What it is#

Grafana-style dashboards built on the metrics Pulsimo collects: a grid of panels that share a time range and variables. Existing Grafana dashboards can be imported.

Where to find it#

Dashboards in the sidebar. Endpoints and sources also link to their dashboards (Connected monitoring).

Panels#

Panel	Use it for
Time series	Anything over time: lines, areas, bars, points, stacked or percent-stacked, thresholds as lines or bands, legends as a list or a table with min/mean/max
Stat	One number with a sparkline, coloured by thresholds (e.g. "Application status")
Gauge, bar gauge	A value against a range (CPU, disk, pool usage)
Bar chart, pie/donut	Comparisons across routes, status codes, hosts
Table	Many series side by side
Text	Notes in Markdown

Working with a dashboard#

Time: quick ranges up to the retention period, an absolute range, ← → to shift, zoom out, and drag across any time series to zoom in. Auto-refresh pauses while the browser tab is hidden.
Variables at the top (e.g. route, instance) filter every panel. They support label_values(...), metrics(...), query_result(...), custom lists, constants, intervals and text boxes, multi-select and "All". Selections are kept in the URL.
Changes button: shows deploys, restarts and new routes as dashed vertical lines on every time series (on by default; Change markers). Hover a line to see what changed.
Monitors: chips under the title link to the endpoints this dashboard describes.
⋯ menu: TV mode (full screen), JSON model, export JSON, version history, delete.
Star a dashboard to pin it to the top of the list.

Editing#

Edit turns on edit mode: drag panels by the title, resize from the corner, add panels and collapsible rows. Save creates a new version (the last 50 are kept and can be restored). If someone else saved after you opened the dashboard, you choose to reload theirs or overwrite.

The panel editor#

A live preview on top with the dashboard's time and variables.
Queries below (A, B, C…), each in Code (the official PromQL editor with autocomplete for functions, metrics with their descriptions, labels, values and variables) or Builder (pick a metric, add label filters, chain operations such as rate, sum by, topk, histogram_quantile). Both stay in sync.
Metrics browser: search metrics and descriptions, click label values to narrow down, see how many series match, then insert the selector or a ready rate()/p95 query.
Hints point out common mistakes with a one-click fix: a counter without rate(), histogram buckets without histogram_quantile(), rate() on a gauge.
Query inspector: the final queries after variables, timings, data and raw response.
Options on the right: visualization, legend, unit, min/max/decimals, colours, thresholds, value mappings, per-series overrides.

Apply (Ctrl+S) puts the panel back; Save keeps the dashboard.

`$__rate_interval`#

Use rate(x[$__rate_interval]) rather than a fixed window. Pulsimo sizes it from the dashboard step and your sources' real scrape intervals (at least 4 scrapes), so a source pulled every 60 s doesn't show gaps or zeros. $__interval and $__range work as in Grafana.

Importing from Grafana#

Dashboards → Import, then paste the JSON or upload the file (Grafana: Dashboard settings → JSON Model, or Share → Export; community dashboards from grafana.com work too). Before anything is saved you get a report:

which panels mapped and to what (graph → time series, singlestat → stat or gauge…);
which became placeholders (types Pulsimo can't draw yet keep their place, size and queries, so nothing is lost);
which variables were imported or skipped (data-source variables aren't needed);
notes, such as queries for non-Prometheus data sources.

Tested with Node Exporter Full (1860), cAdvisor exporter (14282) and PostgreSQL Database (9628). For Node Exporter Full, job is the source type (node) and instance is host:port, so it works without edits.

Ready-made dashboards#

Metrics → Sources → Connect → Create dashboard builds one from what a source exposes (HTTP app, host, PostgreSQL, MySQL, containers) and attaches it to the endpoints linked to that source.

Performance#

Panels query only when scrolled into view, one batched request per panel; the step follows the panel width; lines are downsampled keeping their shape; at most 50 series per time series panel ("50 of N").