Deep Redis (10) Thread IO model

Thread IO Model

The first thing to remember is that Redis is a single-threaded thread.

Why is a single thread so fast?

All Redis data is in memory, and all operations are memory-level operations, so the speed is faster than on-disk operations. But it is also because it is single-threaded, so be careful with the time-complexity O (n) instructions.

How does a single thread handle so many concurrent client connections?

Multiplexing.

Non-blocking IO

The socket method reads and writes are blocked by default and can be set to non-blocking IO via Socket.socket (). setblocking (False) in Python.

Time Polling (multiplexing)

The problem with non-blocking IO is that thread read and write data may be returned as part of the process, when the data continues to be read, and when the data continues to be written, there should be a way to notify them.

The time polling API is to solve this problem, the early Select, later the poll, now the Epoll and Kqueque.

  #基于selectimport socketimport selectimport queueserver = Socket.socket () server.bind (' localhost ', 8000)) Server.listen () server.setblocking (0) msg = {}inputs = [server]outputs = []while true:readable, Writea ble, exceptional = select.select (inputs, outputs, inputs) for R in Readable:if R is Server:conn , addr = R.accept () conn.setblocking (0) inputs.append (conn) msg[conn] = queue. Queue () Else:try:data = R.recv (1024x768) msg[r].put (data) out  Puts.append (R) except ConnectionResetError:inputs.remove (R) continue for W in Writeable:data = Msg[w].get () w.send (data) Outputs.remove (W) for E in Exceptional:if e i n Outputs:outputs.remove (E) inputs.remove (e) del msg[e]  

# 自动匹配版select,在linux中自动使用epoll，在windows中只能select，在bsd或mac中自动kqueueimport socketimport selectorsserver = socket.socket()server.bind((‘localhost‘, 8000))server.listen()server.setblocking(0)sel = selectors.DefaultSlector()def accept(server, mask):    conn, addr = server.accept()    conn.setblocking(0)    sel.register(conn, mask, read)def read(conn, mask):    data = conn.recv(1024)    if data:        conn.send(data)    else:        sel.unregister(conn)        conn.close()sel.register(server, selectors.EVENT_READ, accept)while True:    events = sel.select()    for key, mask in events:        callback = key.data        obj = key.fileobj        callback(obj, mask)

But none of this is going to happen, and Redis has implemented this polling mechanism for us, as long as we remember that multiplexing is non-blocking IO, not an asynchronous operation.

Deep Redis (10) Thread IO model