Suite runner: lightningpanda e2e

lightningpanda e2e runs a whole suite in one process. Tests from every spec file go into one queue that runs up to --concurrency of them at a time, so cases of different files interleave. They all share one fork-mode browser, which the runner launches on first use and always closes. Every profile therefore boots once per run, and every test forks its page from it.

Use it instead of bun test --parallel for large suites. bun test runs the files of a worker one after another, so overlapping tests of different files takes worker processes, and each worker launches its own browser and boots its own profiles.

It needs Bun on PATH: the CLI hands e2e over to Bun even when started from Node.

A spec file

Specs import describe, test, the hooks and expect from @povio/lightning-panda/e2e instead of bun:test, and get the browser from browser():

// e2e/todos.test.ts
import { beforeAll, browser, defer, describe, expect, test } from '@povio/lightning-panda/e2e';
import { button, check, label, testId, text } from '@povio/lightning-panda';
import { app, signedInProfile } from './app';

describe('todos', () => {
  beforeAll(async () => {
    await signedInProfile();                       // shared by every file, registered once
    const lp = await browser();
    await lp.profile('todos', { base: 'app', url: `${app()}/`, setup: [check(label('New todo')).visible()] });
    defer(() => lp.releaseProfile('todos'));       // after this describe's last test
  });

  test('adds a todo', async () => {
    await using page = await (await browser()).page({ as: 'todos' });
    const r = await page.run([
      label('New todo').fill('Buy milk'),
      button('Add').click(),
      check.request('POST', '/api/items').status(201),
      text(/^Added /).read('status'),
    ]);
    expect(r.values.status).toBe('Added Buy milk');
  });

  test('deletes a seeded todo', async () => {
    const item = await (await fetch(`${app()}/api/seed`, { method: 'POST', body: JSON.stringify({ title: 'Walk' }) })).json();
    await using page = await (await browser()).page({ as: 'todos', route: '/' });
    await page.run([
      button('Delete Walk').click(),
      button('Delete', { exact: true }).click(),
      check(testId(`item-${item.id}`)).count(0),
    ]);
  });
});

A profile that many files share is registered once per run. Module state is shared by every file (one process), so memoize the promise:

// e2e/app.ts
import { browser, setupResult } from '@povio/lightning-panda/e2e';
import { button, check, label, text } from '@povio/lightning-panda';

export const app = () => setupResult<{ app: string }>().app;

let signedIn: Promise<void> | undefined;
export function signedInProfile() {
  return (signedIn ??= (async () => {
    const lp = await browser();
    await lp.profile('signed-out', { url: `${app()}/` });
    await lp.profile('app', {
      base: 'signed-out',
      url: `${app()}/`,
      setup: [label('Email').fill('ada@example.com'), button('Sign in').click(), check(text('Signed in')).visible()],
    });
  })());
}

$ bunx lightningpanda e2e e2e --setup e2e/setup.ts -j 4 --timeout 10000 --junit junit.xml
lightningpanda: CDP server listening on ws://127.0.0.1:49843/ (fork mode)
✓ e2e/settings.test.ts > opens settings [320.10ms]
✓ e2e/todos.test.ts > todos > adds a todo [159.22ms]
✓ e2e/todos.test.ts > todos > deletes a seeded todo [166.88ms]
✓ e2e/todos.test.ts > todos > runs alone [29.62ms]

 4 pass
Ran 4 tests across 2 files. [415.78ms]

The first tests of a run include the profile boots they waited for.

Options

lightningpanda e2e [options] [path | dir | glob | filter ...]

Option	Default	Meaning
-j, --concurrency <n>	CPU count	Tests running at once across all files
--max-concurrency <n>	no limit	Tests of one file running at once (as in bun test)
--timeout <ms>	5000	Default timeout of tests and hooks
-t, --test-name-pattern <re>		Only tests whose name (describe names and test name joined by spaces) matches the regular expression
--only		Only tests and describes marked .only
--shard <i>/<n>		Run shard i of n; files are split by --timings
--timings <file>		Per-file durations, {"version": 1, "files": {"<path>": ms}}
--update-timings		Write this run's per-file durations to --timings
--setup <module>		A setup module (see below)
--preload <module>		Loaded before the specs; its top-level hooks apply to every test. Repeatable
--browser-arg <arg>		Extra lightningpanda serve argument. Repeatable
--reporter junit	console	Also write a JUnit report, to --reporter-outfile
--reporter-outfile <file>		Where the JUnit report goes
--junit <file>		Both at once
-u, --update-snapshots		Write screenshot baselines instead of comparing
--bail [n]		Stop after n failures (1 without a number)
-h, --help		Print the usage

Options that take a value also accept --option=value. An unknown option or a bad value prints the usage and exits with code 2.

Paths

Arguments are files, directories, globs or filters:

• A directory is searched for files matching *.test.*, *.spec.*, *_test.* or *_spec.* with a js, ts, jsx, tsx, mjs, cjs, mts or cts extension, skipping node_modules and dot directories.
• A file runs as given, whatever its name.
• A glob (an argument with *, ?, [ or {) runs every file it matches. Use one for other naming schemes: lightningpanda e2e 'e2e/**/*.e2e.ts'.
• Anything else is a filter: only discovered files whose path contains it run, as with bun test.

Without paths the runner searches the working directory. No spec files found is an error (exit code 1).

The setup module

--setup <module> names a module that prepares the run. Export the functions one by one, or as a default export (type it with satisfies Setup<T>):

// e2e/setup.ts
import type { Setup } from '@povio/lightning-panda/e2e';
import { startServer } from '../server';

type Runtime = { app: string; stop: () => void };

export default {
  async setup() {
    const server = startServer();
    return { app: server.url.origin, stop: () => server.stop(true) };
  },
  browser: () => ({ args: [] }),
  teardown(rt) {
    rt.stop();
  },
} satisfies Setup<Runtime>;

Export	Called	Contract
setup()	Once, before any spec file loads	Start servers, seed data. Its (awaited) return value is the run's setup result
browser(result)	When the first test calls browser()	Returns { args?: string[]; bin?: string }: extra lightningpanda serve arguments (--static, --proxy, ...) and the browser binary (default $LIGHTNINGPANDA_BIN, else the bundled one)
teardown(result)	Once, after the run, also when it is interrupted	Stop what setup started. A failure counts as a hook failure

Specs read the result with setupResult<T>(), even while they load. Without --setup, setupResult() throws.

The runner launches the browser with --fork when the binary supports it (it checks lightningpanda version for fork: true), plus --cdp-max-connections set to the larger of 64 and 4 × concurrency + 8 unless you pass that flag yourself. Arguments from browser(result) come first, then --browser-arg values.

Spec API

Everything below is exported from @povio/lightning-panda/e2e. Signatures follow bun:test; expect is Bun's own.

Export	Meaning
describe(name, fn)	A group. fn must register its tests synchronously
describe.serial, .skip, .todo, .only, .if(cond), .skipIf(cond)	Modifiers, as below
test(name, fn, timeout?), it	A test. timeout is a number or { timeout }
test.serial	Runs alone in its describe (see Serial tests)
test.concurrent	Same as test: every test is concurrent unless it is serial
test.skip, test.todo(name, fn?), test.only, test.if(cond), test.skipIf(cond)	As in bun:test; .only takes effect with --only
beforeAll, afterAll, beforeEach, afterEach	Hooks of the describe (or file) they are declared in
setDefaultTimeout(ms)	Default timeout of tests and hooks without their own
browser()	The run's browser (a Lightningpanda), launched on first use
defer(fn)	Cleanup after what is running now
setupResult<T>()	What the setup module's setup() returned
testInfo()	{ file, name, titlePath } of the running test, or undefined
expect	Bun's expect

Hooks and tests can be async, or take a done callback. beforeEach hooks run outer first, afterEach inner first. Registering a test from inside a test or hook throws.

browser()

const lp = await browser();
await using page = await lp.page({ as: 'todos' });

The first call launches lightningpanda serve --fork with the setup module's options. Every later call returns the same instance. The runner closes it when the run ends, also on SIGINT, SIGTERM and SIGHUP. If the runner is killed outright, the browser's --parent-pid watch ends it.

Pages a test opens through browser() are closed when the test ends, including a test that timed out, so a hung test does not keep its forked process. Still write await using: it closes the page as soon as the test body is done.

defer()

defer(fn) runs fn once what is running now is done:

Called from	fn runs
A test, or its beforeEach/afterEach	After that test
A describe's (or file's) beforeAll/afterAll, or while it registers	After that describe's last test and its afterAll hooks

Later calls run first. The usual use is retiring a per-describe profile:

beforeAll(async () => {
  const lp = await browser();
  await lp.profile('orders', { base: 'app', url, setup });
  defer(() => lp.releaseProfile('orders'));
});

Scheduling

• All tests of all files share one queue; up to -j run at once. Tests start in file order (longest file first when --timings is given) and declaration order.
• A describe's beforeAll runs before its first test starts. Its other tests wait for it without taking a slot. Its afterAll runs after its last test.
• --max-concurrency n caps the tests of one file running at once. -j 8 --max-concurrency 1 runs like bun test --parallel=8 --max-concurrency=1, with one browser instead of eight.
• A test that fails a beforeAll of its scopes fails with HookError: beforeAll of describe '...' failed: ....

Serial tests

test.serial runs alone in its describe: after every test declared before it has finished, and before any test declared after it starts. Tests of other describes and files keep running beside it. describe.serial makes every test in it serial, so they run one at a time in order.

Use serial tests for steps that depend on each other's server state. Prefer independent tests that each seed what they need.

Timeouts

A test's timeout covers its beforeEach hooks, the test, its afterEach hooks and its deferred cleanups. A test that times out fails with TimeoutError: test timed out after N ms, and its pages are closed. Hooks without a timeout of their own use the default. Step timeouts inside page.run are separate (5 s per step by default); see Steps and checks.

Failures and exit code

Each test prints a line as it finishes, with failures in full. The summary lists every failure, then counts of passed, failed, skipped, todo, filtered out and not run tests.

The exit code is 1 when any test failed, a hook failed, a file failed to load, or queued tests never ran (after --bail or an interruption); 0 otherwise. An unhandled promise rejection or an uncaught exception during the run counts as a hook failure instead of ending the process. A spec that fails to load is reported and the other files still run.

--bail [n] stops starting tests after n failures. Running tests finish, open describes run their afterAll hooks and deferred cleanups, then the browser closes and teardown runs.

A first SIGINT (or SIGTERM, SIGHUP) stops the run the same way and prints stopping (the browser closes, teardown runs). A second one exits at once.

Sharding and timings

# record durations once (on a full run)
bunx lightningpanda e2e e2e --setup e2e/setup.ts --timings e2e/timings.json --update-timings

# split across 4 machines
bunx lightningpanda e2e e2e --setup e2e/setup.ts --timings e2e/timings.json --shard 1/4

--shard i/n sorts the files by their recorded duration, longest first, and gives each to the least loaded shard. Files without a recorded duration count as the average. Every shard computes the same split from the same file, so commit timings.json. A shard with no files prints No spec files in shard i/n. and exits 0.

--update-timings writes each file's busy time (its tests and hooks) merged into the existing file, sorted longest first. It does not write when -t or --only filtered tests, or when the run stopped early. With --timings and no --shard, files still start longest first.

JUnit

--junit junit.xml writes one testsuite per file and one testcase per test, with skipped and todo tests marked <skipped/>. Hook failures, load failures and teardown failures appear as test cases of their own, so CI shows them.

Screenshots

-u sets update mode for matchScreenshot steps (the same as LP_UPDATE_SNAPSHOTS=1): missing and mismatching baselines are written instead of compared. See Visual tests.

Environment

The runner sets NODE_ENV=test unless it is already set. $LIGHTNINGPANDA_BIN selects the browser binary when the setup module does not.

Migrating from bun test --parallel

1. Replace bun:test imports in spec files with @povio/lightning-panda/e2e. The functions take the same arguments.
2. Replace your own Lightningpanda.launch() or connect() with await browser(). Do not call lp.close(): the runner owns the browser.
3. Move global setup (servers, seeding, environment) into a setup module: setup() returns what specs need, browser(result) returns the server flags (--static, --local-storage, ...), teardown(result) stops it all.
4. Register shared profiles once (a memoized promise in a shared module), per-file or per-describe profiles in that describe's beforeAll with defer(() => lp.releaseProfile(name)).
5. Remove per-file globals that assumed a process per file: every file shares one module graph.
6. Run with the same concurrency first (-j 8 --max-concurrency 1 for bun test --parallel=8 --max-concurrency=1), then raise -j.

Porting ToS (2,158 cases) took the import changes and about 20 lines of harness code (PR #289). At equal concurrency on 62 files, bun test took 59.0 s and 59.3 s, lightningpanda e2e 51.9 s and 58.9 s; on the full suite, lightningpanda e2e -j 16 with one browser took 313 s against 380 s for bun test --parallel=8 with eight.

The runner has no watch mode, no coverage option and no HTML report.