Problem
On-call just got paged for a dashboard timing out mid-incident: the query was scanning raw samples straight off the storage tier, and separately, a runaway service is quietly multiplying the number of distinct time series the cluster has to track. Build the metrics platform every other team's service pushes numbers into: counters, gauges, and histograms scraped or pushed at high frequency (10M samples/sec in production). Design it from a blank canvas so a sudden spike in ingestion volume never loses samples, one misbehaving service can't blow up the number of distinct series the whole cluster has to track, dashboards stay fast over a day of history including the last few un-flushed seconds, and high-resolution data doesn't sit on the most expensive storage tier forever once nobody queries it at full resolution anymore.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.