巨硬出品的playwright是一个非常不错的自动化工具，能够用同一套API去控制Chromium, Firefox 和 WebKit ，看到这儿大家也许会想到可以用这玩意儿做爬虫，相比puppeteer或者pyppeteer，他们适配的浏览器更多，适配的语言更多。但是其官方不提供远程调用的方式。这儿是一个简单的远程调用playwright的梳理流程和简单的代码。以playwright-python版本作为示例，其他语言的playwright的做法是一样的。

安装和启动方式

python版本的playwright在启动浏览器进程的时候是用nodejs版本的进行启动的

在安装的时候setup.py会去下载对应版本的nodejs的版本

启动代码为

https://github.com/microsoft/playwright-python/blob/03e5cd01fdda2125cea47ab443d34564f767af13/playwright/_impl/_transport.py#L57

class Transport:
    async def run(self) -> None:
        self._loop = asyncio.get_running_loop()
        self._stopped_future: asyncio.Future = asyncio.Future()

        self._proc = proc = await asyncio.create_subprocess_exec(
            str(self._driver_executable),#self._driver_executable就是 /Users/lozzo/.virtualenvs/py37/lib/python3.7/site-packages/playwright/driver/playwright.sh
            "run-driver",
            stdin=asyncio.subprocess.PIPE,
            stdout=asyncio.subprocess.PIPE,
            stderr=_get_stderr_fileno(),
            limit=32768,
        )
        assert proc.stdout
        assert proc.stdin
        self._output = proc.stdin

        while not self._stopped:
            try:
                buffer = await proc.stdout.readexactly(4)
                length = int.from_bytes(buffer, byteorder="little", signed=False)
                buffer = bytes(0)
                while length:
                    to_read = min(length, 32768)
                    data = await proc.stdout.readexactly(to_read)
                    length -= to_read
                    if len(buffer):
                        buffer = buffer + data
                    else:
                        buffer = data
                obj = json.loads(buffer)

                if "DEBUGP" in os.environ:  # pragma: no cover
                    print("\x1b[33mRECV>\x1b[0m", json.dumps(obj, indent=2))
                self.on_message(obj)
            except asyncio.IncompleteReadError:
                break
            await asyncio.sleep(0)
        self._stopped_future.set_result(None)

所以要使playwright能够进行远程访问，只需要修改python版本的启动方式，然后后面的远程的nodejs版本更具请求参数进行转换，然后将ws暴露出来，即可进行远程访问

python端和nodejs端是通过进程间通信进行通信的，所以我们只需要在python和nodejs外各包一层，然后在让他们外包的一层之间用socket进行通信即可实现远程调用。

代码实现

python端代码修改

修改代码https://github.com/microsoft/playwright-python/blob/29cddbd5174ab262e5cb57b2d8c8fbcf8df3e171/playwright/_impl/_driver.py#L24

def compute_driver_executable() -> Path:
    return Path("/Users/lozzo/.virtualenvs/py37/lib/python3.7/site-packages/playwright/driver/playwright.sh")
    # package_path = Path(inspect.getfile(playwright)).parent
    # platform = sys.platform
    # if platform == "win32":
    #     return package_path / "driver" / "playwright.cmd"
    # return package_path / "driver" / "playwright.sh"

其中/Users/lozzo/.virtualenvs/py37/lib/python3.7/site-packages/playwright/driver/playwright.sh内容为

#!/bin/sh
#SCRIPT_PATH="$(cd "$(dirname "$0")" ; pwd -P)"
#$SCRIPT_PATH/node $SCRIPT_PATH/package/lib/cli/cli.js "$@"
cd /Users/lozzo/workdir/sovietironfist/test
ts-node processPipe.ts

processPipe.ts内容为

import net from "net"
(async()=>{
    const socket = new net.Socket()
    socket.connect({host:"127.0.0.1",port:12345})
    process.stdin.pipe(socket)
    socket.pipe(process.stdout)
})()

需要注意的是请不要在这个脚本中使用console.*进行任何标准输入输出操作，这些操作会被本地的playwright-python进行读取，产生异常

当然也可以修改playwright/_impl/_transport.py Transport类,使之直接和远端的socket链接，少走一层

class Transport:
       async def run(self) -> None:
        self._loop = asyncio.get_running_loop()
        self._stopped_future: asyncio.Future = asyncio.Future()
        
        reader, writer = await asyncio.open_connection(host='127.0.0.1',port=12345)
        self._output = writer

        while not self._stopped:
            try:
                buffer = await reader.readexactly(4)
                length = int.from_bytes(buffer, byteorder="little", signed=False)
                buffer = bytes(0)
                while length:
                    to_read = min(length, 32768)
                    data = await reader.readexactly(to_read)
                    length -= to_read
                    if len(buffer):
                        buffer = buffer + data
                    else:
                        buffer = data
                obj = json.loads(buffer)

                if "DEBUGP" in os.environ:  # pragma: no cover
                    print("\x1b[33mRECV>\x1b[0m", json.dumps(obj, indent=2))
                self.on_message(obj)
            except asyncio.IncompleteReadError:
                break
            await asyncio.sleep(0)
        self._stopped_future.set_result(None)

node端代码修改

常驻进程为socketPipe.ts

import net from 'net'
import { spawn, ChildProcessWithoutNullStreams } from 'child_process'
;(async () => {
    let child: ChildProcessWithoutNullStreams | undefined
    const server = net.createServer()
    server.listen(12345)
    const close = () => {
        if (child) {
            child.kill(0)
            child = undefined
        }
    }
    server.on('connection', (socket: net.Socket) => {
        console.log('connection')
        if (!child) {
            child = spawn('/Users/lozzo/.virtualenvs/py37/lib/python3.7/site-packages/playwright/driver/b.sh')
        }
        child.on('exit', (number, signal) => {
            console.log('exit', number, signal)
        })
        child.stdout.pipe(socket)
        socket.pipe(child.stdin)
        socket.on('error', close)
    })

    server.on('close', close)
    server.on('error', close)
})()

其中/Users/lozzo/.virtualenvs/py37/lib/python3.7/site-packages/playwright/driver/b.sh内容为

#!/bin/sh
SCRIPT_PATH="$(cd "$(dirname "$0")" ; pwd -P)"
$SCRIPT_PATH/node $SCRIPT_PATH/package/lib/cli/cli.js 'run-driver'

服务端启动ts-node socketPipe.ts

然后就可以远程调用了(本地无感使用)，同理，其他的语言的服务是可以一样操作的