Config and the test runner

lightningpanda test is the counterpart of playwright test. It reads your config, finds each project's test files, starts web servers and global setup once, then runs the files in worker processes. Each worker is a bun test process with its own Pyro browser. This page lists every flag, every config field that is honoured, and how workers split the work.

Usage

lightningpanda test [options] [test-filter...]

A test-filter is a regular expression. A file runs when its absolute or working-directory-relative path matches any filter. Playwright's file.spec.ts:42 line filters are not supported.

The command needs Bun on PATH. When the CLI is started by Node, it starts itself again under Bun, and fails with "needs Bun (https://bun.sh) on PATH" if Bun is missing.

Flags

Flag	Meaning
-c, --config <file>	Config file. Default: lightningpanda.config.*, else playwright.config.* in the working directory.
--project <name>	Run only this project. Repeatable. Without it, every project's files run.
-j, --workers <n|n%>	Worker processes. Default: the config's workers, else half the CPUs. 50% is a share of the CPUs.
-g, --grep <regex>	Only tests whose title path (file, describes, title, tags) matches.
--grep-invert <regex>	Skip tests whose title path matches.
--retries <n>	Retries of a failing test. Overrides the config.
--timeout <ms>	Test timeout. Overrides the config and every project.
-u, --update-snapshots [mode]	Rewrite snapshots and screenshot baselines. changed (the default), all, missing or none. See Snapshots.
--fully-parallel	Run every file's tests in parallel, as fullyParallel: true.
-x	Stop after the first failure (same as --max-failures 1).
--max-failures <n>	Stop after n failures.
--reporter <name>	list (default), line or dot.
--list	List the tests without running them.
-q, --quiet	Do not print the tests' own output.
-h, --help	Print the help.
--headed, --debug, --ui	Print a warning and have no effect: Pyro is headless.

Any other flag (--shard, --trace, --repeat-each, --last-failed, --forbid-only, --output, ...) is an error: the command prints the help and exits with status 2.

Config fields

defineConfig(config, ...overrides) merges later configs over earlier ones, with use and expect merged key by key, as Playwright does. A plain exported object works too.

Honoured

Field	Where	Notes
use	config, project	The project's use is merged over the config's. See use options.
projects	config	Each has name, use, testDir, testMatch, testIgnore, outputDir, timeout, retries, snapshotDir, snapshotPathTemplate, expect, metadata.
testDir	config, project	Default: the config file's directory.
testMatch	config, project	Default `**/*.@(spec
testIgnore	config, project	Same matching as testMatch.
timeout	config, project	Test timeout. Default 30 000 ms.
retries	config, project	Default 0.
expect.timeout	config	Default timeout of web-first assertions and expect.poll. Default 5 000 ms.
expect.toHaveScreenshot	config, project	threshold, maxDiffPixels, maxDiffPixelRatio defaults, and pathTemplate.
expect.toMatchSnapshot	config, project	threshold, maxDiffPixels, maxDiffPixelRatio defaults.
expect.toMatchAriaSnapshot	config, project	children (contain, equal, deep-equal) and pathTemplate.
outputDir	config, project	Default test-results. Each test writes into a sub-directory named after its title path.
snapshotDir, snapshotPathTemplate	config, project	Where snapshots live. See Snapshots.
updateSnapshots	config	Default update mode when -u is not given. Default missing.
webServer	config	One or a list. See below.
globalSetup, globalTeardown	config	A path or a list of paths. See below.
workers	config	A number or a percentage string.
fullyParallel	config	Top level only.
metadata	config, project	Merged into testInfo.project.metadata.

A project's expect replaces the top-level expect instead of merging with it, as in Playwright.

Accepted and ignored

These typecheck but do nothing: reporter, globalTimeout, maxFailures (use --max-failures), forbidOnly, reportSlowTests, quiet, preserveOutput, captureGitInfo, tsconfig, shard, and on projects grep, grepInvert, dependencies, repeatEach, teardown and fullyParallel. In webServer, gracefulShutdown, name and ignoreHTTPSErrors are ignored.

Project dependencies are worth a note: a "setup" project that logs in and saves storageState does not run first. Run that code in globalSetup instead.

Projects

lightningpanda test collects the files of every project (or those named with --project). Each file runs once, under the first project whose testDir, testMatch and testIgnore include it. Playwright would run such a file once per project.

Under plain bun test, the project is $LP_PROJECT if set, else the project whose files include the test file, else the first one.

use options

All Playwright use options are accepted. These take effect:

Option	Effect
baseURL	Resolves relative URLs in goto, waitForURL, toHaveURL, routes and request.
viewport	Default { width: 1280, height: 720 }. null leaves the browser's own size.
storageState	Cookies and localStorage (path or object). See Network.
extraHTTPHeaders	Sent with every page request and by request.
userAgent	Overrides the page's user agent.
colorScheme, reducedMotion	Media emulation for @media and matchMedia.
httpCredentials	Used by the request fixture only (Basic auth); pages do not use it.
ignoreHTTPSErrors	Used by the request fixture only.
actionTimeout, navigationTimeout	Default timeouts. With neither set, an action or navigation waits up to 30 s (Playwright waits until the test times out).
testIdAttribute	The attribute getByTestId matches. Default data-testid.
contextOptions	Merged under the named options.
acceptDownloads	Default true.
lightningpanda	How tests reach the browser. See below.

These are accepted and have no effect on pages: bypassCSP, deviceScaleFactor, geolocation, hasTouch, isMobile, javaScriptEnabled, locale, offline, permissions, proxy, timezoneId, serviceWorkers, and the worker options browserName, channel, headless, launchOptions, connectOptions, screenshot, video, trace. A proxy goes on the browser instead: lightningpanda: { args: ['--proxy', 'http://host:3128'] }.

browserName still sets the browserName fixture, so test.skip(({ browserName }) => browserName === 'webkit') behaves as the config says.

use.lightningpanda

use: {
  lightningpanda: {
    wsEndpoint: 'ws://127.0.0.1:9222/',     // connect to a running server instead of launching one
    bin: '/path/to/lightningpanda',          // the binary to launch
    args: ['--static', 'http://app.test=./dist'], // extra `lightningpanda serve` flags
    boot: [{ url: '/platform', routes: '/platform/**' }], // pages booted ahead
  },
},

Field	Default	Meaning
wsEndpoint	$LIGHTNINGPANDA_WS	A running lightningpanda serve to connect to. Without it, each worker launches its own browser.
bin	$LIGHTNINGPANDA_BIN, else the bundled binary	The binary each worker launches.
args	none	Extra lightningpanda serve flags, for example --static, --proxy, --hyphen-digit-breaks on. See serve options.
boot	none	Pages booted ahead of the tests. See Booting pages ahead.

lightningpanda is a worker option, so test.use({ lightningpanda: { ... } }) in a file works. Each distinct value gets its own browser in the worker.

With wsEndpoint (or $LIGHTNINGPANDA_WS), all workers share one server, one connection each. Start it yourself, for example lightningpanda serve --port 9222, or in globalSetup.

Environment variables

Variable	Meaning
LP_CONFIG	Config file path. PLAYWRIGHT_CONFIG is read when it is unset.
LP_PROJECT	Project for plain bun test runs.
LP_UPDATE_SNAPSHOTS	Update mode (all, changed, missing, none; 1 or true mean changed). Set by -u.
LP_RETRIES, LP_TIMEOUT	Override retries and test timeout. Set by --retries and --timeout.
LP_CONCURRENT	1 runs the tests of a parallel describe concurrently inside one worker (see Workers).
LIGHTNINGPANDA_WS	Connect to this server instead of launching one.
LIGHTNINGPANDA_BIN	Binary to launch.

See Environment variables for the browser's own variables.

Workers

The runner starts workers processes at once. Each runs bun test with its own browser and its own module state. Files are dealt out whole: sorted by size, largest first, round robin over the workers.

• A file's tests run in order, in one worker, unless they are parallel.
• Parallel tests (fullyParallel, --fully-parallel, or a describe with mode: 'parallel') spread over the workers by their position in the file. A describe's beforeAll and afterAll run in each worker that runs some of its tests, and in no other.
• A describe in serial mode, or default mode set with test.describe.configure, keeps its tests on the file's worker.
• Inside a worker, parallel tests still run one after another, since they may share state set in hooks. Set LP_CONCURRENT=1 to run a parallel describe's tests concurrently in the worker (Bun's test.concurrent).
• Web servers and global setup run once, in the parent process, before any worker starts.
• A worker that gets no tests exits before it starts a browser.

--list uses one worker.

The worker count is not capped by the number of files, because parallel describes can spread one file over every worker. With few files and no parallel tests, lower -j to avoid idle processes.

Web servers

webServer is one object or a list. For each:

1. If url (or port) already answers, the server is reused, unless reuseExistingServer: false, which is an error.
2. Otherwise command runs in a shell, in cwd (relative to the config file) with env added, in its own process group.
3. The runner polls url (any status below 404, or 405) or the TCP port every 100 ms until it answers, for up to timeout (default 60 000 ms). It fails if the command exits first.

stdout: 'pipe' shows the server's output; stderr is shown unless stderr: 'ignore'. The servers are stopped after the run, also when the run ends early.

Global setup and teardown

globalSetup and globalTeardown take a path or a list of paths, relative to the config file. Each module's default export (or the module itself) is called with the config. A function returned by a global setup runs at teardown, as in Playwright. Teardowns run after the last worker ends.

Reporting

The report shows each test as it finishes: passes, failures, skips (declared, or test.skip() at run time), flaky tests (passed on a retry) and expected failures (test.fail()). The tests' own console.log output is printed per worker, prefixed with [w0], [w1], ... After the run come the failures with their errors and a summary.

Running 2 files using 4 workers

  ✓   1 e2e/todo.spec.ts › todos › adds a todo (359ms)
  ✘   2 e2e/todo.spec.ts › todos › ua and role (4962ms)
  -   3 e2e/todo.spec.ts › todos › runtime skip (skipped at run time)

  1 failed
  1 skipped
  1 passed (5.9s)

• list prints one line per test.
• line prints the same lines as list in this version.
• dot prints one character per test: · pass, ± flaky, ° skipped, F failed.

There is no HTML, JSON or JUnit reporter. A worker that crashes outside a test (a syntax error in a test file, for example) is reported with the last 40 lines of Bun's output.

The exit status is 1 when any test failed or a worker crashed, else 0.

Plain bun test

LP_CONFIG=playwright.config.ts bun test ./e2e/foo.spec.ts

This runs the files in one process with one browser. The config, web servers and global setup are loaded on the first file; teardowns run after the last file. Bun's own flags apply (-t for names, --bail, -u for snapshots).

Differences from lightningpanda test: no workers or fullyParallel spreading, Bun's reporter instead of the runner's, and a run-time test.skip() is reported by Bun as a pass (it prints ↳ skipped: <title> to the log).