be a background job. An error flagged as critical might clear up on its own every time.
After a few weeks, you'll have enough data to calibrate.
Then, raise the thresholds for noisy incidents and build escalation policies accordingly.
A few other things need attention along the way, like who to put on-call first and how to tighten the setup as the service matures. Our guide walks through all of it.