Metrics Explorer

The Metrics Explorer is where you build a query against the Metrics Store, visualize the result, and act on it. This page covers every part of that: the query builder (metric, filters, operations), combining queries with formulas, the eight chart types, switching to raw PromQL, the time range and step size, and creating an alert directly from a chart. See Anatomy of a query for how the query builder’s three fields relate to PromQL structurally.

Access the Explorer

Click Metrics in the top navigation, then Explorer from the dropdown. The Explorer opens with an empty query and the Timeseries chart type selected.

The query builder

Each query row has three fields, filled in left to right:

Metric

A searchable dropdown listing every metric name in the Metrics Store. Type to filter as you search.

Filters

Optional label operator value conditions. Click Add Filter, pick a label from the autocomplete list, choose an operator (=, !=, =~ regex match, !~ regex exclude), then pick or type a value — the value field also autocompletes from what’s actually present for that label. Adding more than one filter combines them with AND logic; remove a filter with the × at its end.

Operations

An aggregation or function applied to the metric, with an optional by clause to group the result. Avg grouped by everything (a single series) is the default.

The chart above the query updates as you build it, and the Total series / Fetched series / Took line above the chart reports how many series the query matched and how long it took to run.

Operations: aggregations and functions

Click Add Operation to open the function picker, organized into categories:

Category What it contains

Range functions

rate, irate, increase, delta, idelta, deriv, changes, resets, predict_linear

Rollup

Rollup functions for downsampling a range of samples into one value per step

Aggregations

sum, avg, min, max, count, stddev, stdvar, group, topk, bottomk

Functions

General-purpose transforms (clamping, absolute value, rounding, and others)

Histogram

Functions specific to histogram metrics — see Histogram

Trigonometric

sin, cos, tan, and their inverse and hyperbolic forms

Time functions

Functions that return or operate on timestamps

Binary operations

Arithmetic and comparison operators between two operands

Time shift

Compares the current value to the same query at an earlier point in time

Add more than one operation to chain them — for example, rate followed by sum by (kube_cluster_name) turns a per-pod counter into a per-cluster rate. Remove any operation with its ×. This is the same picker used everywhere a query is built — in Derived Metrics and when creating an alert from a chart — and every function’s exact PromQL semantics are documented once, in PromQL, rather than repeated per screen.

Builder vs. Code

Each query row has a Builder/Code toggle at its right edge:

Builder

The structured metric/filters/operations form described above.

Code

A raw PromQL editor. Press kbd:[Enter] to run, kbd:[Shift+Enter] for a new line, or click Run.

Both edit the same underlying query — switching modes doesn’t lose your work. Reach for Code when an expression is easier to type than to build, or when it needs syntax the Builder doesn’t expose directly.

Multiple queries and formulas

Click + Query to add a second, independently-configured query row (labeled A, B, C, …​) to the same chart — useful for comparing two metrics, or two filtered views of the same metric, on one set of axes.

Click + Formula to combine existing queries mathematically. Reference each query by its letter — B / A * 100, for example, to turn a raw error count (B) and a raw request count (A) into an error percentage. The formula renders as its own series alongside the queries it references, and accepts the same Create alert action as any other series — see Alerting from a chart.

Remove a query or formula row from its own controls; the interface always needs at least one query.

Chart types

The tab strip above the query (Timeseries · Bar Chart · Stat · Gauge · Bar Gauge · Table · Top List · Pie Chart) re-renders the same query result as a different visualization — it doesn’t change what you queried, only how the result is drawn. Switch tabs freely while building a query to find the clearest view of it.

Chart options

Click Options next to the Builder/Code toggle to set:

Legend

How much detail the legend under the chart shows for each series (for example, the full label set vs. an abbreviated form).

Step (s)

The interval between plotted points. auto (the default) picks a step based on the current time range so the chart stays readable; enter an explicit value in seconds to override it — a smaller step shows more granularity but returns more data points.

Time range

The time range control (top right) defaults to the last 1 hour and accepts quick ranges or an absolute From/To range. It’s shared by every query row and every chart type on the page — changing it re-runs every query currently on screen.

Acting on a chart

Each chart has its own row of controls, independent of the others on the page:

Compare

Overlay the same query from an earlier period (for example, the same time yesterday) on the current chart, to spot a change at a glance.

Download

Export the chart’s underlying data.

Split view

Open a second panel alongside the chart for side-by-side comparison.

Full screen

Expand the chart to fill the screen, showing the same series at a larger size over the same time range; a download control remains available there.

Alerting from a chart

Click the alert (bell) icon on any query’s chart to open Create Metrics alert, pre-filled with that query exactly as built — filters, operations, and all — so you don’t re-enter it. Choose a detection method before configuring the rest of the alert:

Threshold

Fires when the query crosses a fixed value you set.

Change

Fires on a significant change in the query’s value over a comparison window.

Anomaly

Fires when the query deviates from its learned normal pattern.

Forecast

Fires when the query is predicted to cross a threshold within a future window.

Outliers

Fires when one series in a multi-series query diverges from the rest.

See Create a Metric Alert for the full alert configuration reference — field-by-field, for every detection method.

See also

  • Metrics Architecture — how the query builder’s fields map to PromQL, and how this screen relates to the other three

  • PromQL — the full PromQL operator and function reference

  • Metric Types — what counters, gauges, and histograms let you compute

  • Derived Metrics — precompute an expensive query on a schedule instead of re-running it live