Repository navigation
[Bug]: BestFirstCrawlingStrategy fetches a full batch past max_pages #2381
Description
crawl4ai version
0.9.5 (same on 0.9.0)
Expected Behavior
BestFirstCrawlingStrategy with max_pages=N fetches about N pages: the last batch is cut to the remaining budget.
Current Behavior
_arun_best_first computes batch_size = min(BATCH_SIZE, remaining) but pops range(BATCH_SIZE) items, so the last batch is always 10 URLs. They are all fetched, and the results past the limit are dropped. With max_pages=12, 21 pages are fetched for 12 results (24 on 0.9.0).
Is this reproducible?
Yes
Code snippets
import asyncio
from aiohttp import web
from crawl4ai import AsyncWebCrawler, CrawlerRunConfig
from crawl4ai.deep_crawling import BestFirstCrawlingStrategy
hits = []
async def handle(request):
hits.append(request.path)
links = "".join(f'<a href="/p{i}">p{i}</a>' for i in range(30)) if request.path == "/" else ""
return web.Response(text=f"<html><body>{links}<p>{'text ' * 200}</p></body></html>", content_type="text/html")
async def main():
app = web.Application()
app.router.add_get("/{tail:.*}", handle)
runner = web.AppRunner(app)
await runner.setup()
await web.TCPSite(runner, "127.0.0.1", 8766).start()
config = CrawlerRunConfig(deep_crawl_strategy=BestFirstCrawlingStrategy(max_depth=1, max_pages=12), stream=True)
async with AsyncWebCrawler() as crawler:
results = [r async for r in await crawler.arun("http://127.0.0.1:8766/", config=config)]
print("results:", len(results), "pages fetched:", len([h for h in hits if h != "/robots.txt"])) # 12 vs 21
await runner.cleanup()
asyncio.run(main())Suggested fix
for _ in range(batch_size): in the batch-filling loop. The open PR #2246 (configurable batch_size) keeps the same pattern: it computes effective_batch_size but loops over range(self.batch_size).
Activity
fix: honor remaining max_pages budget in BestFirstCrawlingStrategy fi…
fix(deep_crawling): cap best-first batch by remaining max_pages budget
yashikam19 commentedon Oct 15, 2026
Thanks @loickouao, good catch, and you're right that #2246 had the same problem.
I have pushed a fix to #2246: the batch loop now iterates over
effective_batch_size, so the final round is capped by the remainingmax_pagesbudget.