@crawlee/browser-pool 4.0.0-beta.11 → 4.0.0-beta.111

package/puppeteer/puppeteer-plugin.js

@@ -1,11 +1,18 @@
 import { readFile } from 'node:fs/promises';
 import { BrowserPlugin } from '../abstract-classes/browser-plugin.js';
 import { anonymizeProxySugar } from '../anonymize-proxy.js';
-import { log } from '../logger.js';
 import { noop } from '../utils.js';
 import { PuppeteerController } from './puppeteer-controller.js';
 const PROXY_SERVER_ARG = '--proxy-server=';
 export class PuppeteerPlugin extends BrowserPlugin {
+    /** Pages share cookies/storage on the remote browser (Puppeteer defaults to non-incognito). */
+    useRemoteConnection(connection, parameters = {}) {
+        super.useRemoteConnection(connection, parameters);
+        if (!this.useIncognitoPages) {
+            this.log.info('Remote Puppeteer connection — pages will share cookies and storage on the remote ' +
+                'browser instance (useIncognitoPages defaults to false).');
+        }
+    }
     async _launch(launchContext) {
         let oldPuppeteerVersion = false;
         try {
@@ -17,58 +24,80 @@ export class PuppeteerPlugin extends BrowserPlugin {
         catch {
             // ignore
         }
-        const { launchOptions, userDataDir, useIncognitoPages, proxyUrl } = launchContext;
-        launchOptions.userDataDir = launchOptions.userDataDir ?? userDataDir;
-        if (launchOptions.headless === false) {
-            if (Array.isArray(launchOptions.args)) {
-                launchOptions.args.push('--disable-site-isolation-trials');
-            }
-            else {
-                launchOptions.args = ['--disable-site-isolation-trials'];
-            }
-        }
-        if (launchOptions.headless === true && oldPuppeteerVersion) {
-            launchOptions.headless = 'new';
-        }
+        const { useIncognitoPages, proxyUrl, ignoreProxyCertificate } = launchContext;
         let browser;
-        {
-            const [anonymizedProxyUrl, close] = await anonymizeProxySugar(proxyUrl);
-            if (proxyUrl) {
-                const proxyArg = `${PROXY_SERVER_ARG}${anonymizedProxyUrl ?? proxyUrl}`;
+        if (this.remoteConnection) {
+            browser = await this.connectToRemoteBrowser(launchContext, async (url) => {
+                const connectOptions = this.remoteConnectionParameters?.connectOptions ?? {};
+                this.log.info('Connecting to remote browser via connect (CDP).');
+                return this.library.connect({ ...connectOptions, browserWSEndpoint: url });
+            });
+        }
+        else {
+            const { launchOptions, userDataDir, experimentalContainers } = launchContext;
+            if (experimentalContainers) {
+                throw new Error('Experimental containers are only available with Playwright');
+            }
+            launchOptions.userDataDir = launchOptions.userDataDir ?? userDataDir;
+            if (launchOptions.headless === false) {
                 if (Array.isArray(launchOptions.args)) {
-                    launchOptions.args.push(proxyArg);
+                    launchOptions.args.push('--disable-site-isolation-trials');
                 }
                 else {
-                    launchOptions.args = [proxyArg];
+                    launchOptions.args = ['--disable-site-isolation-trials'];
                 }
             }
-            try {
-                browser = await this.library.launch(launchOptions);
-                if (anonymizedProxyUrl) {
-                    browser.on('disconnected', async () => {
-                        await close();
-                    });
-                }
+            if (launchOptions.headless === true && oldPuppeteerVersion) {
+                launchOptions.headless = 'new';
             }
-            catch (error) {
-                await close();
-                this._throwAugmentedLaunchError(error, launchContext.launchOptions?.executablePath, '`apify/actor-node-puppeteer-chrome`', "Try installing a browser, if it's missing, by running `npx @puppeteer/browsers install chromium --path [path]` and pointing `executablePath` to the downloaded executable (https://pptr.dev/browsers-api)");
+            {
+                const [anonymizedProxyUrl, close] = await anonymizeProxySugar(proxyUrl, undefined, undefined, {
+                    ignoreProxyCertificate: launchContext.ignoreProxyCertificate,
+                });
+                if (proxyUrl) {
+                    const proxyArg = `${PROXY_SERVER_ARG}${anonymizedProxyUrl ?? proxyUrl}`;
+                    if (Array.isArray(launchOptions.args)) {
+                        launchOptions.args.push(proxyArg);
+                    }
+                    else {
+                        launchOptions.args = [proxyArg];
+                    }
+                }
+                try {
+                    browser = await this.library.launch(launchOptions);
+                    if (anonymizedProxyUrl) {
+                        browser.on('disconnected', async () => {
+                            await close();
+                        });
+                    }
+                }
+                catch (error) {
+                    await close();
+                    this.throwAugmentedLaunchError(error, launchContext.launchOptions?.executablePath, '`apify/actor-node-puppeteer-chrome`', "Try installing a browser, if it's missing, by running `npx @puppeteer/browsers install chromium --path [path]` and pointing `executablePath` to the downloaded executable (https://pptr.dev/browsers-api)");
+                }
             }
         }
-        browser.on('targetcreated', async (target) => {
+        const targetCreatedHandler = async (target) => {
             try {
                 const page = await target.page();
                 if (page) {
                     page.on('error', (error) => {
-                        log.exception(error, 'Page crashed.');
+                        this.log.exception(error, 'Page crashed.');
                         page.close().catch(noop);
                     });
                 }
             }
             catch (error) {
-                log.exception(error, 'Failed to retrieve page from target.');
+                this.log.exception(error, 'Failed to retrieve page from target.');
             }
-        });
+        };
+        browser.on('targetcreated', targetCreatedHandler);
+        // Clean up the listener when a remote browser disconnects to prevent leaks
+        if (this.remoteConnection) {
+            browser.once('disconnected', () => {
+                browser.off('targetcreated', targetCreatedHandler);
+            });
+        }
         const boundMethods = [
             'newPage',
             'close',
@@ -78,6 +107,7 @@ export class PuppeteerPlugin extends BrowserPlugin {
             'version',
             'on',
             'process',
+            'pages',
         ].reduce((map, method) => {
             map[method] = browser[method]?.bind(browser);
             return map;
@@ -89,22 +119,30 @@ export class PuppeteerPlugin extends BrowserPlugin {
                     return async (...args) => {
                         let page;
                         if (useIncognitoPages) {
-                            const [anonymizedProxyUrl, close] = await anonymizeProxySugar(proxyUrl);
+                            // Skip proxy setup for remote connections — proxy is managed by the remote service.
+                            const effectiveProxyUrl = this.remoteConnection ? undefined : proxyUrl;
+                            const [anonymizedProxyUrl, close] = effectiveProxyUrl
+                                ? await anonymizeProxySugar(effectiveProxyUrl, undefined, undefined, {
+                                    ignoreProxyCertificate,
+                                })
+                                : [undefined, noop];
+                            const proxyServer = anonymizedProxyUrl ?? effectiveProxyUrl;
+                            const contextOptions = proxyServer ? { proxyServer } : {};
+                            const context = (await browser[method](contextOptions));
                             try {
-                                const context = (await browser[method]({
-                                    proxyServer: anonymizedProxyUrl ?? proxyUrl,
-                                }));
                                 page = await context.newPage(...args);
-                                if (anonymizedProxyUrl) {
-                                    page.on('close', async () => {
-                                        await close();
-                                    });
-                                }
                             }
                             catch (error) {
+                                await context.close().catch(noop);
                                 await close();
                                 throw error;
                             }
+                            page.once('close', async () => {
+                                if (anonymizedProxyUrl) {
+                                    await close();
+                                }
+                                await context.close().catch(noop);
+                            });
                         }
                         else {
                             page = await boundMethods.newPage(...args);
@@ -139,7 +177,7 @@ export class PuppeteerPlugin extends BrowserPlugin {
     createController() {
         return new PuppeteerController(this);
     }
-    async _addProxyToLaunchOptions(_launchContext) {
+    async addProxyToLaunchOptions(_launchContext) {
         /*
         // DO NOT USE YET! DOING SO DISABLES CACHE WHICH IS 50% PERFORMANCE HIT!
         launchContext.launchOptions ??= {};
@@ -166,8 +204,7 @@ export class PuppeteerPlugin extends BrowserPlugin {
         }
         */
     }
-    _isChromiumBasedBrowser(_launchContext) {
+    isChromiumBasedBrowser(_launchContext) {
         return true;
     }
 }
-//# sourceMappingURL=puppeteer-plugin.js.map

package/remote-browser-pool.d.ts

@@ -0,0 +1,164 @@
+import type { IBrowserPool, NewPageOptions, PageState } from '@crawlee/types';
+import type { BrowserPlugin } from './abstract-classes/browser-plugin.js';
+import { BrowserPool } from './browser-pool.js';
+import type { BrowserPoolHooks, BrowserPoolOptions } from './browser-pool.js';
+import { RemoteBrowserProvider } from './remote-browser-provider.js';
+/**
+ * The result of resolving a remote browser endpoint: the URL to connect to plus an optional opaque
+ * `context` object that is handed back to `release`.
+ */
+export interface ResolvedRemoteEndpoint {
+    /** The browser endpoint URL to connect to. */
+    url: string;
+    /** Opaque metadata passed back to `release()` — e.g. session IDs, API tokens. */
+    context?: Record<string, unknown>;
+}
+/**
+ * A remote browser endpoint: either a static URL string, or a function called once per browser launch
+ * that returns a URL (optionally with a `context` for `release`).
+ *
+ * The function receives the `proxyUrl` resolved by Crawlee's proxy configuration for the launch, so it
+ * can forward it to the remote service's proxy API.
+ */
+export type RemoteBrowserEndpoint = string | ((options?: {
+    proxyUrl?: string;
+}) => string | ResolvedRemoteEndpoint | Promise<string | ResolvedRemoteEndpoint>);
+/**
+ * The bridge a {@link RemoteBrowserPool} injects into a {@link BrowserPlugin} so the plugin can
+ * connect to a remote browser without owning any remote-session policy.
+ *
+ * The plugin only knows how to make the library-specific `connect()` call; everything else — resolving
+ * the endpoint, calling the user's `release()`, and guaranteeing release fires at most once — lives in
+ * the pool. The plugin calls {@link RemoteConnection.resolve|resolve} before connecting, stores the
+ * returned `token` on its launch context, and the controller later calls
+ * {@link RemoteConnection.release|release} with that token when the browser closes.
+ */
+export interface RemoteConnection {
+    /** Resolves the endpoint for a single browser launch. The `token` identifies the session for release. */
+    resolve(options?: {
+        proxyUrl?: string;
+    }): Promise<{
+        url: string;
+        token: number;
+    }>;
+    /** Releases the remote session for `token`. Idempotent — safe to call from both `close()` and `kill()`. */
+    release(token: number): Promise<void>;
+}
+/**
+ * Per-plugin remote connection parameters, passed to {@link BrowserPlugin.useRemoteConnection}.
+ * The endpoint is supplied per-launch via {@link RemoteConnection}; these are the static connect()
+ * parameters (protocol, headers, timeouts, …).
+ */
+export interface RemoteConnectionParameters {
+    /**
+     * Playwright only: which protocol to connect with. `'cdp'` uses `connectOverCDP()` (the default),
+// @ts-ignore optional peer dependency or compatibility with es2022
+     * `'playwright'` uses `connect()` (Playwright's own WebSocket protocol). Ignored by Puppeteer.
+     */
+    protocol?: 'cdp' | 'playwright';
+    /** Extra options forwarded to the library `connect()` / `connectOverCDP()` call (endpoint excluded). */
+    connectOptions?: Record<string, unknown>;
+}
+export interface RemoteBrowserPoolOptions {
+    /**
+     * The browser plugin(s) used to connect to the remote service — e.g. `new PlaywrightPlugin(playwright.chromium)`
+     * or `new PuppeteerPlugin(puppeteer)`. The pool configures them for remote connection; do not set a local
+     * `launchOptions` on them.
+     */
+    browserPlugins: BrowserPlugin[];
+    /**
+     * The remote browser endpoint: a static URL, a function returning one per launch, or a
+     * {@link RemoteBrowserProvider} instance encapsulating a session create/release lifecycle.
+     */
+    endpoint: RemoteBrowserEndpoint | RemoteBrowserProvider<any>;
+    /**
+     * Cleanup callback invoked when a browser closes, crashes, or the pool is destroyed. Receives the
+     * `context` returned by a function endpoint. Errors are caught and logged. Ignored when `endpoint`
+     * is a {@link RemoteBrowserProvider} (its own `release()` is used instead).
+     */
+    release?: (info: {
+        endpoint: string;
+        context?: Record<string, unknown>;
+    }) => unknown;
+    /**
+     * Maximum number of remote browsers open at once. When reached, {@link RemoteBrowserPool.newPage|newPage}
+     * waits for a browser to close before connecting a new one. Set it to your service's concurrent-session limit
+     * to avoid `429` errors. Defaults to the {@link RemoteBrowserProvider.maxOpenBrowsers|provider's value}, or
+     * `Infinity`.
+     */
+    maxOpenBrowsers?: number;
+    /** Static connect() parameters (Playwright protocol selection, headers, timeouts, …). */
+    connection?: RemoteConnectionParameters;
+    /** Extra {@link BrowserPool} options (lifecycle hooks, page limits, fingerprinting, …). */
+    browserPoolOptions?: Omit<BrowserPoolOptions, 'browserPlugins'> & BrowserPoolHooks<any, any, any>;
+    /** Fallback poll interval (ms) while waiting for a free browser slot. The wait is event-driven; this only bounds it. @default 500 */
+    slotPollIntervalMillis?: number;
+}
+/**
+ * The remote-connection configuration a browser crawler accepts on its `remoteBrowser` option. It is the
+ * {@link RemoteBrowserPoolOptions} a user supplies *minus* the parts the crawler provides itself — the
+ * `browserPlugins` (the crawler builds the correct one for its browser) and `browserPoolOptions` (taken from
+ * the crawler's own `browserPoolOptions`). This is what makes the crawler path both terse and mismatch-proof.
+ */
+export type CrawlerRemoteBrowserOptions = Omit<RemoteBrowserPoolOptions, 'browserPlugins' | 'browserPoolOptions'>;
+/**
+ * An {@link IBrowserPool} implementation for remote browser services.
+ *
+ * Unlike configuring a remote browser through a crawler's `launchContext`, this pool is the single owner
+ * of all remote-session concerns:
+ * - **endpoint resolution** — static URL, per-launch function, or {@link RemoteBrowserProvider};
+ * - **release lifecycle** — `release()` fires exactly once per session on close/crash/teardown (no leaks,
+ *   no double-release);
+ * - **concurrency** — {@link RemoteBrowserPoolOptions.maxOpenBrowsers|maxOpenBrowsers} is enforced inside
+ *   {@link RemoteBrowserPool.newPage|newPage}, which waits for a free slot rather than overshooting.
+ *
+ * The wrapped {@link BrowserPool} and its plugin only perform the library-specific `connect()` call.
+ *
+ * Pass an instance as the crawler's `browserPool` option:
+ *
+ * ```typescript
+ * import { PlaywrightPlugin, RemoteBrowserPool } from '@crawlee/browser-pool';
+ * import { PlaywrightCrawler } from 'crawlee';
+// @ts-ignore optional peer dependency or compatibility with es2022
+ * import playwright from 'playwright';
+ *
+ * const browserPool = new RemoteBrowserPool({
+ *     browserPlugins: [new PlaywrightPlugin(playwright.chromium)],
+ *     endpoint: 'wss://production-sfo.browserless.io?token=xxx',
+ *     maxOpenBrowsers: 2,
+ * });
+ *
+ * const crawler = new PlaywrightCrawler({ browserPool });
+ * ```
+ *
+ * @category Browser management
+ */
+export declare class RemoteBrowserPool<Page = unknown> implements IBrowserPool<Page> {
+    #private;
+    /** The wrapped pool that performs the remote connections and serves pages. */
+    readonly browserPool: BrowserPool;
+    constructor(options: RemoteBrowserPoolOptions);
+    /** Maximum number of remote browsers that may be open at the same time. */
+    get maxOpenBrowsers(): number;
+    set maxOpenBrowsers(value: number);
+    /**
+     * Opens a new page, waiting first until {@link RemoteBrowserPoolOptions.maxOpenBrowsers|maxOpenBrowsers}
+     * allows it (either a new browser slot is free, or an active browser still has page capacity).
+     */
+    newPage(options?: NewPageOptions): Promise<Page>;
+    closePage(page: Page, options?: {
+        error?: Error;
+    }): Promise<void>;
+    extractPageState(page: Page): Promise<PageState>;
+    injectPageState(page: Page, state: PageState): Promise<void>;
+    /** Closes all browsers, releases any still-open remote sessions, and tears down the wrapped pool. */
+    destroy(): Promise<void>;
+    /** Resolves once the wrapped pool can serve another page without exceeding `maxOpenBrowsers`. */
+    private waitForFreeSlot;
+    /**
+     * Resolves on the next browser-retired / page-closed event, or after `slotPollIntervalMillis`. All
+     * concurrently-waiting `newPage` calls share a single promise (and a single pair of event listeners)
+     * per tick, so a fleet of saturated callers doesn't fan out into N listener pairs on the pool.
+     */
+    private nextCapacityChange;
+}

package/remote-browser-pool.js

@@ -0,0 +1,191 @@
+import { serviceLocator } from '@crawlee/core';
+import { BrowserPool } from './browser-pool.js';
+import { RemoteBrowserProvider } from './remote-browser-provider.js';
+/**
+ * Owns the lifecycle of remote browser sessions for a single {@link RemoteBrowserPool}: endpoint
+ * resolution, the user's `release()` callback, and a release-at-most-once guarantee. Implements
+ * {@link RemoteConnection} so it can be injected into a plugin.
+ */
+class RemoteSessionRegistry {
+    #sessions = new Map();
+    #nextToken = 0;
+    #endpoint;
+    #onRelease;
+    #log;
+    constructor(endpoint, onRelease, log) {
+        this.#endpoint = endpoint;
+        this.#onRelease = onRelease;
+        this.#log = log;
+    }
+    async resolve(options) {
+        const resolved = typeof this.#endpoint === 'function' ? await this.#endpoint(options) : this.#endpoint;
+        let result;
+        if (typeof resolved === 'string') {
+            if (!resolved)
+                throw new Error('Remote browser endpoint resolved to an empty string.');
+            result = { url: resolved };
+        }
+        else if (!resolved?.url) {
+            throw new Error("Remote browser endpoint() must return a URL string or an object with a non-empty 'url'.");
+        }
+        else {
+            result = resolved;
+        }
+        const token = this.#nextToken++;
+        this.#sessions.set(token, { url: result.url, context: result.context, released: false });
+        return { url: result.url, token };
+    }
+    async release(token) {
+        const session = this.#sessions.get(token);
+        // Release at most once per session — guards a close()/teardown race (the `released` flag is set
+        // synchronously before the awaited onRelease, so releaseAll() can't double-fire an in-flight release).
+        if (!session || session.released)
+            return;
+        session.released = true;
+        try {
+            await this.#onRelease?.({ endpoint: session.url, context: session.context });
+        }
+        catch (err) {
+            this.#log.warning('Remote browser release() failed.', { error: err?.message });
+        }
+        finally {
+            this.#sessions.delete(token);
+        }
+    }
+    /** Releases every session that is still open. Called on pool teardown so no remote session leaks. */
+    async releaseAll() {
+        await Promise.all([...this.#sessions.keys()].map(async (token) => this.release(token)));
+    }
+}
+/**
+ * An {@link IBrowserPool} implementation for remote browser services.
+ *
+ * Unlike configuring a remote browser through a crawler's `launchContext`, this pool is the single owner
+ * of all remote-session concerns:
+ * - **endpoint resolution** — static URL, per-launch function, or {@link RemoteBrowserProvider};
+ * - **release lifecycle** — `release()` fires exactly once per session on close/crash/teardown (no leaks,
+ *   no double-release);
+ * - **concurrency** — {@link RemoteBrowserPoolOptions.maxOpenBrowsers|maxOpenBrowsers} is enforced inside
+ *   {@link RemoteBrowserPool.newPage|newPage}, which waits for a free slot rather than overshooting.
+ *
+ * The wrapped {@link BrowserPool} and its plugin only perform the library-specific `connect()` call.
+ *
+ * Pass an instance as the crawler's `browserPool` option:
+ *
+ * ```typescript
+ * import { PlaywrightPlugin, RemoteBrowserPool } from '@crawlee/browser-pool';
+ * import { PlaywrightCrawler } from 'crawlee';
+ * import playwright from 'playwright';
+ *
+ * const browserPool = new RemoteBrowserPool({
+ *     browserPlugins: [new PlaywrightPlugin(playwright.chromium)],
+ *     endpoint: 'wss://production-sfo.browserless.io?token=xxx',
+ *     maxOpenBrowsers: 2,
+ * });
+ *
+ * const crawler = new PlaywrightCrawler({ browserPool });
+ * ```
+ *
+ * @category Browser management
+ */
+export class RemoteBrowserPool {
+    /** The wrapped pool that performs the remote connections and serves pages. */
+    browserPool;
+    /** The wrapped pool viewed through the {@link IBrowserPool} contract (the bare type widens pages to `never`). */
+    #pool;
+    #registry;
+    #slotPollIntervalMillis;
+    #log;
+    /** Shared by all `newPage` callers waiting for a free slot, so they don't each register their own listeners. */
+    #capacityChange;
+    constructor(options) {
+        const { browserPlugins, endpoint, release, maxOpenBrowsers, connection = {}, browserPoolOptions = {}, slotPollIntervalMillis = 500, } = options;
+        this.#log = serviceLocator.getLogger().child({ prefix: 'RemoteBrowserPool' });
+        this.#slotPollIntervalMillis = slotPollIntervalMillis;
+        // A RemoteBrowserProvider carries its own endpoint, release, and maxOpenBrowsers.
+        const provider = endpoint instanceof RemoteBrowserProvider ? endpoint : undefined;
+        const resolvedEndpoint = provider
+            ? (opts) => provider.connect(opts)
+            : endpoint;
+        const resolvedRelease = provider
+            ? ({ context }) => provider.release(context)
+            : release;
+        const resolvedMax = maxOpenBrowsers ?? provider?.maxOpenBrowsers;
+        this.#registry = new RemoteSessionRegistry(resolvedEndpoint, resolvedRelease, this.#log);
+        // Wire every plugin for remote connection.
+        for (const plugin of browserPlugins) {
+            plugin.useRemoteConnection(this.#registry, connection);
+        }
+        this.browserPool = new BrowserPool({ ...browserPoolOptions, browserPlugins });
+        this.#pool = this.browserPool;
+        // Release a browser's remote session once it closes. The registry dedupes (close() schedules a delayed
+        // kill(), so BROWSER_CLOSED can fire twice), and destroy()'s releaseAll() backstops any that never close.
+        this.browserPool.on("browserLaunched" /* BROWSER_POOL_EVENTS.BROWSER_LAUNCHED */, (controller) => {
+            controller.once("browserClosed" /* BROWSER_CONTROLLER_EVENTS.BROWSER_CLOSED */, () => {
+                const token = controller.launchContext.remoteToken;
+                if (token !== undefined)
+                    void this.#registry.release(token);
+            });
+        });
+        if (resolvedMax !== undefined) {
+            this.browserPool.maxOpenBrowsers = resolvedMax;
+        }
+    }
+    /** Maximum number of remote browsers that may be open at the same time. */
+    get maxOpenBrowsers() {
+        return this.browserPool.maxOpenBrowsers;
+    }
+    set maxOpenBrowsers(value) {
+        this.browserPool.maxOpenBrowsers = value;
+    }
+    /**
+     * Opens a new page, waiting first until {@link RemoteBrowserPoolOptions.maxOpenBrowsers|maxOpenBrowsers}
+     * allows it (either a new browser slot is free, or an active browser still has page capacity).
+     */
+    async newPage(options) {
+        await this.waitForFreeSlot();
+        return this.#pool.newPage(options);
+    }
+    async closePage(page, options) {
+        return this.#pool.closePage(page, options);
+    }
+    async extractPageState(page) {
+        return this.#pool.extractPageState(page);
+    }
+    async injectPageState(page, state) {
+        return this.#pool.injectPageState(page, state);
+    }
+    /** Closes all browsers, releases any still-open remote sessions, and tears down the wrapped pool. */
+    async destroy() {
+        await this.browserPool.destroy();
+        // Backstop: release any sessions whose browser never emitted a close (e.g. dropped on teardown).
+        await this.#registry.releaseAll();
+    }
+    /** Resolves once the wrapped pool can serve another page without exceeding `maxOpenBrowsers`. */
+    async waitForFreeSlot() {
+        while (!this.browserPool.hasFreeBrowserSlot() && !this.browserPool.hasActiveBrowserWithFreeCapacity()) {
+            await this.nextCapacityChange();
+        }
+    }
+    /**
+     * Resolves on the next browser-retired / page-closed event, or after `slotPollIntervalMillis`. All
+     * concurrently-waiting `newPage` calls share a single promise (and a single pair of event listeners)
+     * per tick, so a fleet of saturated callers doesn't fan out into N listener pairs on the pool.
+     */
+    nextCapacityChange() {
+        this.#capacityChange ??= new Promise((resolve) => {
+            const done = () => {
+                clearTimeout(timer);
+                this.browserPool.off("browserRetired" /* BROWSER_POOL_EVENTS.BROWSER_RETIRED */, done);
+                this.browserPool.off("pageClosed" /* BROWSER_POOL_EVENTS.PAGE_CLOSED */, done);
+                this.#capacityChange = undefined;
+                resolve();
+            };
+            const timer = setTimeout(done, this.#slotPollIntervalMillis);
+            timer.unref?.();
+            this.browserPool.once("browserRetired" /* BROWSER_POOL_EVENTS.BROWSER_RETIRED */, done);
+            this.browserPool.once("pageClosed" /* BROWSER_POOL_EVENTS.PAGE_CLOSED */, done);
+        });
+        return this.#capacityChange;
+    }
+}

package/remote-browser-provider.d.ts

@@ -0,0 +1,83 @@
+/**
+ * Abstract base class for remote browser service providers.
+ *
+ * Implement this class to encapsulate the lifecycle of a remote browser session
+ * (creation, connection URL resolution, and cleanup). {@link RemoteBrowserPool}
+ * calls {@link connect} once per browser launch and {@link release} when the browser
+ * closes, crashes, the pool is destroyed, or the connection fails during launch.
+ *
+ * Pass the provider instance as the `endpoint` of a {@link RemoteBrowserPool}, then
+ * hand the pool to a crawler via its `browserPool` option:
+ *
+ * ```typescript
+ * const browserPool = new RemoteBrowserPool({
+ *     browserPlugins: [new PlaywrightPlugin(playwright.chromium)],
+ *     endpoint: new MyProvider(),
+ * });
+ *
+ * const crawler = new PlaywrightCrawler({ browserPool });
+ * ```
+ *
+ * **Example — simple static endpoint (e.g. Browserless):**
+ * ```typescript
+ * class BrowserlessProvider extends RemoteBrowserProvider {
+ *     maxOpenBrowsers = 2; // respect the service's concurrent session limit
+ *
+ *     async connect() {
+ *         return { url: `wss://production-sfo.browserless.io?token=${token}` };
+ *     }
+ * }
+ * ```
+ *
+ * **Example — session lifecycle with concurrency limit (e.g. Browserbase):**
+ * ```typescript
+ * class BrowserbaseProvider extends RemoteBrowserProvider<{ id: string }> {
+ *     maxOpenBrowsers = 2; // respect the service's concurrent session limit
+ *
+ *     async connect({ proxyUrl } = {}) {
+ *         const session = await createSession(apiKey, projectId, {
+ *             proxies: proxyUrl ? [{ type: 'external', server: proxyUrl }] : undefined,
+ *         });
+ *         return { url: session.connectUrl, context: { id: session.id } };
+ *     }
+ *
+ *     async release(context: { id: string }) {
+ *         await releaseSession(apiKey, context.id);
+ *     }
+ * }
+ * ```
+ */
+export declare abstract class RemoteBrowserProvider<TContext extends Record<string, unknown> = Record<string, unknown>> {
+    /**
+     * Maximum number of browsers that can be open at the same time.
+     * Set this to your remote service's concurrent session limit to avoid 429 errors.
+     */
+    maxOpenBrowsers?: number;
+    /**
+     * Called once per browser launch. Return the WebSocket/CDP endpoint URL
+     * and an optional `context` object that will be passed back to {@link release}.
+     *
+     * @param options.proxyUrl - The proxy URL resolved by Crawlee's proxy configuration
+     * for this browser session. Pass it to your remote service's proxy API if supported.
+     */
+    abstract connect(options?: {
+        proxyUrl?: string;
+    }): Promise<{
+        url: string;
+        context?: TContext;
+    }> | {
+        url: string;
+        context?: TContext;
+    };
+    /**
+     * Called when the browser closes, crashes, the pool is destroyed, or the
+     * connection fails right after {@link connect} succeeds.
+     * Override this to clean up remote sessions, release API resources, etc.
+     *
+     * Errors thrown here are caught and logged as warnings — they never crash the crawler.
+     * Safe to assume this is called at most once per {@link connect} call.
+     *
+     * @param _context The same `context` object returned by {@link connect}.
+     */
+    release(_context: TContext): Promise<void>;
+}