Dashboards

The four ways to create a dashboard, and its panels, sections, variables, and templates

One dashboard is one definition document, made of four parts.

{
  "schemaVersion": 1,
  "variables": [ ... ],   // values picked at the top of the screen
  "sections":  [ ... ],   // groups of panels
  "panels":    [ ... ],   // the tiles; each has one or more queries
  "settings":  { ... }    // default time range, refresh, timezone
}

Building one in the editor stores the same document. Field-level syntax is in the Definition JSON reference.

Four ways to create one

Path	When	Result
From a template	Starting out — copy a built-in template.	A copy is saved in your workspace; the template itself is untouched.
Blank dashboard	Composing panels from scratch	Starts from an empty definition you fill in the editor.
Import JSON	Uploading a document exported elsewhere, or one written by hand	Validated first, then saved under a name you choose.
Connect a repository	Managing definitions in Git	JSON files under the connected directory become dashboards, and each commit is applied.

Dashboards that come from a repository can still be edited on screen, and the edited definition can be written back to the file with "Push to repository". Edits you do not push are overwritten the next time that file is committed.

Panels

A panel is one tile. It is placed on a 12-column grid with layout (x, y, w, h), and visualization.type decides how it is drawn.

The visualization does not decide the shape of the result. The query's output.type does; the visualization only has to be compatible with it.

Visualization	Required result shape	Used for
`TIME_SERIES`	`TIME_SERIES`	A trend over time. Multiple series come from `groupBy`.
`NUMBER`	`SCALAR`	A single number — error rate, availability, anything answered with one value.
`TABLE`	`TABLE`	A table with axis columns and value columns.
`BAR`	`TABLE`	Bars per category, using one axis of the table as the category.
`PIE`	`TABLE`	Share per category.
`HISTOGRAM`	`HISTOGRAM`	Bucket distribution. Only histogram metrics can produce it.
`LOGS`	`TABLE`	Raw log lines. Used with a `records` query.
`TRACES`	`TABLE`	A list of traces; a row opens the trace detail.

Queries

A panel can hold several queries (up to 20), each identified by refId. Queries in the same panel can be divided with ratio — an error rate is two queries (errors, total) plus one query that points at them.

Sections

A section groups panels: a title plus panelIds. If it points at a panel that does not exist, the definition is rejected.

Variables

Variables are the values picked at the top of the screen. The picked value is not stored in the definition — the definition only holds the name and where to read candidate values; the selection arrives with each request.

Type	Value list	Used for
`SERVICE`	Service names in the workspace	Picking a service — the most common variable.
`ATTRIBUTE`	Attribute values actually collected	Values of the axis given by `source.signal` and `source.field` — environment, pod name, and so on
`TEXT`	None (typed in)	Values with no candidates to list, such as a search term
`CUSTOM`	A fixed list written in the definition	A fixed list that does not read from storage

Filters reference a variable as ${name}. The whole value must be the reference — prefix-${name} is rejected.

ALL is special: picking it removes the filter that uses that variable entirely — "all" is not a value, it means "do not filter". A multi variable accepts several values, and when more than one is bound, even an EQ filter is promoted to IN.

Settings

Field	Values	Default
`defaultTimeRange`	`LAST_30_MIN` · `LAST_1_HOUR` · `LAST_6_HOURS` · `LAST_24_HOURS` · `LAST_7_DAYS`	`LAST_1_HOUR`
`refreshInterval`	`NONE` · `10s` · `30s` · `1m` · `5m`	`30s`
`timezone`	`BROWSER` · `UTC`	`BROWSER`

One refresh issues as many queries as there are panels. A short interval on a large dashboard costs accordingly.

Built-in templates

A shared catalog, independent of any workspace. Picking one copies its definition into a new dashboard.

Template	Category	Shows
Service RED	APPLICATION	Rate, errors, and duration for a service, plus the deployed version
Trace Analysis	APPLICATION	Slow spans, failing paths, service-to-service calls
API Endpoint Performance	APPLICATION	Throughput, errors, latency per route (services sending `http.route`)
Application Metrics	APPLICATION	HTTP throughput and latency plus runtime state (`http.server.*` metrics)
Application Logs	APPLICATION	Log volume by service and level, plus the raw lines
Logs & Errors	APPLICATION	Error log trend and its spread over severity and service
Database Dependencies	DATA	Slow queries, call volume, errors per dependency (`db.system` client spans)
Deploy Impact	DEPLOY	Rate, errors, latency before and after a deploy, compared by version
Service Resources	INFRA	Pod CPU, memory, network, restarts (`k8s.*` metrics)
Kubernetes Workloads	INFRA	Restarts, CPU, memory per namespace and pod (`k8s.*` metrics)
Infrastructure Capacity	INFRA	Host CPU, memory, disk, network saturation (`system.*` metrics)
LLM Gateway	AI	Call volume, latency, token usage per model (`gen_ai.*` attributes)

If you are not sending the signals a template reads, its panels stay empty. A workspace that does not collect k8s.* or system.* metrics, for example, gets no values in the infrastructure templates. Check the instrumentation each template expects first.

Limits

Subject	Limit
Grid width	12 columns (`layout.x` + `layout.w` cannot exceed 12)
Panel height	24
Queries per panel	20
Panels executed at once	30
`groupBy` axes	5
Value columns per table (`measures`)	8
Time series	20 by default, 100 maximum (`limit`)
Points per series	1,500 — beyond that the bucket width grows
Raw record rows	200 per page — continued with a cursor

Field-by-field meaning and validation rules are in the Definition JSON reference; import, export, and repository sync are in Building dashboards as JSON.