Panel testing strategy

This skill builds on frontend-testing-strategy — read that first for the general principles every Grafana frontend test is held to (the inverted testing diamond, asserting real behavior instead of existence, avoiding AI slop, verifying a test reaches its target branch, generic anti-flake rules, and the SDLC-phase gating). This skill covers what's specific to visualization code on top of that: data-frame/panel-prop builders, the canvas draw-call snapshot harness, panel accessibility and interaction-snapshot E2E, and canvas/uPlot-specific anti-flake rules. The visualization codeowner paths are opted into the gating check-frontend-test-coverage.yml check, so coverage that drops fails CI.

Step 1 — Set up data with the repo's builders

Build data frames with the @grafana/data builders — pick one and don't mix toDataFrame and createDataFrame in the same file:

import { createDataFrame, toDataFrame, arrayToDataFrame, FieldType, LoadingState } from '@grafana/data';

Use a single canonical builder per file with a Partial<> overrides object, rather than bespoke frames per test:

function makeFrame(overrides: Partial<Options> = {}) {
  /* … */
}

To render a panel component, use the shared panel-props builder instead of hand-rolling props:

import { getPanelProps } from '../test-utils'; // public/app/plugins/panel/test-utils.ts
render(<BarChartPanel {...getPanelProps(defaultOptions, { fieldConfig })} />);

Gotcha — field config. A panel unit test must call applyFieldOverrides itself with a createFieldConfigRegistry; the panel framework normally does this, so without it your custom fieldConfig.custom never reaches the render and every case looks identical.

Gotcha — type inference. If you're testing guessFieldTypes (or any inference), feed untyped raw fields (as unknown as DataFrameDTO). createDataFrame pre-sets type, so the function under test becomes a no-op and the test gives false confidence.

Step 2 — HTML5 canvas / rendering panels: use the draw-call snapshot harness

Panels that draw to canvas (timeseries, heatmap, xychart, timeline, piechart, sparkline) are tested by capturing the ctx draw-call stream, not by pixel-diffing. Follow the established harness:

// In the harness (public/app/plugins/panel/timeseries/TimeSeriesPanel.canvasTestUtils.tsx):
import {
  applyDefaultUPlotAxisMeasureTextMock,
  installCanvasPath2DShim,
  removeCanvasTransforms,
} from '@grafana/test-utils/canvas';

// In each *.canvas.test.tsx, mock grafana-ui's text measurement so layout is deterministic:
jest.mock('@grafana/ui/src/utils/measureText', () =>
  require('@grafana/test-utils/canvas').createGrafanaUiMeasureTextJestMock(() =>
    require('./TimeSeriesPanel.canvasTestUtils').getUPlotInstance()
  )
);

• Split suites by concern: *.lines.canvas.test.tsx, *.fills.…, *.annotations.…, *.axisPlacement.…, *.axisRange.… — each a focused it.each of cases.
• Assert with the custom matcher: expect(events).toMatchCanvasSnapshot(context, { width, height }).
• Keep it deterministic (this is where flake comes from): fixed width/height, UTC timestamps (Date.UTC(...), timeZone: 'utc'), and wait for the renderer to be ready before asserting — await waitFor(() => expect(uPlotInstance?.status).toBe(1)) (a waitFor callback must throw to retry, so it needs expect, not a bare boolean).

Step 3 — E2E for interaction, accessibility, and interaction snapshots

The DataViz strategy is unit-first. Reserve Playwright for cross-component interaction and per-panel smoke coverage. When you do write E2E:

• Add the selector to the versioned @grafana/e2e-selectors package first, wire data-testid into the JSX, then query it (use the add-e2e-selectors skill).
• Query by selector, never brittle CSS — in E2E dashboardPage.getByGrafanaSelector(...), in unit screen.getByTestId(selectors.components...).

import { test, expect } from '@grafana/plugin-e2e';

test.describe('Panels test: BarChart render', { tag: ['@panels', '@barchart'] }, () => {
  test('renders without error', async ({ gotoDashboardPage, selectors }) => {
    const page = await gotoDashboardPage({ uid: DASHBOARD_UID }); // provisioned devenv dashboard
    await expect(page.getByGrafanaSelector(selectors.components.Panels.Panel.headerCornerInfo('error'))).toBeHidden();
  });
});

Accessibility — every panel gets an a11y check

Every panel must have an E2E accessibility test. Use the scanForA11yViolations fixture and the toHaveNoA11yViolations() matcher, in a describe/test tagged @a11y. Load the panel, wait for it to actually render (assert the panel title and the chart element are visible — an empty panel trivially passes), then scan:

test.describe('a11y', { tag: ['@a11y'] }, () => {
  test('run a11y report', async ({ gotoDashboardPage, scanForA11yViolations, selectors, page }) => {
    const dashboardPage = await gotoDashboardPage({
      uid: DASHBOARD_UID,
      queryParams: new URLSearchParams({ viewPanel: 'panel-4' }),
    });
    await expect(dashboardPage.getByGrafanaSelector(selectors.components.Panels.Panel.title('…'))).toBeVisible();
    await expect(page.locator('.uplot')).toBeVisible(); // panel has drawn

    const report = await scanForA11yViolations({
      options: { runOnly: ['wcag2a', 'wcag2aa', 'wcag21a', 'wcag21aa', 'best-practice'] },
    });
    expect(report).toHaveNoA11yViolations();
  });
});

• Only pass ignoredRules for a documented, tracked pre-existing violation — add a @todo with the tracking issue rather than silently ignoring (e.g. page-has-heading-one, region, color-contrast are common app-shell noise, not panel bugs).
• Exemplars: e2e-playwright/panels-suite/{histogram,xychart,table-nested,table-kitchenSink}.spec.ts, and keyboard-a11y in e2e-playwright/various-suite/panel-presets.spec.ts.

Interaction snapshots — cover a variety of states, not just first render

A panel's accessibility and structure change as the user interacts. For each panel, drive a variety of interaction states and snapshot the resulting accessibility tree with toMatchAriaSnapshot, re-running the a11y scan in the states that matter. Typical states per panel type: default render, hover / tooltip open, legend item toggled, sort / filter applied (table), series selected, panel edit mode, and empty / no-data.

const panel = dashboardPage.getByGrafanaSelector(selectors.components.Panels.Panel.content);
await expect(panel).toMatchAriaSnapshot(); // baseline structure
await panel.locator('.uplot').hover({ position: { x: 120, y: 80 } });
await expect(panel).toMatchAriaSnapshot(); // tooltip-open state
expect(await scanForA11yViolations()).toHaveNoA11yViolations(); // a11y holds mid-interaction

Keep these deterministic — see the canvas/uPlot anti-flake rules below (pin data, scope locators, wait for the renderer). Aria snapshots capture semantic structure, not pixels; leave pixel-level visual regression to Meticulous.

Canvas / uPlot anti-flake rules

These are the viz-specific additions to frontend-testing-strategy's generic anti-flake list. Avoid → Do:

1. Non-deterministic data / relative time ranges. Avoid random_walk + now-30m when asserting on shapes or coordinates. Do pin an absolute time range and a fixed seed / startValue so the render is identical every run. (timeseries tooltip #128617)
2. Coordinate-based hover/click on canvas. Avoid hardcoded x/y on .uplot. Do make data deterministic first; derive coords from rendered geometry, not constants. (xychart tooltip remains skipped #128389 for this reason)
3. Broad locators. Avoid page.locator('.uplot') — it also matches option-pane preview thumbnails. Do scope: getByGrafanaSelector(Panels.Panel.content).locator('.uplot').
4. Asserting before the renderer is ready. Do waitFor(() => expect(uPlotInstance?.status).toBe(1)) before any canvas snapshot/output assertion. (Sparkline/Heatmap/XYChart #127557)

Rules checklist

• Read frontend-testing-strategy first — its checklist (Principles 1-4, test naming, mocking convention, generic anti-flake, SDLC gating) applies here too.
• Step 1 — one data builder per file; getPanelProps + applyFieldOverrides for panel renders.
• Step 2 — canvas panels → draw-call harness, deterministic, wait for status === 1.
• Step 3 — E2E selectors-first; every panel gets an @a11y test plus interaction aria-snapshots.
• Canvas/uPlot anti-flake — apply all 4 rules above, on top of the generic 7.

Exemplar files

Additional exemplars not already cited inline above (Step 1 has the panel-props builder, Step 2 the canvas harness, Step 3 the a11y specs):

• Behavior-specific util tests with typed uPlot mocks & it.each: public/app/plugins/panel/barchart/bars.test.ts
• Concrete-value assertions & clear descriptions: packages/grafana-ui/src/components/Table/{utils,cellUtils}.test.ts, packages/grafana-ui/src/components/uPlot/config/gradientFills.test.ts
• E2E panel spec + shared helpers: e2e-playwright/panels-suite/table-footer.spec.ts, e2e-playwright/panels-suite/table-utils.ts

See also the add-e2e-selectors skill, contribute/style-guides/e2e-playwright.md, and packages/grafana-e2e-selectors/src/selectors/README.md.

Verify

• yarn test <path> (add --watchAll=false) — the new tests pass and actually fail when the asserted value is broken (mutate the expected value once to confirm it's not a no-op).
• For E2E: yarn e2e:playwright <spec> (it starts its own server).
• yarn typecheck if selectors or casts were added.