Playwright

pip install playwright

Playwright 基本框架 ^async/await

from playwright.async_api import async_playwright

async with async_playwright() as p:
  browser = await p.chromium.launch()
  page = await browser.new_page()
  await page.goto("...")
  title = await page.title()

  ...

  await browser.close()

page / context

• page: 相當於一個瀏覽器 (browser) 中的分頁 (tab)
• context: 一組 session (cookie, localStorage, …)
  可由 page.context 取得

Jupyter Notebook (.ipynb) on Windows

在 Windows 的某些執行環境 (如：Jupyter Notebook) 中
預設使用的 asyncio event loop policy (SelectorEventLoopPolicy)不支援 pipes/subprocess transport
因此 Playwright 在嘗試建立 browser subprocess 或 pipe 通訊時會拋出NotImplementedError

import platform, asyncio
if platform.system() == "Windows":
  asyncio.set_event_loop_policy(asyncio.WindowsProactorEventLoopPolicy())

🙅‍♂️ sync

import traceback
from playwright.sync_api import sync_playwright

def run_playwright():
  try:
    with sync_playwright() as p:
      browser = p.chromium.launch()
      page = browser.new_page()
      page.goto('...')

      ...

      page.wait_for_event("close", timeout=0)
  except Exception as e:
    print(f"An error occurred: {repr(e)}")
    traceback.print_exc()

import threading
thread = threading.Thread(target=run_playwright)
thread.start()
thread.join() # 等 thread 結束才繼續

在 Win + VS Code 的 .ipynb 環境中
主執行緒由 IPython kernel 持有 asyncio event loop
透過另開 thread 將其隔離至不受既有 asyncio event loop 影響的獨立執行環境
使 Playwright Sync/Async API 得以運行

import traceback
from playwright.async_api import async_playwright

async def run_playwright():
  try:
    async with async_playwright() as p:
      browser = await p.chromium.launch()
      page = await browser.new_page()
      await page.goto('...')

      ...

      await page.wait_for_event("close", timeout=0)
  except Exception as e:
    print(f"An error occurred: {repr(e)}")
    traceback.print_exc()

import asyncio
def thread_entry():
  asyncio.run(run_playwright())

import threading
thread = threading.Thread(target=thread_entry)
thread.start()
thread.join() # 等 thread 結束才繼續

Python 開啟 Chrome

import subprocess, requests, time

import platform, os
project_name = ""
match platform.system():
  case 'Windows':
    chrome_path = r"C:\Program Files\Google\Chrome\Application\chrome.exe"
    user_data_dir = os.path.join(
      os.environ["LOCALAPPDATA"],
      "Chrome-CDP" + (project_name if project_name == "" else f"_{project_name}")
    )
  case 'Linux':
    chrome_path = 'google-chrome'
    os.path.join(os.path.expanduser("~/.config/google-chrome/"), project_name)
cdp_port = 9453
cdp_url = f"http://127.0.0.1:{cdp_port}/json/version"

subprocess.Popen([
  chrome_path,
  f"--remote-debugging-port={cdp_port}",
  f"--user-data-dir={user_data_dir}",
  # "--no-first-run", "--no-default-browser-check", "--disable-sync",
  # "--headless=new", 
])

# 等待 CDP ready
for _ in range(s := 10):  # 最多等 s 秒
  try:
    r = requests.get(cdp_url, timeout=0.2)
    if r.status_code == 200:
      break
  except requests.exceptions.RequestException:
    time.sleep(1)
else:
  raise RuntimeError("Chrome CDP did not start")

Playwright 連接至 啟用 remote debugging 的 Chrome

with ..._playwright() as p:
  browser = p.chromium.connect_over_cdp(cdp_url)
  context = browser.contexts[-1]
  page = context.pages[-1]
  ...

context

context.cookies()

context.cookies(): list[dict]

傳遞 storage_state

匯出

• state = await context.storage_state()
• 匯出為檔案：await context.storage_state(path="... .json")

匯入

context = await browser.new_context(storage_state="... .json")
page = await context.new_page()
...

沿用既存的 user_data_dir

async with async_playwright() as p:
  context = await p.chromium.launch_persistent_context(user_data_dir=user_data_dir)
  browser = context.browser
  page = await context.new_page()
  ...

page

wait

page.wait_for_url()

page.wait_for_url(“**/xxx/**”)

• 等待 URL 變成指定路徑
• **:= 萬用字元/任意字串（含/）

當 URL 有變動

page.wait_for_url(lambda url: url != old_url)
old_url = page.url

page.expect_navigation()

with page.expect_navigation():
  ...

• 等待「瀏覽器層級的導航事件」，包含：
  • URL 改變
  • document reload
  • window.location 改變
  • form submit 導致離開頁面
• timeout預設為 30000 (ms)
  • 不設timeout：with page.expect_navigation(timeout=0):

page.wait_for_load_state()

page.wait_for_load_state('domcontentloaded')

• ✅ HTML 已下載並 parse 完成
• ✅ DOM tree 已建立
• 💁‍♂️ JS、圖片、CSS 可能還在載

page.wait_for_load_state()
≡ page.wait_for_load_state(‘load’)

• ✅ DOM ready

• ✅ 同步 JS 完成

• 🙆‍♂️ 圖片 / CSS 大多已載入

page.wait_for_load_state('networkidle')

• ⏱️ 連續 500ms 沒有任何 network request

• ⚠️ 不只是主頁，包含 fetch / xhr

page.wait_for_selector()

page.screenshot()

await page.screenshot(path="page.png"[, full_page=True])

• 引數full_page預設為False
  只擷取可視範圍
• full_page=True
  自動捲動頁面並拼貼為整頁內容

Jupyter Notebook

from IPython.display import Image, display

img_bytes = await page.screenshot()
display(Image(data=img_bytes))

locator

locator()

• nodes = page.locator('<css_selector (default)>')³CSS Selector
• nodes = page.locator('xpath=<xpath>')⁴XPath

count(), nth(), first, last, all()

• if await nodes.count() > 0 : ...
• nodes.nth(i):= nods 中第 i + 1 個結果，i = 0, 1, 2, …, nodes.count() – 1
• nodes.first ≡ nodes.nth(0)
nodes.last ≡ nodes.nth(await nodes.count() - 1)
• ❌ nodes.first()
  ❌ nodes.last()
• for i in range(await nodes.count()): ...nodes.nth(i)...
  ≡ for node in await nodes.all(): ...
• subnodes = nodes.locator('<css_selector>')
• for subnode in await nodes.locator('<css_selector>').all(): ...

get_attribute(), inner_text()

• attribute = await node.get_attribute('<attribute>')
• text = await node.inner_text()

模擬瀏覽器操作

開啟連結

urljoin()

from urllib.parse import urljoin

full_url = urljoin(page.url, await a.get_attribute('href'))
await page.goto(full_url)

click()

locator([selector]).click()

上一頁

await page.go_back()

Last Updated on 2026/01/26 by A1go

References

• 1
  The Chromium Projects: Run Chromium with command-line switches
• 2
  List of Chromium Command Line Switches
• 3
  CSS Selector
• 4
  XPath