Skip to content

Latest commit

 

History

History
123 lines (92 loc) · 3.39 KB

async.md

File metadata and controls

123 lines (92 loc) · 3.39 KB
Raw

Async

What is async for?

  • Minimises resources for idle connections

C10K

Why is async hard to do?

  • You need to store state
  • Needs to remember for a frontend client, what to do when the backend responds
  • Storing state has two axes:
    • Memory/connection
    • Coding difficulty
  • Callbacks = low mem/connection, difficult to code
  • Threads = high mem/connection, easy to use
  • Still a live and developing area of research

What IS async?

  • Single-threaded
  • Can do I/O concurrency
  • Can't do computation concurrently
  • Non-blocking sockets (epoll)
  • Uses an event loop to continuously process event notifications
  • Examples: Node, Tornado, Twisted, Tulip

asyncio

  • AKA "Tulip"
  • Made by Guido
  • In the Python 3.4 standard lib
  • Standard event loop
  • Coroutines
  • Implements PEP 3156

Async App Example

Layers

  • app
  • protocol -e.g. HTTP
    • example uses autobahn lib
  • transport -e.g. TCP
  • event loop
    • finds out about events and dispatches them to event handlers, which are callbacks
  • selectors
    • abstracts away platform specific differences between event notifications
    • magic

Implementation

  • sock.setblocking(False): all operations will send or receive instantly, but can't block. Will raise an exception if blocking
  • pass the file descriptor of the socket to _selector to abstract to the operating system
  • also pass in a 'reader' and 'writer' callback - what to do on read, or on write
  • for a listening socket, just pass in a reader

_accept_connection

  • takes a protocol_factory (ChatProtocol) and a socket
  • accepts the socket and returns a connected socket
  • calls the protocol factory
  • calls _SelectorSocketTransport with the connected socket and the result of the protocol factory

_SelectorSocketTransport

  • takes connected socket and protocol

BaseEventLoop

  • run_forever is an infinite loop
  • while True, recieves an event list from the _selector
  • list of events that are currently ready on non-blocking sockets
  • these contain file descriptors, mask(?) and data
  • data contains reader and writer callbacks
  • add ready callbacks to a ready queue
  • check length of ready queue
  • iterate through ready queue for the length of the current ready queue
    • this makes sure that any callbacks that add new callbacks aren't executed until the next iteration of the event loop

Review

  • asyncio uses non-blocking sockets
  • Event loop tracks sockets and the callbacks waiting
  • Selectors handles abstraction of events on operating system

When should I use it?

Yes

  • Slow backend
  • Websockets
  • Many connections

No

  • CPU-bound
  • No async driver
  • No async expertise

Resources

Questions

  • Debugging?

    • Callbacks usually lose your traceback
    • Chain of events that led to it being added to the runnable queue is gone
    • Coroutines are better than callbacks for debugging

  • Asynchronous drivers for other kinds of blocking I/O?

    • gevent monkey-patches stdlib I/O with an async layer
    • theres a stdlib email module, then a non-blocking Twisted version
    • will be a few years before there is a separation between business logic and I/O scheduling

  • Why didn't you just untangle existing PyMongo lib?

    • Too intertwined scheduling I/O and business logic