Playwright review: tutorial, alternatives and pricing

Playwright keeps one persistent, two-way connection to each browser it launches and sends commands over it, rather than issuing a separate HTTP request per instruction as classic WebDriver does. That removes per-command overhead and lets the browser push events, such as requests, console messages and new pages, to the test as they happen. Chromium is driven through its own DevTools protocol; for Firefox and WebKit, Playwright ships builds carrying its own patches, which is why it cannot drive branded Firefox or Safari. The Python, Java and .NET packages share the same underlying implementation as the Node.js library, so the browser API is the same in every language even though the test runners differ.

Before a click, Playwright checks that the locator resolves to exactly one element, and that the element is visible, stable rather than animating, able to receive events because nothing covers it, enabled and not inert. Fill checks visibility, enabled and editable. It retries those checks until they pass or the action times out. Locators are strict, so a selector matching two elements fails rather than silently taking the first. Assertions such as toHaveText retry too, for 5 seconds by default inside a 30 second test timeout. Together these mean synchronisation is the default behaviour of the API rather than code you write, which is the main practical difference from Selenium.

A browser context is an isolated session with its own cookies, storage, cache and permissions, created in milliseconds inside an already running browser. Playwright Test gives every test a fresh context and page, so tests are isolated without paying browser start-up time each time. Several contexts in one test are how you simulate two users, and several pages in one context are how you handle tabs and popups. Contexts are also how authentication is reused: sign in once, save the context’s storage state to a file, and start every later context from that file already signed in.

Every Playwright version is built and tested against specific browser builds. Version 1.64 pins Chromium 156, Firefox 157 and WebKit 27.2, and was also tested against Chrome and Edge 155 stable. Since 1.57 the Chromium builds are Chrome for Testing: headed runs use the full browser and headless runs use chrome-headless-shell. npx playwright install downloads the set for your version, so the browser moves only when you upgrade Playwright, which makes runs reproducible. If you need the browser your users run, set channel to chrome or msedge to drive the installed branded build instead.

Playwright Test runs test files in parallel in worker processes, by default half the machine’s logical CPU cores. Tests inside a file run in order in one worker unless fullyParallel is on. Each worker starts its own browser, and a worker is replaced after a test fails so later tests start clean. Projects define the matrix: the same tests under different browsers, devices or settings, with dependencies so a setup project runs first. Across machines, --shard=2/4 runs the second quarter of the suite, and the blob reporter plus merge-reports stitches the shards back into one HTML report.

Group specs by feature, keep page objects in their own folder, and expose them to tests through fixtures rather than constructing them in every test. A fixture declared with test.extend is created only for tests that ask for it, can run setup before the test and cleanup after it, and keeps tests free of plumbing. Page objects should hold locators, which are lazy and re-evaluated on every use, rather than element handles, so they never go stale. Keep assertions in tests, except for the check that a page has finished loading, which belongs in the page object’s navigation method.

The documentation’s recommended pattern for a shared account is a setup project: a small test that signs in through the UI, waits until the app has set its cookies, and saves the context’s storage state to a file. Every other project lists setup as a dependency and loads that file, so every test starts signed in without repeating the login. Keep the file in playwright/.auth and add that folder to .gitignore, because it holds live session cookies. If tests change server-side state that other tests read, use one account per worker instead, as the authentication guide describes. The webServer option starts your application before the run and reuses a running one locally.

The states users complain about, empty lists, failed requests and slow responses, are the hardest to produce against a real backend. page.route makes each one a few lines: register the route before navigating so the first request is caught, then fulfil it with JSON, an error status or a delay. To synchronise with a real request rather than a stub, start waitForResponse before the click that triggers it, then await it afterwards; awaiting it first misses fast responses. Keep a smaller set of unstubbed journeys running against the real API, so stubs cannot drift from the contract unnoticed.

One CI machine runs as many workers as its cores allow; beyond that, shard. Each matrix job runs --shard=n/4, writes a blob report and uploads it, and a final job downloads every blob and merges them into one HTML report. Turn on fullyParallel so shards are balanced by test rather than by file, and keep fail-fast off so one failing shard does not cancel the others. The workflow installs browsers with their system dependencies on each job; on large suites, the official Playwright Docker image as a job container saves that time. Upload reports even when tests fail, since that is when you need them.

Set trace to on-first-retry, as the generated config does, so a failing test is retried with a full trace and passing runs record nothing. Open the merged HTML report, click the failed test and step through the trace before trying to reproduce anything locally. Use the new --shuffle option occasionally to flush out tests that depend on execution order, and test locks, added in 1.63, for tests that must not run alongside each other. When upgrading to 1.64, check four breaking changes: device descriptors now forward screen size, so window.screen and media queries see the emulated screen; elements inside hidden iframes count as hidden; JSX in test files follows tsconfig.json; and --update-snapshots=missing now passes the run.