Merge pull request #479 from ganisback/support-stream

feat: support stream api

Author: yuvipanda

jupyter_server_proxy/handlers.py

@@ -5,6 +5,7 @@
 """
 
 import os
+import re
 import socket
 from asyncio import Lock
 from copy import copy
@@ -287,7 +288,7 @@ def get_client_uri(self, protocol, host, port, proxied_path):
 
         return client_uri
 
-    def _build_proxy_request(self, host, port, proxied_path, body):
+    def _build_proxy_request(self, host, port, proxied_path, body, **extra_opts):
         headers = self.proxy_request_headers()
 
         client_uri = self.get_client_uri("http", host, port, proxied_path)
@@ -307,6 +308,7 @@ def _build_proxy_request(self, host, port, proxied_path, body):
             decompress_response=False,
             headers=headers,
             **self.proxy_request_options(),
+            **extra_opts,
         )
         return req
 
@@ -365,7 +367,6 @@ async def proxy(self, host, port, proxied_path):
                 body = b""
             else:
                 body = None
-
         if self.unix_socket is not None:
             # Port points to a Unix domain socket
             self.log.debug("Making client for Unix socket %r", self.unix_socket)
@@ -374,8 +375,97 @@ async def proxy(self, host, port, proxied_path):
                 force_instance=True, resolver=UnixResolver(self.unix_socket)
             )
         else:
-            client = httpclient.AsyncHTTPClient()
+            client = httpclient.AsyncHTTPClient(force_instance=True)
+        # check if the request is stream request
+        accept_header = self.request.headers.get("Accept")
+        if accept_header == "text/event-stream":
+            return await self._proxy_progressive(host, port, proxied_path, body, client)
+        else:
+            return await self._proxy_buffered(host, port, proxied_path, body, client)
+
+    async def _proxy_progressive(self, host, port, proxied_path, body, client):
+        # Proxy in progressive flush mode, whenever chunks are received. Potentially slower but get results quicker for voila
+        # Set up handlers so we can progressively flush result
+
+        headers_raw = []
+
+        def dump_headers(headers_raw):
+            for line in headers_raw:
+                r = re.match("^([a-zA-Z0-9\\-_]+)\\s*\\:\\s*([^\r\n]+)[\r\n]*$", line)
+                if r:
+                    k, v = r.groups([1, 2])
+                    if k not in (
+                        "Content-Length",
+                        "Transfer-Encoding",
+                        "Content-Encoding",
+                        "Connection",
+                    ):
+                        # some header appear multiple times, eg 'Set-Cookie'
+                        self.set_header(k, v)
+                else:
+                    r = re.match(r"^HTTP[^\s]* ([0-9]+)", line)
+                    if r:
+                        status_code = r.group(1)
+                        self.set_status(int(status_code))
+            headers_raw.clear()
+
+        # clear tornado default header
+        self._headers = httputil.HTTPHeaders()
+
+        def header_callback(line):
+            headers_raw.append(line)
+
+        def streaming_callback(chunk):
+            # record activity at start and end of requests
+            self._record_activity()
+            # Do this here, not in header_callback so we can be sure headers are out of the way first
+            dump_headers(
+                headers_raw
+            )  # array will be empty if this was already called before
+            self.write(chunk)
+            self.flush()
+
+        # Now make the request
+
+        req = self._build_proxy_request(
+            host,
+            port,
+            proxied_path,
+            body,
+            streaming_callback=streaming_callback,
+            header_callback=header_callback,
+        )
+
+        # no timeout for stream api
+        req.request_timeout = 7200
+        req.connect_timeout = 600
+
+        try:
+            response = await client.fetch(req, raise_error=False)
+        except httpclient.HTTPError as err:
+            if err.code == 599:
+                self._record_activity()
+                self.set_status(599)
+                self.write(str(err))
+                return
+            else:
+                raise
+
+        # For all non http errors...
+        if response.error and type(response.error) is not httpclient.HTTPError:
+            self.set_status(500)
+            self.write(str(response.error))
+        else:
+            self.set_status(
+                response.code, response.reason
+            )  # Should already have been set
+
+            dump_headers(headers_raw)  # Should already have been emptied
+
+            if response.body:  # Likewise, should already be chunked out and flushed
+                self.write(response.body)
 
+    async def _proxy_buffered(self, host, port, proxied_path, body, client):
         req = self._build_proxy_request(host, port, proxied_path, body)
 
         self.log.debug(f"Proxying request to {req.url}")

tests/resources/eventstream.py

@@ -0,0 +1,36 @@
+import asyncio
+
+import tornado.escape
+import tornado.ioloop
+import tornado.options
+import tornado.web
+import tornado.websocket
+from tornado.options import define, options
+
+
+class Application(tornado.web.Application):
+    def __init__(self):
+        handlers = [
+            (r"/stream/(\d+)", StreamHandler),
+        ]
+        super().__init__(handlers)
+
+
+class StreamHandler(tornado.web.RequestHandler):
+    async def get(self, seconds):
+        for i in range(int(seconds)):
+            await asyncio.sleep(0.5)
+            self.write(f"data: {i}\n\n")
+            await self.flush()
+
+
+def main():
+    define("port", default=8888, help="run on the given port", type=int)
+    options.parse_command_line()
+    app = Application()
+    app.listen(options.port)
+    tornado.ioloop.IOLoop.current().start()
+
+
+if __name__ == "__main__":
+    main()

tests/resources/jupyter_server_config.py

@@ -79,6 +79,9 @@ def my_env():
             "X-Custom-Header": "pytest-23456",
         },
     },
+    "python-eventstream": {
+        "command": [sys.executable, "./tests/resources/eventstream.py", "--port={port}"]
+    },
     "python-unix-socket-true": {
         "command": [
             sys.executable,

tests/test_proxies.py

@@ -1,13 +1,14 @@
 import gzip
 import json
 import sys
+import time
 from http.client import HTTPConnection
 from io import BytesIO
 from typing import Tuple
 from urllib.parse import quote
 
 import pytest
-from tornado.httpclient import HTTPClientError
+from tornado.httpclient import AsyncHTTPClient, HTTPClientError
 from tornado.websocket import websocket_connect
 
 # use ipv4 for CI, etc.
@@ -343,6 +344,40 @@ def test_server_content_encoding_header(
         assert f.read() == b"this is a test"
 
 
+async def test_eventstream(a_server_port_and_token: Tuple[int, str]) -> None:
+    PORT, TOKEN = a_server_port_and_token
+    # The test server under eventstream.py will send back monotonically increasing numbers
+    # starting at 0 until the specified limit, with a 500ms gap between them. We test that:
+    # 1. We get back as many callbacks from our streaming read as the total number,
+    #    as the server does a flush after writing each entry.
+    # 2. The streaming entries are read (with some error margin) around the 500ms mark, to
+    #    ensure this is *actually* being streamed
+    limit = 3
+    last_cb_time = time.perf_counter()
+    times_called = 0
+    stream_read_intervals = []
+    stream_data = []
+
+    def streaming_cb(data):
+        nonlocal times_called, last_cb_time, stream_read_intervals
+        time_taken = time.perf_counter() - last_cb_time
+        last_cb_time = time.perf_counter()
+        stream_read_intervals.append(time_taken)
+        times_called += 1
+        stream_data.append(data)
+
+    url = f"http://{LOCALHOST}:{PORT}/python-eventstream/stream/{limit}?token={TOKEN}"
+    client = AsyncHTTPClient()
+    await client.fetch(
+        url,
+        headers={"Accept": "text/event-stream"},
+        streaming_callback=streaming_cb,
+    )
+    assert times_called == limit
+    assert all([0.45 < t < 3.0 for t in stream_read_intervals])
+    assert stream_data == [b"data: 0\n\n", b"data: 1\n\n", b"data: 2\n\n"]
+
+
 async def test_server_proxy_websocket_messages(
     a_server_port_and_token: Tuple[int, str]
 ) -> None: