rewrite: Implement shard IPC server.

This commit is contained in:
2022-11-03 15:36:39 +02:00
parent 88861f3880
commit fd04b825f2
10 changed files with 460 additions and 17 deletions

0
bot/meta/ipc/__init__.py Normal file
View File

200
bot/meta/ipc/client.py Normal file
View File

@@ -0,0 +1,200 @@
from typing import Optional
import asyncio
import logging
import pickle
logging.basicConfig(level=logging.DEBUG)
logger = logging.getLogger(__name__)
class AppClient:
routes = {} # route_name -> Callable[Any, Awaitable[Any]]
def __init__(self, appid, client_address, server_address):
self.appid = appid
self.address = client_address
self.server_address = server_address
self.peers = {appid: client_address} # appid -> address
self._listener: Optional[asyncio.Server] = None # Local client server
self._server = None # Connection to the registry server
self.register_route('new_peer')(self.new_peer)
self.register_route('drop_peer')(self.drop_peer)
self.register_route('peer_list')(self.peer_list)
def register_route(self, name=None):
def wrapper(coro):
route = AppRoute(coro, name)
self.routes[route.name] = route
return route
return wrapper
async def server_connection(self):
"""Establish a connection to the registry server"""
try:
reader, writer = await asyncio.open_connection(**self.server_address)
payload = ('connect', (), {'appid': self.appid, 'address': self.address})
writer.write(pickle.dumps(payload))
writer.write(b'\n')
await writer.drain()
data = await reader.readline()
peers = pickle.loads(data)
self.peers = peers
self._server = (reader, writer)
except Exception:
logger.exception("Could not connect to registry server. Trying again in 30 seconds.")
await asyncio.sleep(30)
asyncio.create_task(self.server_connection())
else:
logger.info("Connected to the registry server, launching keepalive.")
asyncio.create_task(self._server_keepalive())
async def _server_keepalive(self):
if self._server is None:
raise ValueError("Cannot keepalive non-existent server!")
reader, write = self._server
try:
await reader.read()
except Exception:
logger.exception("Lost connection to address server. Reconnecting...")
else:
# Connection ended or broke
logger.info("Lost connection to address server. Reconnecting...")
await asyncio.sleep(30)
asyncio.create_task(self.server_connection())
async def new_peer(self, appid, address):
self.peers[appid] = address
async def peer_list(self, peers):
self.peers = peers
async def drop_peer(self, appid):
self.peers.pop(appid, None)
async def close(self):
# Close connection to the server
# TODO
...
async def request(self, appid, payload: 'AppPayload'):
try:
if appid not in self.peers:
raise ValueError(f"Peer '{appid}' not found.")
logger.debug(f"Sending request to app '{appid}' with payload {payload}")
address = self.peers[appid]
reader, writer = await asyncio.open_connection(**address)
writer.write(payload.encoded())
await writer.drain()
writer.write_eof()
result = await reader.read()
writer.close()
decoded = payload.route.decode(result)
return decoded
except Exception:
logging.exception(f"Failed to send request to {appid}'")
return None
async def requestall(self, payload):
results = await asyncio.gather(*(self.request(appid, payload) for appid in self.peers))
return dict(zip(self.peers.keys(), results))
async def handle_request(self, reader, writer):
data = await reader.read()
loaded = pickle.loads(data)
route, args, kwargs = loaded
logger.debug(f"AppClient {self.appid} handling request on route '{route}' with args {args} and kwargs {kwargs}")
if route in self.routes:
try:
await self.routes[route].run((reader, writer), args, kwargs)
except Exception:
logger.exception(f"Fatal exception during route '{route}'. This should never happen!")
else:
logger.warning(f"Appclient '{self.appid}' recieved unknown route {route}. Ignoring.")
writer.write_eof()
async def connect(self):
"""
Start the local peer server.
Connect to the address server.
"""
# Start the client server
self._listener = await asyncio.start_server(self.handle_request, **self.address, start_serving=True)
logger.info(f"Serving on {self.address}")
await self.server_connection()
class AppPayload:
__slots__ = ('route', 'args', 'kwargs')
def __init__(self, route, *args, **kwargs):
self.route = route
self.args = args
self.kwargs = kwargs
def __await__(self):
return self.route.execute(*self.args, **self.kwargs).__await__()
def encoded(self):
return pickle.dumps((self.route.name, self.args, self.kwargs))
class AppRoute:
__slots__ = ('func', 'name')
def __init__(self, func, name=None):
self.func = func
self.name = name or func.__name__
def __call__(self, *args, **kwargs):
return AppPayload(self, *args, **kwargs)
def encode(self, output):
return pickle.dumps(output)
def decode(self, encoded):
# TODO: Handle exceptions here somehow
if len(encoded) > 0:
return pickle.loads(encoded)
else:
return ''
def encoder(self, func):
self.encode = func
def decoder(self, func):
self.decode = func
async def execute(self, *args, **kwargs):
"""
Execute the underlying function, with the given arguments.
"""
return await self.func(*args, **kwargs)
async def run(self, connection, args, kwargs):
"""
Run the route, with the given arguments, using the given connection.
"""
# TODO: ContextVar here for logging? Or in handle_request?
# Get encoded result
# TODO: handle exceptions in the execution process
try:
result = await self.execute(*args, **kwargs)
payload = self.encode(result)
except Exception:
logger.exception(f"Exception occured running route '{self.name}' with args: {args} and kwargs: {kwargs}")
payload = b''
_, writer = connection
writer.write(payload)
await writer.drain()
writer.close()

173
bot/meta/ipc/server.py Normal file
View File

@@ -0,0 +1,173 @@
import asyncio
import pickle
import logging
import string
import random
from ..logger import log_action, log_context, log_app
logger = logging.getLogger(__name__)
uuid_alphabet = string.ascii_lowercase + string.digits
def short_uuid():
return ''.join(random.choices(uuid_alphabet, k=10))
class AppServer:
routes = {} # route name -> bound method
def __init__(self):
self.clients = {} # AppID -> (info, connection)
self.route('ping')(self.route_ping)
self.route('whereis')(self.route_whereis)
self.route('peers')(self.route_peers)
self.route('connect')(self.client_connection)
@classmethod
def route(cls, route_name):
"""
Decorator to add a route to the server.
"""
def wrapper(coro):
cls.routes[route_name] = coro
return coro
return wrapper
async def route_ping(self, connection):
"""
Pong.
"""
reader, writer = connection
writer.write(b"Pong")
writer.write_eof()
async def route_whereis(self, connection, appid):
"""
Return an address for the given client appid.
Returns None if the client does not have a connection.
"""
reader, writer = connection
if appid in self.clients:
writer.write(pickle.dumps(self.clients[appid][0]))
else:
writer.write(b'')
writer.write_eof()
async def route_peers(self, connection):
"""
Send back a map of current peers.
"""
reader, writer = connection
peers = self.peer_list()
payload = pickle.dumps(('peer_list', (peers,)))
writer.write(payload)
writer.write_eof()
async def client_connection(self, connection, appid, address):
"""
Register and hold a new client connection.
"""
log_action.set("CONN " + appid)
reader, writer = connection
# Add the new client
self.clients[appid] = (address, connection)
# Send the new client a client list
peers = self.peer_list()
writer.write(pickle.dumps(peers))
writer.write(b'\n')
await writer.drain()
# Announce the new client to everyone
await self.broadcast('new_peer', (), {'appid': appid, 'address': address})
# Keep the connection open until socket closed or EOF (indicating client death)
try:
await reader.read()
finally:
# Connection ended or it broke
logger.info(f"Lost client '{appid}'")
await self.deregister_client(appid)
async def handle_connection(self, reader, writer):
data = await reader.readline()
route, args, kwargs = pickle.loads(data)
rqid = short_uuid()
log_context.set("RQID:" + rqid)
log_action.set("SERV ROUTE " + route)
logger.info(f"AppServer handling request on route '{route}' with args {args} and kwargs {kwargs}")
if route in self.routes:
# Execute route
try:
await self.routes[route]((reader, writer), *args, **kwargs)
except Exception:
logger.exception(f"AppServer recieved exception during route '{route}'")
else:
logger.warning(f"AppServer recieved unknown route '{route}'. Ignoring.")
def peer_list(self):
return {appid: address for appid, (address, _) in self.clients.items()}
async def deregister_client(self, appid):
self.clients.pop(appid, None)
await self.broadcast('drop_peer', (), {'appid': appid})
async def broadcast(self, route, args, kwargs):
logger.debug(f"Sending broadcast on route '{route}' with args {args} and kwargs {kwargs}.")
payload = pickle.dumps((route, args, kwargs))
if self.clients:
await asyncio.gather(
*(self._send(appid, payload) for appid in self.clients),
return_exceptions=True
)
async def message_client(self, appid, route, args, kwargs):
"""
Send a message to client `appid` along `route` with given arguments.
"""
logger.debug(f"Sending '{route}' to '{appid}' with args {args} and kwargs {kwargs}.")
if appid not in self.clients:
raise ValueError(f"Client '{appid}' is not connected.")
payload = pickle.dumps((route, args, kwargs))
return await self._send(appid, payload)
async def _send(self, appid, payload):
"""
Send the encoded `payload` to the client `appid`.
"""
address, _ = self.clients[appid]
try:
reader, writer = await asyncio.open_connection(**address)
writer.write(payload)
writer.write_eof()
await writer.drain()
writer.close()
except Exception as ex:
# TODO: Close client if we can't connect?
logger.exception(f"Failed to send message to '{appid}'")
raise ex
async def start(self, address):
log_app.set("APPSERVER")
server = await asyncio.start_server(self.handle_connection, **address)
logger.info(f"Serving on {address}")
async with server:
await server.serve_forever()
async def start_server():
address = {'host': '127.0.0.1', 'port': '5000'}
server = AppServer()
await server.start(address)
if __name__ == '__main__':
asyncio.run(start_server())