BasicCrawler decides whether to enter and exit a collaborator by whether it is already active (_basic_crawler.py around line 832), not by who created it. An injected Statistics, SessionPool or BrowserPool that is not yet active is entered and torn down by the first crawler that runs, so sharing one between crawlers breaks: the first to finish exits it while the other still uses it. A user-supplied Statistics and SessionPool are also reset on every repeated run() (around line 728). The Playwright crawler always enters a passed browser_pool, and AdaptivePlaywrightCrawler initializes and clears a passed rendering_type_predictor.

Per-run state lives on the crawler itself (_running, _has_finished_before, _last_run_failed, _unexpected_stop and others), and repeated runs depend on resetting it by hand.

JS v4 settled both: a crawler owns, starts, resets and stops only what it built, and each run() gets a fresh run object.

Proposed solution

• Track owned vs injected collaborators explicitly. The crawler creates and manages its defaults; injected instances are borrowed, so the caller enters and exits them, and the crawler never resets them.
• Apply this to the event manager, statistics, session pool, HTTP client, browser pool, rendering type predictor and request manager.
• Extract per-run state into an internal CrawlerRun built on every run().

Acceptance criteria

• Two crawlers can share one SessionPool, Statistics or BrowserPool and run concurrently.
• Re-running a crawler does not reset injected collaborators.
• Upgrade notes describing who enters and exits each injected object.

JS references

• Ownership pattern: Unify the inject-or-default ownership pattern crawlee#3888, implemented in refactor: Introduce OwnedOrInjected dependency tracking helper crawlee#3891
• Pre-configured Statistics instance: Accept a preconfigured Statistics instance instead of statisticsOptions crawlee#3889, refactor!: Accept a pre-configured Statistics instance in crawlers instead of an option bag crawlee#3966
• Predictor ownership: Fix renderingTypePredictor lifecycle ownership in AdaptivePlaywrightCrawler crawlee#3887, refactor: Introduce IRenderingTypePredictor and unify the ownership pattern crawlee#3944
• Browser pool factories: refactor!: replace browserPoolOptions with browser pool factories crawlee#4026
• CrawlerRun: Extract data relevant to a single BasicCrawler.run into a separate class crawlee#3087, implemented in refactor: Extract per-run crawler state into a CrawlerRun class crawlee#4111
• Shared SessionPool: Allow sharing a SessionPool instance across multiple crawlers crawlee#3445

Related: #2133 and #2132 build on this.

This will be a breaking change.

✍️ Drafted by Claude Code