hennge / aapns Goto Github PK

View Code? Open in Web Editor NEW

12.0 34.0 2.0 284 KB

Asynchronous Apple Push Notifications

Home Page: https://aapns.readthedocs.io/en/latest/

License: Other

Python 96.42% Go 3.58%

asyncio apns2 push-notifications python3 apns apple ios

aapns's Introduction

AAPNS

Asynchronous Apple Push Notification Service client.

Requires TLS 1.2 or better
Requires Python 3.8 or better

Quickstart

from aapns.api import Server
from aapns.config import Priority
from aapns.models import Notification, Alert, Localized

async def send_hello_world():
    client = await Server.production('/path/to/push/cert.pem').create_client()
    apns_id = await client.send_notification(
        'my-device-token',
        Notification(
            alert=Alert(
                body=Localized(
                    key='Hello World!',
                    args=['foo', 'bar']
                ),
            ),
            badge=42
        ),
        priority=Priority.immediately
    )
    print(f'Sent push notification with ID {apns_id}')
    await client.close()

aapns's People

Contributors

Stargazers

Watchers

Forkers

st3fan michaelaw320

aapns's Issues

NPN protocol does not work

NPN protocol does not initialize even if openssl version is > 1.0.1 (default initialization without explicit ssl_context passing)

aapns version 19.1

Error log:

The NPN extension requires OpenSSL 1.0.1 or later.

System:
Python 3.7.2 Docker, alpine 3.9 and alpine 3.8
openssl versions tested 1.0.2q on alpine 3.8
1.1.1a on alpine 3.9

NPN successfully initialized on mac with LibreSSL 2.6.5

Maybe must go JWT by 2020-11

Pr0Ger/PyAPNs2#103 (comment) states:

Starting from November 2020 Apple will no longer support cert-based authentication and will only support JWT APNS communication.

Invalid input ConnectionInputs.RECV_HEADERS in state ConnectionState.CLOSED

Got this error and traceback in production:

ProtocolError
Invalid input ConnectionInputs.RECV_HEADERS in state ConnectionState.CLOSED

h2/connection.py in process_input at line 228

h2/connection.py in _receive_headers_frame at line 1554

h2/connection.py in _receive_frame at line 1486

h2/connection.py in receive_data at line 1463

aapns/connection.py in background_read at line 300

Current code in question is:

aapns/src/aapns/connection.py

Lines 291 to 301 in 84ce577

 while not self.closed: 

 data = await self.read_stream.read(2 ** 16) 

 if not data: 

 raise ConnectionError("Server closed the connection") 

 for event in self.protocol.receive_data(data): 

 logger.debug("APN: %s", event) 

 stream_id = getattr(event, "stream_id", 0) 

 error = getattr(event, "error_code", None) 

 channel = self.channels.get(stream_id)

Also, looks like that prod was not running the newest aapns 🤔

APNS token-based authentication support (*.p8 file)

Instead of certificate-based authentication Apple also offers stateless token-based authentication, which is recommended way of connecting to APNS. It does not require a client certificate, so it never expires.

Here’s an article describing this method:
https://developer.apple.com/documentation/usernotifications/setting_up_a_remote_notification_server/establishing_a_token-based_connection_to_apns

And a post by python-apns author showing the implementation:
http://gobiko.com/blog/token-based-authentication-http2-example-apns/

Provide project metadata and description

In short, make https://pypi.org/project/aapns/ nice :)

add support for `mutable-content` in notifications

See https://developer.apple.com/documentation/usernotifications/modifying_content_in_newly_delivered_notifications

Move on from circleci and custom docker image

This has to go:

./.circleci/config.yml:      - image: ojii/aapns:latest`

typo in interpolation

[aapns] Exception: ('Request timed out: %s', 'timeout')

Looks like it should be "xx %s" % foo and not "xx %s", foo 😅

RuntimeError('read() called while another coroutine is already waiting for incoming data')

Happens when doing concurrent request to HTTP/2 endpoint

encode/httpx#527
encode/httpx#535

Update aapns when PR is merged and released

Separate request timeout for request and pool

Sometimes, the connection to APN just dies with much data in the pipe, in which case the last sent request is way ahead of last processed response:

Connection(host='api.push.apple.com', port=443, ᠁ channels={38907: Channel(wakeup=<asyncio.locks.Event object at 0x7f135abdb7f0 [unset, waiters:1]>,᠁}, ᠁closed=False, outcome=None, max_concurrent_streams=1000, last_stream_id_got=38907, last_stream_id_sent=38907

vs.

Channel(wakeup=<asyncio.locks.Event ᠁ [set]>, events=[<ResponseReceived stream_id:38645, headers:[(':status', '200'), ᠁]>, ]

Thus, 131 requests were lost in the void (Δ/2 as outbound requests have odd id's).

raw-h2: handle server closing connection

Today

Bad error <ConnectionTerminated error_code:ErrorCodes.NO_ERROR, last_stream_id:0, additional_data:7b22726561736f6e223a22426164436572746966>

The hex translates to {"reason":"BadCertif...

To do

report connection being closed
decode reason in additional data
[possibly] treat this as graceful shutdown, when h2 is ready python-hyper/h2#1181
[possibly] refuse to open new connections with same cert, or
[possibly] throttle connection open rate, as not to flood the server

Handle graceful shutdown

if/when python-hyper/h2#1181 happens, aapns should handle graceful connection shutdown.

Add `apns-expiration` request header

https://developer.apple.com/documentation/usernotifications/setting_up_a_remote_notification_server/sending_notification_requests_to_apns/

Add apns-expiration request header and if it's set, align request timeout/deadline to that value.

raw-h2: pool test / fixture fails

/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/asyncio/base_events.py:1854:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <selectors.KqueueSelector object at 0x7fde205bddc0>, timeout = 0.6137869281683797

    def select(self, timeout=None):
        timeout = None if timeout is None else max(timeout, 0)
        max_ev = len(self._fd_to_key)
        ready = []
        try:
>           kev_list = self._selector.control(None, max_ev, timeout)
E           KeyboardInterrupt

/Library/Frameworks/Python.framework/Versions/3.9/lib/python3.9/selectors.py:558: KeyboardInterrupt
==================================== 13 deselected in 5.90 seconds =====================================
Task was destroyed but it is pending!
task: <Task pending name='bg-write' coro=<Connection.background_write() running at /Users/dima.tisnek/misc/aapns/src/aapns/connection.py:373> wait_for=<Future pending cb=[<TaskWakeupMethWrapper object at 0x7fde18134550>()]>>
Task was destroyed but it is pending!
task: <Task pending name='Task-16' coro=<test_termination() running at /Users/dima.tisnek/misc/aapns/tests/stress/test_pool.py:71> wait_for=<Future pending cb=[<TaskWakeupMethWrapper object at 0x7fde18126070>()]>>
Exception ignored in: <coroutine object test_termination at 0x7fde1811de40>
Traceback (most recent call last):
  File "/Users/dima.tisnek/misc/aapns/tests/stress/test_pool.py", line 72, in test_termination
  File "/Users/dima.tisnek/Library/Caches/pypoetry/virtualenvs/aapns-qVMfpW3G-py3.9/lib/python3.9/site-packages/_pytest/python_api.py", line 715, in __exit__
  File "/Users/dima.tisnek/Library/Caches/pypoetry/virtualenvs/aapns-qVMfpW3G-py3.9/lib/python3.9/site-packages/_pytest/_code/code.py", line 405, in __init__
ImportError: sys.meta_path is None, Python is likely shutting down

Auto reconnect might cause memory leak

I'm not quite sure but it's probably better to ensure await self.connect() block here to be executed exactly once when there are multiple send_notification requests.

investigate httpx as a replacement for aapns.connection

reference: https://github.com/encode/httpx

Analyse competition

Suddenly there's this: https://github.com/Fatal1ty/aioapns and it's used by 3 projects:

https://github.com/matrix-org/sygnal (looks alive)
https://github.com/smart-campus/django-notifier (no code)
https://github.com/reliktor/aiohttp-server (one off?)

Perhaps there's no need for 2 separate projects? 🤔

raw-h2: connection pool design

Background

@ojii and I have 2 use cases for aapns:

batch app badge update (latency is not important; throughput could be, but weirdly is quite small)
real-time actionable notification (latency is key)

I've tested 1 connection using the new raw-h2 branch against apple's development server. Here are the results from my home:

start python, establish 1 connection, send a few notifications, close connections, quit: 0.83s
same as above, +100 requests: 0.98s, thus 𝚫 is 0.15s / 100 req
latency for first few requests: 0.12s
+100, latency for last few requests: 0.36s

Apple allows up to 1000 concurrent requests per connection, but I can't test that because above ~400 requests to the same device, I start getting 429 status code back 😓. The latency in that case jumps to ~1.2s.

The results could be skewed by my having only 1 device and thus testing with 1 device token. It's possible that Apple server serialises requests per device. Apple responds much faster when token is bogus, typically ~0.54s for me.

Specifications

simple use case - few notifications now and then
batch use case - throughput over latency
real-time use case - latency trumps all

I'm thinking following connection [pool] design for these cases:

Simple, connection on demand

initialise and enter context, nothing created yet
on first request create a connection (this makes 1st request slow)
connection is kept around
subsequent requests reuse the connection and are fast
eventually connection dies (today that means pending requests error out)
on next request, create a new connection
exit context, connection, if any, is closed

Batch use case

initialise and enter context, that creates N connections (perhaps N=1)
requests are assigned to connections round-robin
a connection may be blocked (has 1000 concurrent requests), then another is used
when connections die, they are replaced in the background
exit the context, all connections are closed

Real-time use case

initialise and enter context, that creates N connections (N>=2, ideally largish N)
requests are assigned to connections weighted round-robin
a connection does not allow more than e.g. 10 concurrent requests
if a connection is busy, another is used
[perhaps] there should be a skew to round-robin mechanism, i.e. small preference for older connections, lest perfect round-robin allocation result in simultaneous exhaustion of all connections, resulting in latency spike when all connections are to be replaced at the same time
[perhaps] alternatively, connections are retired well before they are exhausted
when connections die they are replaced in the background
exit the context, all connections are closed

Add readme to pypi description

Today pypi:aapns states:

Project description
The author of this package has not provided a project description

Meanwhile, we've got a nice readme...

Make `click` an optional dep.

Docs state:

Installation
pip install aapns

If you wish to use the command line interface, use pip install aapns[cli]

However, pyproject.toml add click regardless today.
The [cli] part needs to be done.

	while not self.closed:
	data = await self.read_stream.read(2 ** 16)
	if not data:
	raise ConnectionError("Server closed the connection")

	for event in self.protocol.receive_data(data):
	logger.debug("APN: %s", event)
	stream_id = getattr(event, "stream_id", 0)
	error = getattr(event, "error_code", None)
	channel = self.channels.get(stream_id)