Metrics in Insights
Break a metric down by label and read the result. There is no PromQL to write before you see something.
Open Insights → Metrics.
Find a metric
Type into the metric search. Every term you type has to appear somewhere in the
metric name, and order does not matter, so total http and http total return
the same set. _ and : stay inside a term, so node_cpu is one term rather
than two.
Pick a metric and Insights charts it over the time range in the toolbar.
Break it down
The breakdown grid splits the selected metric by one label and draws a panel per value. Use it for "which one" questions: which pod, which status code, which region.
- Choose the label from the selector above the grid. All instead draws one panel per label, which is a quick way to see which labels the metric even carries.
- Switch between Single, Grid and Rows to trade panel size against how many you see at once.
- Search the value tiles by name when there are more than a screenful.
- Include or Exclude on a tile narrows the whole session, not just that panel.
The grid shows up to 24 value tiles, and All mode up to 12 panels. On a high-cardinality label, add a filter first so the values you care about are in that window.
A label with thousands of values makes a slow query and an unreadable grid. Filter, then break down. Limits & quotas lists the ingest-side cardinality limits.
Write the query yourself
The toolbar's query button opens a drawer holding the raw PromQL for the current view. Edit it to override what the controls built. An active override is labelled in the toolbar, so results that no longer match the filters above them say so.
What to do with a result
Add to dashboard writes the current query into a dashboard panel.
Create alert opens the rule editor seeded with the expression and the data source.
Share copies a link to the view as it stands. The whole session lives in the
query string, filters included. A relative window such as now-6h re-resolves
in whoever opens it, at their clock, so the menu also offers a variant with the
time range pinned to absolute timestamps. Use that one when you are sharing
during an incident.
The same data over HTTP
This page queries the Prometheus-compatible endpoint you can call yourself. See Query API overview for instant and range queries over HTTP.