zulip/tools/run-dev.py

398 lines
13 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
import argparse
import os
import pwd
import signal
import subprocess
import sys
import traceback
from typing import Any, Callable, Generator, List, Sequence
from urllib.parse import urlunparse
2016-07-30 00:51:14 +02:00
# check for the venv
from lib import sanity_check
sanity_check.check_venv(__file__)
from tornado import gen, httpclient, httputil, web
from tornado.ioloop import IOLoop
TOOLS_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, os.path.dirname(TOOLS_DIR))
from tools.lib.test_script import assert_provisioning_status_ok
if 'posix' in os.name and os.geteuid() == 0:
raise RuntimeError("run-dev.py should not be run as root.")
DESCRIPTION = '''
Starts the app listening on localhost, for local development.
This script launches the Django and Tornado servers, then runs a reverse proxy
which serves to both of them. After it's all up and running, browse to
http://localhost:9991/
Note that, while runserver and runtornado have the usual auto-restarting
behavior, the reverse proxy itself does *not* automatically restart on changes
to this file.
'''
parser = argparse.ArgumentParser(description=DESCRIPTION,
formatter_class=argparse.RawTextHelpFormatter)
parser.add_argument('--test',
action='store_true',
help='Use the testing database and ports')
parser.add_argument('--minify',
action='store_true',
help='Minifies assets for testing in dev')
parser.add_argument('--interface',
action='store',
default=None, help='Set the IP or hostname for the proxy to listen on')
parser.add_argument('--no-clear-memcached',
action='store_false', dest='clear_memcached',
default=True, help='Do not clear memcached')
parser.add_argument('--streamlined',
action="store_true",
default=False, help='Avoid thumbor, etc.')
parser.add_argument('--force',
action="store_true",
default=False, help='Run command despite possible problems.')
parser.add_argument('--enable-tornado-logging',
action="store_true",
default=False, help='Enable access logs from tornado proxy server.')
options = parser.parse_args()
assert_provisioning_status_ok(options.force)
if options.interface is None:
user_id = os.getuid()
user_name = pwd.getpwuid(user_id).pw_name
if user_name in ["vagrant", "zulipdev"]:
# In the Vagrant development environment, we need to listen on
# all ports, and it's safe to do so, because Vagrant is only
# exposing certain guest ports (by default just 9991) to the
# host. The same argument applies to the remote development
# servers using username "zulipdev".
options.interface = None
else:
# Otherwise, only listen to requests on localhost for security.
options.interface = "127.0.0.1"
elif options.interface == "":
options.interface = None
runserver_args: List[str] = []
2016-11-02 04:27:44 +01:00
base_port = 9991
if options.test:
2016-11-02 04:27:44 +01:00
base_port = 9981
settings_module = "zproject.test_settings"
# Don't auto-reload when running casper tests
runserver_args = ['--noreload']
else:
settings_module = "zproject.settings"
manage_args = [f'--settings={settings_module}']
os.environ['DJANGO_SETTINGS_MODULE'] = settings_module
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
from scripts.lib.zulip_tools import CYAN, ENDC, FAIL, WARNING
2016-11-02 04:27:44 +01:00
proxy_port = base_port
django_port = base_port + 1
tornado_port = base_port + 2
webpack_port = base_port + 3
thumbor_port = base_port + 4
os.chdir(os.path.join(os.path.dirname(__file__), '..'))
# Clean up stale .pyc files etc.
subprocess.check_call('./tools/clean-repo')
if options.clear_memcached:
print("Clearing memcached ...")
subprocess.check_call('./scripts/setup/flush-memcached')
# Set up a new process group, so that we can later kill run{server,tornado}
# and all of the processes they spawn.
os.setpgrp()
# Save pid of parent process to the pid file. It can be used later by
# tools/stop-run-dev to kill the server without having to find the
# terminal in question.
if options.test:
pid_file_path = os.path.join(os.path.join(os.getcwd(), 'var/casper/run_dev.pid'))
else:
pid_file_path = os.path.join(os.path.join(os.getcwd(), 'var/run/run_dev.pid'))
# Required for compatibility python versions.
if not os.path.exists(os.path.dirname(pid_file_path)):
os.makedirs(os.path.dirname(pid_file_path))
with open(pid_file_path, 'w+') as f:
f.write(str(os.getpgrp()) + "\n")
def server_processes() -> List[List[str]]:
main_cmds = [
['./manage.py', 'runserver'] +
manage_args + runserver_args + [f'127.0.0.1:{django_port}'],
['env', 'PYTHONUNBUFFERED=1', './manage.py', 'runtornado'] +
manage_args + [f'127.0.0.1:{tornado_port}'],
]
if options.streamlined:
# The streamlined operation allows us to do many
# things, but search/thumbor/etc. features won't work.
return main_cmds
other_cmds = [
['./manage.py', 'process_queue', '--all'] + manage_args,
['env', 'PGHOST=127.0.0.1', # Force password authentication using .pgpass
'./puppet/zulip/files/postgresql/process_fts_updates', '--quiet'],
['./manage.py', 'deliver_scheduled_messages'],
['/srv/zulip-thumbor-venv/bin/thumbor', '-c', './zthumbor/thumbor.conf',
'-p', f'{thumbor_port}'],
]
# NORMAL (but slower) operation:
return main_cmds + other_cmds
def do_one_time_webpack_compile() -> None:
# We just need to compile webpack assets once at startup, not run a daemon,
# in test mode. Additionally, webpack-dev-server doesn't support running 2
# copies on the same system, so this model lets us run the casper tests
# with a running development server.
tests: Clean up calls to tools/webpack. Before this change, the way we loaded webpack for various tools was brittle. First, I addressed test-api and test-help-documentation. These tools used to be unable to run standalone on a clean provision, because they were (indirectly) calling tools/webpack without the `--test` option. The problem was a bit obscure, since running things like `./tools/test-backend` or `./tools/test-all` in your workflow would create `./var/webpack-stats-test.json` for the broken tools (and then they would work). The tools themselves weren't broken; they were the only relying on the common `test_server_running` helper. And even that helper wasn't broken; it was just that `run-dev.py` wasn't respecting the `--test` option. So I made it so that `./tools/run-dev` passes in `--test` to `./tools/webpack`. To confuse matters even more, for some reason Casper uses `./webpack-stats-production.json` via various hacks for its webpack configuration, so when I fixed the other tests, it broke Casper. Here is the Casper-related hack in zproject/test_settings.py, which was in place before my change and remains after it: if CASPER_TESTS: WEBPACK_FILE = 'webpack-stats-production.json' else: WEBPACK_FILE = os.path.join('var', 'webpack-stats-test.json') I added similar logic in tools/webpack: if "CASPER_TESTS" in os.environ: build_for_prod_or_casper(args.quiet) I also made the helper functions in `./tools/webpack` have nicer names. So, now tools should all be able to run standalone and not rely on previous tools creating webpack stats files for them and leaving them in the file system. That's good. Things are still a bit janky, though. It's not completely clear to me why `test-js-with-casper` should work off of a different webpack configuration than the other tests. For now most of the jankiness is around Casper, and we have hacks in two different places, `zproject/test_settings.py` and `tools/webpack` to force it to use the production stats file instead of the "test" one, even though Casper uses test-like settings for other things like which database you're using.
2018-09-06 21:01:45 +02:00
subprocess.check_call(['./tools/webpack', '--quiet', '--test'])
def start_webpack_watcher() -> None:
webpack_cmd = ['./tools/webpack', '--watch', '--port', str(webpack_port)]
if options.minify:
webpack_cmd.append('--minify')
if options.interface is None:
# If interface is None and we're listening on all ports, we also need
# to disable the webpack host check so that webpack will serve assets.
webpack_cmd.append('--disable-host-check')
if options.interface:
webpack_cmd += ["--host", options.interface]
else:
webpack_cmd += ["--host", "0.0.0.0"]
subprocess.Popen(webpack_cmd)
def transform_url(protocol: str, path: str, query: str, target_port: int, target_host: str) -> str:
# generate url with target host
host = ":".join((target_host, str(target_port)))
# Here we are going to rewrite the path a bit so that it is in parity with
# what we will have for production
if path.startswith('/thumbor'):
path = path[len('/thumbor'):]
newpath = urlunparse((protocol, host, path, '', query, ''))
return newpath
@gen.engine
def fetch_request(url: str, callback: Any, **kwargs: Any) -> "Generator[Callable[..., Any], Any, None]":
# use large timeouts to handle polling requests
req = httpclient.HTTPRequest(
url,
connect_timeout=240.0,
request_timeout=240.0,
decompress_response=False,
**kwargs,
)
client = httpclient.AsyncHTTPClient()
# wait for response
response = yield gen.Task(client.fetch, req)
callback(response)
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
class BaseHandler(web.RequestHandler):
# target server ip
target_host: str = '127.0.0.1'
# target server port
target_port: int
def _add_request_headers(
self, exclude_lower_headers_list: Sequence[str] = [],
) -> httputil.HTTPHeaders:
headers = httputil.HTTPHeaders()
for header, v in self.request.headers.get_all():
if header.lower() not in exclude_lower_headers_list:
headers.add(header, v)
return headers
def get(self) -> None:
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
pass
def head(self) -> None:
pass
def post(self) -> None:
pass
def put(self) -> None:
pass
def patch(self) -> None:
pass
def options(self) -> None:
pass
def delete(self) -> None:
pass
def handle_response(self, response: Any) -> None:
if response.error and not isinstance(response.error, httpclient.HTTPError):
self.set_status(500)
self.write('Internal server error:\n' + str(response.error))
else:
self.set_status(response.code, response.reason)
self._headers = httputil.HTTPHeaders() # clear tornado default header
for header, v in response.headers.get_all():
# some header appear multiple times, eg 'Set-Cookie'
self.add_header(header, v)
if response.body:
self.write(response.body)
self.finish()
@web.asynchronous
def prepare(self) -> None:
if 'X-REAL-IP' not in self.request.headers:
self.request.headers['X-REAL-IP'] = self.request.remote_ip
if 'X-FORWARDED_PORT' not in self.request.headers:
self.request.headers['X-FORWARDED-PORT'] = str(proxy_port)
url = transform_url(
self.request.protocol,
self.request.path,
self.request.query,
self.target_port,
self.target_host,
)
try:
fetch_request(
url=url,
callback=self.handle_response,
method=self.request.method,
headers=self._add_request_headers(["upgrade-insecure-requests"]),
follow_redirects=False,
body=getattr(self.request, 'body'),
allow_nonstandard_methods=True,
)
except httpclient.HTTPError as e:
if hasattr(e, 'response') and e.response:
self.handle_response(e.response)
else:
self.set_status(500)
self.write('Internal server error:\n' + str(e))
self.finish()
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
class WebPackHandler(BaseHandler):
target_port = webpack_port
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
class DjangoHandler(BaseHandler):
target_port = django_port
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
class TornadoHandler(BaseHandler):
target_port = tornado_port
dependencies: Remove WebSockets system for sending messages. Zulip has had a small use of WebSockets (specifically, for the code path of sending messages, via the webapp only) since ~2013. We originally added this use of WebSockets in the hope that the latency benefits of doing so would allow us to avoid implementing a markdown local echo; they were not. Further, HTTP/2 may have eliminated the latency difference we hoped to exploit by using WebSockets in any case. While we’d originally imagined using WebSockets for other endpoints, there was never a good justification for moving more components to the WebSockets system. This WebSockets code path had a lot of downsides/complexity, including: * The messy hack involving constructing an emulated request object to hook into doing Django requests. * The `message_senders` queue processor system, which increases RAM needs and must be provisioned independently from the rest of the server). * A duplicate check_send_receive_time Nagios test specific to WebSockets. * The requirement for users to have their firewalls/NATs allow WebSocket connections, and a setting to disable them for networks where WebSockets don’t work. * Dependencies on the SockJS family of libraries, which has at times been poorly maintained, and periodically throws random JavaScript exceptions in our production environments without a deep enough traceback to effectively investigate. * A total of about 1600 lines of our code related to the feature. * Increased load on the Tornado system, especially around a Zulip server restart, and especially for large installations like zulipchat.com, resulting in extra delay before messages can be sent again. As detailed in https://github.com/zulip/zulip/pull/12862#issuecomment-536152397, it appears that removing WebSockets moderately increases the time it takes for the `send_message` API query to return from the server, but does not significantly change the time between when a message is sent and when it is received by clients. We don’t understand the reason for that change (suggesting the possibility of a measurement error), and even if it is a real change, we consider that potential small latency regression to be acceptable. If we later want WebSockets, we’ll likely want to just use Django Channels. Signed-off-by: Anders Kaseorg <anders@zulipchat.com>
2019-07-23 01:43:40 +02:00
class ThumborHandler(BaseHandler):
target_port = thumbor_port
class ErrorHandler(BaseHandler):
@web.asynchronous
def prepare(self) -> None:
print(FAIL + 'Unexpected request: ' + ENDC, self.request.path)
self.set_status(500)
self.write('path not supported')
self.finish()
def using_thumbor() -> bool:
return not options.streamlined
class Application(web.Application):
def __init__(self, enable_logging: bool = False) -> None:
handlers = [
(r"/json/events.*", TornadoHandler),
(r"/api/v1/events.*", TornadoHandler),
(r"/webpack.*", WebPackHandler),
(r"/thumbor.*", ThumborHandler if using_thumbor() else ErrorHandler),
(r"/.*", DjangoHandler),
]
super().__init__(handlers, enable_logging=enable_logging)
def log_request(self, handler: BaseHandler) -> None:
if self.settings['enable_logging']:
super().log_request(handler)
def on_shutdown() -> None:
IOLoop.instance().stop()
def shutdown_handler(*args: Any, **kwargs: Any) -> None:
io_loop = IOLoop.instance()
if io_loop._callbacks:
io_loop.call_later(1, shutdown_handler)
else:
io_loop.stop()
def print_listeners() -> None:
print("\nZulip services will listen on ports:")
ports = [
(proxy_port, CYAN + 'web proxy' + ENDC),
(django_port, 'Django'),
(tornado_port, 'Tornado'),
]
if not options.test:
ports.append((webpack_port, 'webpack'))
if using_thumbor():
ports.append((thumbor_port, 'Thumbor'))
for port, label in ports:
print(f' {port}: {label}')
print()
proxy_warning = f"Only the proxy port ({proxy_port}) is exposed."
print(WARNING + "Note to Vagrant users: " + ENDC + proxy_warning + '\n')
if options.test:
do_one_time_webpack_compile()
else:
start_webpack_watcher()
for cmd in server_processes():
subprocess.Popen(cmd)
try:
app = Application(enable_logging=options.enable_tornado_logging)
try:
app.listen(proxy_port, address=options.interface)
except OSError as e:
if e.errno == 98:
print('\n\nERROR: You probably have another server running!!!\n\n')
raise
print_listeners()
ioloop = IOLoop.instance()
for s in (signal.SIGINT, signal.SIGTERM):
signal.signal(s, shutdown_handler)
ioloop.start()
except Exception:
# Print the traceback before we get SIGTERM and die.
traceback.print_exc()
raise
finally:
# Kill everything in our process group.
os.killpg(0, signal.SIGTERM)
# Remove pid file when development server closed correctly.
os.remove(pid_file_path)