@crawlee/playwright 4.0.0-beta.2 → 4.0.0-beta.200

package/internals/adaptive-playwright-crawler.js

@@ -1,46 +1,30 @@
+import { setTimeout as delay } from 'node:timers/promises';
+import { isDeepStrictEqual } from 'node:util';
+import { BasicCrawler, ContextPipeline, RequestHandlerError, resolveBaseUrlForEnqueueLinksFiltering, Router, Statistics, } from '@crawlee/basic';
 import { extractUrlsFromPage } from '@crawlee/browser';
-import { Configuration, RequestHandlerResult, Router, Statistics, withCheckedStorageAccess } from '@crawlee/core';
-import { extractUrlsFromCheerio } from '@crawlee/utils';
-import { load } from 'cheerio';
-import isEqual from 'lodash.isequal';
+import { CheerioCrawler } from '@crawlee/cheerio';
+import { createStorageTransaction, EnqueueStrategy, OwnedOrInjected } from '@crawlee/core';
+import { extractUrlsFromCheerio, parseArgument } from '@crawlee/utils/internal';
+import { z } from 'zod';
 import { addTimeoutToPromise } from '@apify/timeout';
 import { PlaywrightCrawler } from './playwright-crawler.js';
-import { RenderingTypePredictor } from './utils/rendering-type-prediction.js';
-class AdaptivePlaywrightCrawlerStatistics extends Statistics {
-    state = null; // this needs to be assigned for a valid override, but the initialization is done by a reset() call from the parent constructor
-    constructor(options = {}) {
-        super(options);
-        this.reset();
-    }
-    reset() {
-        super.reset();
-        this.state.httpOnlyRequestHandlerRuns = 0;
-        this.state.browserRequestHandlerRuns = 0;
-        this.state.renderingTypeMispredictions = 0;
-    }
-    async _maybeLoadStatistics() {
-        await super._maybeLoadStatistics();
-        const savedState = await this.keyValueStore?.getValue(this.persistStateKey);
-        if (!savedState) {
-            return;
-        }
-        this.state.httpOnlyRequestHandlerRuns = savedState.httpOnlyRequestHandlerRuns;
-        this.state.browserRequestHandlerRuns = savedState.browserRequestHandlerRuns;
-        this.state.renderingTypeMispredictions = savedState.renderingTypeMispredictions;
-    }
-    trackHttpOnlyRequestHandlerRun() {
-        this.state.httpOnlyRequestHandlerRuns ??= 0;
-        this.state.httpOnlyRequestHandlerRuns += 1;
-    }
-    trackBrowserRequestHandlerRun() {
-        this.state.browserRequestHandlerRuns ??= 0;
-        this.state.browserRequestHandlerRuns += 1;
-    }
-    trackRenderingTypeMisprediction() {
-        this.state.renderingTypeMispredictions ??= 0;
-        this.state.renderingTypeMispredictions += 1;
-    }
-}
+import { RenderingTypePredictor, } from './utils/rendering-type-prediction.js';
+const adaptiveStatisticStateSchema = z.object({
+    /** How many requests were handled by the HTTP-only request handler. */
+    httpOnlyRequestHandlerRuns: z.number().default(0),
+    /** How many requests were handled in a browser. */
+    browserRequestHandlerRuns: z.number().default(0),
+    /** How many times the HTTP-only handler produced a result the `resultChecker` rejected. */
+    renderingTypeMispredictions: z.number().default(0),
+});
+/**
+ * The {@link AdaptivePlaywrightCrawlerStatisticState} fields as a {@link Statistics} state extension, defaults
+ * and all. A {@link Statistics} instance to be injected into an {@link AdaptivePlaywrightCrawler} has to carry
+ * them - `deserialize.extend()` your own fields onto this one and pass the result as `stateExtension`.
+ */
+export const adaptivePlaywrightCrawlerStatisticState = {
+    deserialize: adaptiveStatisticStateSchema,
+};
 const proxyLogMethods = [
     'error',
     'exception',
@@ -80,274 +64,507 @@ const proxyLogMethods = [
  *
  * @experimental
  */
-export class AdaptivePlaywrightCrawler extends PlaywrightCrawler {
-    config;
-    adaptiveRequestHandler;
-    renderingTypePredictor;
-    resultChecker;
-    resultComparator;
-    preventDirectStorageAccess;
+export class AdaptivePlaywrightCrawler extends BasicCrawler {
+    #renderingTypePredictor;
+    #resultChecker;
+    #shouldPropagateError;
+    #resultComparator;
+    #staticContextPipeline;
+    #browserContextPipeline;
+    #individualRequestHandlerTimeoutMillis;
+    /**
+     * The write policy of the per-attempt transactions. Defaults the request queue to `deferred`:
+     * a discarded attempt's enqueues must never reach the queue.
+     */
+    #attemptWritePolicy;
+    /** Owns the browser pool this crawler's runs use, so its per-run resources are released with ours. */
+    #browserCrawler;
+    /** Nothing of its state is per-run, but it owns a session pool that outlives one. */
+    #staticCrawler;
+    /**
+     * In-flight rendering type detections, plus the pending results of an asynchronous `storeResult`.
+     */
+    #activeDetections = new Set();
     /**
-     * Default {@link Router} instance that will be used if we don't specify any {@link AdaptivePlaywrightCrawlerOptions.requestHandler|`requestHandler`}.
-     * See {@link Router.addHandler|`router.addHandler()`} and {@link Router.addDefaultHandler|`router.addDefaultHandler()`}.
+     * Set once `teardown()` starts, so that requests still in the pool stop opening new detections.
      */
-    // @ts-ignore
-    router = Router.create();
-    constructor(options = {}, config = Configuration.getGlobalConfig()) {
-        const { requestHandler, renderingTypeDetectionRatio = 0.1, renderingTypePredictor, resultChecker, resultComparator, statisticsOptions, preventDirectStorageAccess = true, ...rest } = options;
-        super(rest, config);
-        this.config = config;
-        this.adaptiveRequestHandler = requestHandler ?? this.router;
-        this.renderingTypePredictor =
-            renderingTypePredictor ?? new RenderingTypePredictor({ detectionRatio: renderingTypeDetectionRatio });
-        this.resultChecker = resultChecker ?? (() => true);
+    #shutDown = false;
+    constructor(options = {}) {
+        const { requestHandler, renderingTypeDetectionRatio = 0.1, renderingTypePredictor, resultChecker, shouldPropagateError, resultComparator, statistics, requestHandlerTimeoutSecs = 60, errorHandler, failedRequestHandler, preNavigationHooks = [], postNavigationHooks = [], extendContext, transactionalStorage, launchContext, headless, browserPool, remoteBrowser, ...rest } = options;
+        // The user's value is replaced by `false` in the `super` call below — validate it separately,
+        // wrapped in an object so the error still names the field.
+        parseArgument({ transactionalStorage }, z.object({ transactionalStorage: BasicCrawler.optionsShape.transactionalStorage }), 'AdaptivePlaywrightCrawlerOptions');
+        // The extra fields are only tracked if the injected instance was built with them - the types enforce that,
+        // but plain JS callers would otherwise silently increment `undefined` into a sticky `NaN`. Extend
+        // `adaptivePlaywrightCrawlerStatisticState` to satisfy this.
+        if (statistics !== undefined) {
+            parseArgument(statistics.state, z.object({
+                httpOnlyRequestHandlerRuns: z.number(),
+                browserRequestHandlerRuns: z.number(),
+                renderingTypeMispredictions: z.number(),
+            }), 'statistics.state');
+        }
+        // Per-attempt buffering is load-bearing here: the handler runs up to twice per request and the
+        // losing attempt's writes must be discardable.
+        if (transactionalStorage === false) {
+            throw new Error('AdaptivePlaywrightCrawler requires transactional storage - it runs the request handler ' +
+                'multiple times per request and must be able to discard the storage writes of losing ' +
+                'attempts. `transactionalStorage: false` is therefore not supported; a write policy ' +
+                'object is accepted and forwarded to the per-attempt transactions.');
+        }
+        super({
+            ...rest,
+            errorHandler,
+            failedRequestHandler,
+            requestHandler,
+            requestHandlerTimeoutSecs,
+            // The base would build a `Statistics` without the adaptive fields, so provide a default that has them.
+            // The cast covers a `StatisticStateExtension` that adds further fields - those can only come from an
+            // injected instance, in which case this default is never built.
+            statistics: statistics ??
+                new Statistics({
+                    logMessage: `${AdaptivePlaywrightCrawler.name} request statistics:`,
+                    stateExtension: adaptivePlaywrightCrawlerStatisticState,
+                }),
+            contextPipelineBuilder: () => this.#buildContextPipeline(),
+            // The base crawler must not wrap requests in a transaction of its own - this crawler opens
+            // one per request handler attempt in `crawlOne` instead, forwarding the write policy of the
+            // user-facing option (validated above) to those.
+            transactionalStorage: false,
+        });
+        this.#individualRequestHandlerTimeoutMillis = requestHandlerTimeoutSecs * 1000;
+        // `renderingTypeDetectionRatio` only configures the default predictor - an injected one brings its own
+        // detection ratio (and its own state), so the option is ignored in that case.
+        this.#renderingTypePredictor = OwnedOrInjected.resolve(renderingTypePredictor, () => new RenderingTypePredictor({ detectionRatio: renderingTypeDetectionRatio }));
+        this.#attemptWritePolicy = {
+            requestQueue: 'deferred',
+            ...(typeof transactionalStorage === 'object' ? transactionalStorage : {}),
+        };
+        this.#resultChecker = resultChecker ?? (() => true);
+        this.#shouldPropagateError = shouldPropagateError ?? (() => false);
         if (resultComparator !== undefined) {
-            this.resultComparator = resultComparator;
+            this.#resultComparator = resultComparator;
         }
         else if (resultChecker !== undefined) {
-            this.resultComparator = (resultA, resultB) => this.resultChecker(resultA) && this.resultChecker(resultB);
+            this.#resultComparator = (resultA, resultB) => this.#resultChecker(resultA) && this.#resultChecker(resultB);
         }
         else {
-            this.resultComparator = (resultA, resultB) => {
+            this.#resultComparator = (resultA, resultB) => {
                 return (resultA.datasetItems.length === resultB.datasetItems.length &&
                     resultA.datasetItems.every((itemA, i) => {
                         const itemB = resultB.datasetItems[i];
-                        return isEqual(itemA, itemB);
+                        return isDeepStrictEqual(itemA, itemB);
                     }));
             };
         }
-        this.stats = new AdaptivePlaywrightCrawlerStatistics({
-            logMessage: `${this.log.getOptions().prefix} request statistics:`,
-            config,
-            ...statisticsOptions,
+        // `extendContext` is forwarded to the inner crawlers, which run it *before* navigation (see
+        // `BasicCrawler`), keeping the behavior consistent with the non-adaptive crawlers: the
+        // extension is visible to the pre/post-navigation hooks and the request handler, but cannot
+        // access navigation-dependent members (`page`, `response`, `$`, ...).
+        //
+        // The adaptive hooks target a subset context (`AdaptiveHookContext`); the casts to the inner
+        // crawlers' `PlaywrightHook` type relax that nominal difference. The `ContextPipeline` merges
+        // each hook's overrides at runtime regardless of the static type.
+        const staticCrawler = new CheerioCrawler({
+            ...rest,
+            statistics: new Statistics({ persistenceOptions: { enable: false } }),
+            preNavigationHooks,
+            postNavigationHooks,
+            extendContext,
+        });
+        const browserCrawler = new PlaywrightCrawler({
+            ...rest,
+            statistics: new Statistics({ persistenceOptions: { enable: false } }),
+            preNavigationHooks: preNavigationHooks,
+            postNavigationHooks: postNavigationHooks,
+            extendContext,
+            launchContext,
+            headless,
+            browserPool,
+            remoteBrowser,
         });
-        this.preventDirectStorageAccess = preventDirectStorageAccess;
+        this.#staticCrawler = staticCrawler;
+        this.#browserCrawler = browserCrawler;
+        this.#staticContextPipeline = staticCrawler.contextPipeline.compose(this.adaptCheerioContext.bind(this));
+        this.#browserContextPipeline = browserCrawler.contextPipeline.compose(this.adaptPlaywrightContext.bind(this));
     }
-    async _runRequestHandler(crawlingContext) {
-        const renderingTypePrediction = this.renderingTypePredictor.predict(crawlingContext.request);
-        const shouldDetectRenderingType = Math.random() < renderingTypePrediction.detectionProbabilityRecommendation;
-        if (!shouldDetectRenderingType) {
-            crawlingContext.log.debug(`Predicted rendering type ${renderingTypePrediction.renderingType} for ${crawlingContext.request.url}`);
-        }
-        if (renderingTypePrediction.renderingType === 'static' && !shouldDetectRenderingType) {
-            crawlingContext.log.debug(`Running HTTP-only request handler for ${crawlingContext.request.url}`);
-            this.stats.trackHttpOnlyRequestHandlerRun();
-            const plainHTTPRun = await this.runRequestHandlerWithPlainHTTP(crawlingContext);
-            if (plainHTTPRun.ok && this.resultChecker(plainHTTPRun.result)) {
-                crawlingContext.log.debug(`HTTP-only request handler succeeded for ${crawlingContext.request.url}`);
-                plainHTTPRun.logs?.forEach(([log, method, ...args]) => log[method](...args));
-                await this.commitResult(crawlingContext, plainHTTPRun.result);
-                return;
-            }
-            if (!plainHTTPRun.ok) {
-                crawlingContext.log.exception(plainHTTPRun.error, `HTTP-only request handler failed for ${crawlingContext.request.url}`);
-            }
-            else {
-                crawlingContext.log.warning(`HTTP-only request handler returned a suspicious result for ${crawlingContext.request.url}`);
-                this.stats.trackRenderingTypeMisprediction();
-            }
-        }
-        crawlingContext.log.debug(`Running browser request handler for ${crawlingContext.request.url}`);
-        this.stats.trackBrowserRequestHandlerRun();
-        // Run the request handler in a browser. The copy of the crawler state is kept so that we can perform
-        // a rendering type detection if necessary. Without this measure, the HTTP request handler would run
-        // under different conditions, which could change its behavior. Changes done to the crawler state by
-        // the HTTP request handler will not be committed to the actual storage.
-        const { result: browserRun, initialStateCopy } = await this.runRequestHandlerInBrowser(crawlingContext);
-        if (!browserRun.ok) {
-            throw browserRun.error;
-        }
-        await this.commitResult(crawlingContext, browserRun.result);
-        if (shouldDetectRenderingType) {
-            crawlingContext.log.debug(`Detecting rendering type for ${crawlingContext.request.url}`);
-            const plainHTTPRun = await this.runRequestHandlerWithPlainHTTP(crawlingContext, initialStateCopy);
-            const detectionResult = (() => {
-                if (!plainHTTPRun.ok) {
-                    return 'clientOnly';
-                }
-                if (this.resultComparator(plainHTTPRun.result, browserRun.result)) {
-                    return 'static';
-                }
-                return 'clientOnly';
-            })();
-            crawlingContext.log.debug(`Detected rendering type ${detectionResult} for ${crawlingContext.request.url}`);
-            this.renderingTypePredictor.storeResult(crawlingContext.request, detectionResult);
-        }
+    async init() {
+        // A crawler can be run again after a teardown.
+        this.#shutDown = false;
+        // Only the predictor we built ourselves is ours to initialize - an injected one is borrowed, so its
+        // lifecycle (including restoring persisted state) stays with whoever created it.
+        await this.#renderingTypePredictor.ifOwned((predictor) => predictor.initialize());
+        return await super.init();
     }
-    async commitResult(crawlingContext, { calls, keyValueStoreChanges }) {
-        await Promise.all([
-            ...calls.pushData.map(async (params) => crawlingContext.pushData(...params)),
-            ...calls.enqueueLinks.map(async (params) => await crawlingContext.enqueueLinks(...params)),
-            ...calls.addRequests.map(async (params) => crawlingContext.addRequests(...params)),
-            ...Object.entries(keyValueStoreChanges).map(async ([storeIdOrName, changes]) => {
-                const store = await crawlingContext.getKeyValueStore(storeIdOrName);
-                await Promise.all(Object.entries(changes).map(async ([key, { changedValue, options }]) => store.setValue(key, changedValue, options)));
-            }),
-        ]);
+    #buildContextPipeline() {
+        const errorMessage = (prop) => `The \`${prop}\` property is not available on the outer context pipeline of AdaptivePlaywrightCrawler - it is provided by the inner (static/browser) pipelines`;
+        return ContextPipeline.create().compose(async ({ request }) => ({
+            get request() {
+                return request;
+            },
+            get response() {
+                throw new Error(errorMessage('response'));
+            },
+            get page() {
+                throw new Error(errorMessage('page'));
+            },
+            get querySelector() {
+                throw new Error(errorMessage('querySelector'));
+            },
+            get querySelectorAll() {
+                throw new Error(errorMessage('querySelectorAll'));
+            },
+            get waitForSelector() {
+                throw new Error(errorMessage('waitForSelector'));
+            },
+            get parseWithCheerio() {
+                throw new Error(errorMessage('parseWithCheerio'));
+            },
+            get enqueueLinks() {
+                throw new Error(errorMessage('enqueueLinks'));
+            },
+        }));
     }
-    allowStorageAccess(func) {
-        return async (...args) => withCheckedStorageAccess(() => { }, async () => func(...args));
+    async adaptCheerioContext(cheerioContext) {
+        return {
+            get page() {
+                throw new Error('Page object was used in HTTP-only request handler');
+            },
+            async querySelector(selector) {
+                return cheerioContext.$(selector).first();
+            },
+            async querySelectorAll(selector) {
+                return cheerioContext.$(selector);
+            },
+            enqueueLinks: async (options = {}) => {
+                const urls = extractUrlsFromCheerio(cheerioContext.$, options.selector, options.baseUrl ?? cheerioContext.request.loadedUrl);
+                return (await this.enqueueLinks(urls, options, cheerioContext.request));
+            },
+            response: cheerioContext.response,
+        };
     }
-    async runRequestHandlerInBrowser(crawlingContext) {
-        const result = new RequestHandlerResult(this.config, AdaptivePlaywrightCrawler.CRAWLEE_STATE_KEY);
-        let initialStateCopy;
+    async adaptPlaywrightContext(playwrightContext) {
+        // Capture the original response to avoid infinite recursion when the getter is copied to the context
+        const originalResponse = playwrightContext.response;
+        return {
+            response: new Response(Uint8Array.from(await originalResponse.body()), {
+                headers: originalResponse.headers(),
+                status: originalResponse.status(),
+                statusText: originalResponse.statusText(),
+            }),
+            async querySelector(selector, timeoutMs = 5000) {
+                const locator = playwrightContext.page.locator(selector).first();
+                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
+                const $ = await playwrightContext.parseWithCheerio();
+                return $(selector).first();
+            },
+            async querySelectorAll(selector, timeoutMs = 5000) {
+                const locator = playwrightContext.page.locator(selector).first();
+                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
+                const $ = await playwrightContext.parseWithCheerio();
+                return $(selector);
+            },
+            enqueueLinks: async (options = {}, timeoutMs = 5000) => {
+                // TODO consider using `context.parseWithCheerio` to make this universal and avoid code duplication
+                const selector = options.selector ?? 'a';
+                const locator = playwrightContext.page.locator(selector).first();
+                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
+                const urls = await extractUrlsFromPage(playwrightContext.page, selector, options.baseUrl ?? playwrightContext.request.loadedUrl);
+                return (await this.enqueueLinks(urls, options, playwrightContext.request));
+            },
+        };
+    }
+    /**
+     * Runs one request handler attempt inside its own {@link StorageTransaction}, wrapping the inner
+     * (static or browser) context pipeline. The transaction is pushed to `transactions` *at creation
+     * time, before the `try`* - the `ok: false` branch of the returned {@link Result} carries no
+     * result, and failed attempts are routine here. The caller owns the outcome and disposal.
+     */
+    async crawlOne(renderingType, context, useStateFunction, transactions) {
+        const transaction = createStorageTransaction({
+            policy: this.#attemptWritePolicy,
+            commitTimeoutMillis: this.internalTimeoutMillis,
+        });
+        transactions.push(transaction);
+        const logs = [];
+        const deferredCleanup = [];
+        const attemptBoundContextHelpers = {
+            useState: useStateFunction,
+            log: this.createLogProxy(context.log, logs),
+            registerDeferredCleanup: (cleanup) => deferredCleanup.push(cleanup),
+        };
+        const subCrawlerContext = Object.defineProperties({}, Object.getOwnPropertyDescriptors(context));
+        // Mark attempt-bound helpers as non-configurable so they survive the sub-crawler context pipeline
+        // (which would otherwise override them with the sub-crawler's own versions, losing the binding).
+        for (const [key, descriptor] of Object.entries(Object.getOwnPropertyDescriptors(attemptBoundContextHelpers))) {
+            Object.defineProperty(subCrawlerContext, key, { ...descriptor, configurable: false });
+        }
         try {
-            await super._runRequestHandler.call(new Proxy(this, {
-                get: (target, propertyName, receiver) => {
-                    if (propertyName === 'userProvidedRequestHandler') {
-                        return async (playwrightContext) => withCheckedStorageAccess(() => {
-                            if (this.preventDirectStorageAccess) {
-                                throw new Error('Directly accessing storage in a request handler is not allowed in AdaptivePlaywrightCrawler');
-                            }
-                        }, () => this.adaptiveRequestHandler({
-                            id: crawlingContext.id,
-                            session: crawlingContext.session,
-                            proxyInfo: crawlingContext.proxyInfo,
-                            request: crawlingContext.request,
-                            response: {
-                                url: crawlingContext.response.url(),
-                                statusCode: crawlingContext.response.status(),
-                                headers: crawlingContext.response.headers(),
-                                trailers: {},
-                                complete: true,
-                                redirectUrls: [],
-                            },
-                            log: crawlingContext.log,
-                            page: crawlingContext.page,
-                            querySelector: async (selector, timeoutMs = 5_000) => {
-                                const locator = playwrightContext.page.locator(selector).first();
-                                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
-                                const $ = await playwrightContext.parseWithCheerio();
-                                return $(selector);
-                            },
-                            async waitForSelector(selector, timeoutMs = 5_000) {
-                                const locator = playwrightContext.page.locator(selector).first();
-                                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
-                            },
-                            async parseWithCheerio(selector, timeoutMs = 5_000) {
-                                if (selector) {
-                                    const locator = playwrightContext.page.locator(selector).first();
-                                    await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
-                                }
-                                return playwrightContext.parseWithCheerio();
-                            },
-                            async enqueueLinks(options = {}, timeoutMs = 5_000) {
-                                const selector = options.selector ?? 'a';
-                                const locator = playwrightContext.page.locator(selector).first();
-                                await locator.waitFor({ timeout: timeoutMs, state: 'attached' });
-                                const urls = await extractUrlsFromPage(playwrightContext.page, selector, options.baseUrl ??
-                                    playwrightContext.request.loadedUrl ??
-                                    playwrightContext.request.url);
-                                await result.enqueueLinks({ ...options, urls });
-                            },
-                            addRequests: result.addRequests,
-                            pushData: result.pushData,
-                            useState: this.allowStorageAccess(async (defaultValue) => {
-                                const state = await result.useState(defaultValue);
-                                if (initialStateCopy === undefined) {
-                                    initialStateCopy = JSON.parse(JSON.stringify(state));
-                                }
-                                return state;
-                            }),
-                            getKeyValueStore: this.allowStorageAccess(result.getKeyValueStore),
-                        }));
-                    }
-                    return Reflect.get(target, propertyName, receiver);
-                },
-            }), crawlingContext);
-            return { result: { result, ok: true }, initialStateCopy };
+            // Any failure - middleware or handler - ends up in the `ok: false` branch below.
+            const rethrow = (error) => {
+                throw error;
+            };
+            const callAdaptiveRequestHandler = async () => {
+                if (renderingType === 'static') {
+                    await this.#staticContextPipeline.call(subCrawlerContext, this.requestHandler.bind(this), rethrow);
+                }
+                else if (renderingType === 'clientOnly') {
+                    await this.#browserContextPipeline.call(subCrawlerContext, this.requestHandler.bind(this), rethrow);
+                }
+            };
+            // this crawler overrides `runRequestHandler` and times each rendering-type run itself, so it has
+            // to resolve any per-route override too - otherwise routes would be silently ignored here
+            const routeTimeoutSecs = this.requestHandler.getTimeoutSecs?.(context.request.label);
+            const timeoutMillis = routeTimeoutSecs === undefined ? this.#individualRequestHandlerTimeoutMillis : routeTimeoutSecs * 1000;
+            await addTimeoutToPromise(async () => transaction.run(callAdaptiveRequestHandler), timeoutMillis, 'Request handler timed out');
+            return { result: transaction, ok: true, logs };
         }
         catch (error) {
-            return { result: { error, ok: false }, initialStateCopy };
+            return { error, ok: false, logs };
+        }
+        finally {
+            await Promise.all(deferredCleanup.map((cleanup) => cleanup()));
         }
     }
-    async runRequestHandlerWithPlainHTTP(crawlingContext, oldStateCopy) {
-        const result = new RequestHandlerResult(this.config, AdaptivePlaywrightCrawler.CRAWLEE_STATE_KEY);
-        const logs = [];
-        const pageGotoOptions = { timeout: this.navigationTimeoutMillis }; // Irrelevant, but required by BrowserCrawler
+    async runRequestHandler(crawlingContext) {
+        const renderingTypePrediction = await this.#renderingTypePredictor.value.predict(crawlingContext.request);
+        const shouldDetectRenderingType = !this.#shutDown && Math.random() < renderingTypePrediction.detectionProbabilityRecommendation;
+        if (!shouldDetectRenderingType) {
+            crawlingContext.log.debug(`Predicted rendering type ${renderingTypePrediction.renderingType} for ${crawlingContext.request.url}`);
+        }
+        // Every transaction created for this request - up to two, since the static-then-browser
+        // fall-through and the browser-then-detection pair are mutually exclusive. Disposed in the
+        // `finally` below, not earlier: the comparators read the journals after `crawlOne` returns.
+        const transactions = [];
         try {
-            await withCheckedStorageAccess(() => {
-                if (this.preventDirectStorageAccess) {
-                    throw new Error('Directly accessing storage in a request handler is not allowed in AdaptivePlaywrightCrawler');
+            if (renderingTypePrediction.renderingType === 'static' && !shouldDetectRenderingType) {
+                crawlingContext.log.debug(`Running HTTP-only request handler for ${crawlingContext.request.url}`);
+                this.statistics.state.httpOnlyRequestHandlerRuns++;
+                const plainHTTPRun = await this.crawlOne('static', crawlingContext, crawlingContext.useState, transactions);
+                if (plainHTTPRun.ok && this.#resultChecker(plainHTTPRun.result)) {
+                    crawlingContext.log.debug(`HTTP-only request handler succeeded for ${crawlingContext.request.url}`);
+                    plainHTTPRun.logs?.forEach(([log, method, ...args]) => log[method](...args));
+                    await plainHTTPRun.result.commit();
+                    return;
                 }
-            }, async () => addTimeoutToPromise(async () => {
-                const hookContext = {
-                    id: crawlingContext.id,
-                    session: crawlingContext.session,
-                    proxyInfo: crawlingContext.proxyInfo,
-                    request: crawlingContext.request,
-                    log: this.createLogProxy(crawlingContext.log, logs),
-                };
-                await this._executeHooks(this.preNavigationHooks, {
-                    ...hookContext,
-                    get page() {
-                        throw new Error('Page object was used in HTTP-only pre-navigation hook');
-                    },
-                }, // This is safe because `executeHooks` just passes the context to the hooks which accept the partial context
-                pageGotoOptions);
-                const response = await crawlingContext.sendRequest({});
-                const loadedUrl = response.url;
-                crawlingContext.request.loadedUrl = loadedUrl;
-                const $ = load(response.body);
-                await this.adaptiveRequestHandler({
-                    ...hookContext,
-                    request: crawlingContext.request,
-                    response,
-                    get page() {
-                        throw new Error('Page object was used in HTTP-only request handler');
-                    },
-                    async querySelector(selector, _timeoutMs) {
-                        return $(selector);
-                    },
-                    async waitForSelector(selector, _timeoutMs) {
-                        if ($(selector).get().length === 0) {
-                            throw new Error(`Selector '${selector}' not found.`);
+                // Execution will "fall through" and try running the request handler in a browser
+                if (!plainHTTPRun.ok) {
+                    const actualError = plainHTTPRun.error instanceof RequestHandlerError
+                        ? plainHTTPRun.error.cause
+                        : plainHTTPRun.error;
+                    if (await this.#shouldPropagateError(actualError, crawlingContext)) {
+                        throw actualError;
+                    }
+                    crawlingContext.log.exception(actualError, `HTTP-only request handler failed for ${crawlingContext.request.url}`);
+                }
+                else {
+                    crawlingContext.log.warning(`HTTP-only request handler returned a suspicious result for ${crawlingContext.request.url}`);
+                    this.statistics.state.renderingTypeMispredictions++;
+                }
+            }
+            crawlingContext.log.debug(`Running browser request handler for ${crawlingContext.request.url}`);
+            this.statistics.state.browserRequestHandlerRuns++;
+            // Run the request handler in a browser. The copy of the crawler state is kept so that we can perform
+            // a rendering type detection if necessary. Without this measure, the HTTP request handler would run
+            // under different conditions, which could change its behavior. Changes done to the crawler state by
+            // the HTTP request handler will not be committed to the actual storage.
+            const stateTracker = {
+                stateCopy: null,
+                async getLiveState(defaultValue = {}) {
+                    const state = await crawlingContext.useState(defaultValue);
+                    if (this.stateCopy === null) {
+                        this.stateCopy = JSON.parse(JSON.stringify(state));
+                    }
+                    return state;
+                },
+                async getStateCopy(defaultValue = {}) {
+                    if (this.stateCopy === null) {
+                        return defaultValue;
+                    }
+                    return this.stateCopy;
+                },
+            };
+            const browserRun = await this.crawlOne('clientOnly', crawlingContext, stateTracker.getLiveState.bind(stateTracker), transactions);
+            if (!browserRun.ok) {
+                throw browserRun.error;
+            }
+            browserRun.logs?.forEach(([log, method, ...args]) => log[method](...args));
+            await browserRun.result.commit();
+            if (shouldDetectRenderingType) {
+                const detectionPromise = (async () => {
+                    crawlingContext.log.debug(`Detecting rendering type for ${crawlingContext.request.url}`);
+                    // The detection attempt's transaction is never committed - its writes exist only for the
+                    // result comparison.
+                    const plainHTTPRun = await this.crawlOne('static', crawlingContext, stateTracker.getStateCopy.bind(stateTracker), transactions);
+                    const detectionResult = (() => {
+                        if (!plainHTTPRun.ok) {
+                            return 'clientOnly';
                         }
-                    },
-                    async parseWithCheerio(selector, _timeoutMs) {
-                        if (selector && $(selector).get().length === 0) {
-                            throw new Error(`Selector '${selector}' not found.`);
+                        const comparisonResult = this.#resultComparator(plainHTTPRun.result, browserRun.result);
+                        if (comparisonResult === true || comparisonResult === 'equal') {
+                            return 'static';
                         }
-                        return $;
-                    },
-                    async enqueueLinks(options = {}) {
-                        const urls = extractUrlsFromCheerio($, options.selector, options.baseUrl ?? loadedUrl);
-                        await result.enqueueLinks({ ...options, urls });
-                    },
-                    addRequests: result.addRequests,
-                    pushData: result.pushData,
-                    useState: async (defaultValue) => {
-                        // return the old state before the browser handler was executed
-                        // when rerunning the handler via HTTP for detection
-                        if (oldStateCopy !== undefined) {
-                            return oldStateCopy ?? defaultValue; // fallback to the default for `null`
+                        if (comparisonResult === false || comparisonResult === 'different') {
+                            return 'clientOnly';
                         }
-                        return this.allowStorageAccess(result.useState)(defaultValue);
-                    },
-                    getKeyValueStore: this.allowStorageAccess(result.getKeyValueStore),
-                });
-                await this._executeHooks(this.postNavigationHooks, crawlingContext, pageGotoOptions);
-            }, this.requestHandlerTimeoutInnerMillis, 'Request handler timed out'));
-            return { result, logs, ok: true };
+                        return undefined;
+                    })();
+                    crawlingContext.log.debug(`Detected rendering type ${detectionResult} for ${crawlingContext.request.url}`);
+                    if (detectionResult !== undefined) {
+                        // Deliberately not awaited: a predictor that persists asynchronously gets to keep
+                        // batching its writes, and the drain below catches whatever is still pending.
+                        const stored = this.#renderingTypePredictor.value.storeResult(crawlingContext.request, detectionResult);
+                        if (stored !== undefined) {
+                            // Nothing downstream awaits this, so a failed write would otherwise be silent.
+                            void this.#trackDetection(Promise.resolve(stored).catch((error) => this.log.exception(error, `Failed to store the rendering type detection result for ${crawlingContext.request.url}`)));
+                        }
+                    }
+                })();
+                await this.#trackDetection(detectionPromise);
+            }
         }
-        catch (error) {
-            return { error, logs, ok: false };
+        finally {
+            // A still-open transaction here belongs to a discarded attempt - roll it back, then release.
+            for (const transaction of transactions) {
+                transaction.rollback();
+                transaction.dispose();
+            }
         }
     }
+    async enqueueLinks(urls, options, request) {
+        const baseUrl = resolveBaseUrlForEnqueueLinksFiltering({
+            enqueueStrategy: options?.strategy,
+            finalRequestUrl: request.loadedUrl,
+            originalRequestUrl: request.url,
+            userProvidedBaseUrl: options?.baseUrl,
+        });
+        const requestsWithDepth = this.addCrawlDepthRequestGenerator(urls, request.crawlDepth + 1);
+        // The per-attempt transaction buffers these (the queue policy defaults to `deferred` here),
+        // so a discarded attempt's enqueues never reach the queue.
+        return await this.addRequests(requestsWithDepth, {
+            ...options,
+            baseUrl,
+            strategy: options.strategy ?? EnqueueStrategy.SameHostname,
+        });
+    }
     createLogProxy(log, logs) {
         return new Proxy(log, {
-            get(target, propertyName, receiver) {
+            get(target, propertyName) {
                 if (proxyLogMethods.includes(propertyName)) {
                     return (...args) => {
                         logs.push([target, propertyName, ...args]);
                     };
                 }
-                return Reflect.get(target, propertyName, receiver);
+                const value = Reflect.get(target, propertyName, target);
+                // Bind non-intercepted methods to the target instance so private #-fields
+                // (e.g. BaseCrawleeLogger.#options, #warningsLogged) do not throw TypeError at runtime.
+                if (typeof value === 'function') {
+                    return value.bind(target);
+                }
+                return value;
             },
         });
     }
+    #trackDetection(promise) {
+        this.#activeDetections.add(promise);
+        // Not `finally()`: the promise it derives would reject on its own and go unhandled. A rejection here
+        // belongs to whoever awaits the original, or to `allSettled` in the drain.
+        void promise.catch(() => { }).then(() => this.#activeDetections.delete(promise));
+        return promise;
+    }
+    /**
+     * Number of rendering type detections that have not settled yet, including results the predictor is
+     * still persisting.
+     */
+    get inFlightRenderingTypeDetectionCount() {
+        return this.#activeDetections.size;
+    }
+    /**
+     * Waits for in-flight rendering type detections to settle, bounded by `timeoutMillis` (defaults to the
+     * internal timeout).
+     */
+    async drainRenderingDetections({ timeoutMillis } = {}) {
+        if (this.#activeDetections.size === 0) {
+            return;
+        }
+        const drained = (async () => {
+            while (this.#activeDetections.size > 0) {
+                await Promise.allSettled(Array.from(this.#activeDetections));
+            }
+        })();
+        const millis = timeoutMillis ?? this.internalTimeoutMillis;
+        // A caller opting out of the bound would otherwise get an immediate spurious timeout - `setTimeout`
+        // clamps a non-finite delay to 1ms.
+        if (!Number.isFinite(millis)) {
+            await drained;
+            return;
+        }
+        const abortTimer = new AbortController();
+        try {
+            const outcome = await Promise.race([
+                drained.then(() => 'drained'),
+                delay(millis, 'timedOut', { signal: abortTimer.signal }).catch(() => 'aborted'),
+            ]);
+            if (outcome === 'timedOut') {
+                this.log.warning(`Timed out after ${millis / 1e3} seconds waiting for ${this.#activeDetections.size} rendering type detection(s) to settle - their results may be lost.`);
+            }
+        }
+        finally {
+            abortTimer.abort();
+        }
+    }
+    /**
+     * Stops the crawler immediately, but not before rendering type detections already under way (and results
+     * the predictor is still persisting) have settled - see
+     * {@link AdaptivePlaywrightCrawler.drainRenderingDetections|`drainRenderingDetections()`}. Requests
+     * that are still running are not waited for, unlike {@link BasicCrawler.stop|`stop()`}.
+     */
+    async teardown() {
+        // Called from outside `run()` - under `keepAlive`, say - the pool keeps dispatching until
+        // `super.teardown()` aborts it, and a request starting during the drain would open a detection the
+        // drain has already passed. Closing that first makes the drain a fence.
+        this.#shutDown = true;
+        await this.drainRenderingDetections();
+        await super.teardown();
+        // Mirrors the owned-only `initialize()` in `init()` - without this, the predictor we built keeps its
+        // PERSIST_STATE listener registered after the crawl and never gets a final write.
+        await this.#renderingTypePredictor.ifOwned((predictor) => predictor.teardown());
+        await this.#browserCrawler.teardown();
+    }
+    async destroy() {
+        await super.destroy();
+        await this.#staticCrawler.destroy();
+        await this.#browserCrawler.destroy();
+    }
+}
+export function createAdaptivePlaywrightRouter(routesOrSchemas) {
+    return Router.create(routesOrSchemas);
 }
-export function createAdaptivePlaywrightRouter(routes) {
-    return Router.create(routes);
+/**
+ * An opt-in {@link AdaptivePlaywrightCrawlerOptions.resultComparator|`resultComparator`} that considers two
+ * request handler results equal only if *all* of their observable effects match - the pushed dataset items, the
+ * enqueued requests, and the key-value store changes. This is stricter than the default comparator, which only
+ * compares dataset items.
+ *
+ * **Beware:** enqueued URLs are compared exactly. The same page rendered in a browser and via plain HTTP often
+ * yields links that differ only in tracking query parameters, for example:
+ * - `https://sdk.apify.com/docs/guides/getting-started`
+ * - `https://sdk.apify.com/docs/guides/getting-started?__hsfp=1136113150&__hssc=7591405.1.173549427712`
+ *
+ * Such links are treated as *different*, which will make the crawler favor browser rendering for those pages.
+ *
+ * **Example usage:**
+ * ```ts
+ * const crawler = new AdaptivePlaywrightCrawler({
+ *     resultComparator: fullResultComparator,
+ *     async requestHandler({ pushData, enqueueLinks }) {
+ *         // ...
+ *     },
+ * });
+ * ```
+ */
+export function fullResultComparator(resultA, resultB) {
+    return (isDeepStrictEqual(resultA.datasetItems, resultB.datasetItems) &&
+        isDeepStrictEqual(resultA.enqueuedUrls, resultB.enqueuedUrls) &&
+        isDeepStrictEqual(resultA.keyValueStoreChanges, resultB.keyValueStoreChanges));
 }
-//# sourceMappingURL=adaptive-playwright-crawler.js.map