OS: SELECT(2), POLL(2), and their variants

Appetizer: What do the numbers in a man page mean?

根据 Andy Dalton & Michael Mrozek:

The man page for man itself (man man) explains it and lists the standard ones.

MANUAL SECTIONS
       The standard sections of the manual include:

       1      User Commands
       2      System Calls
       3      C Library Functions
       4      Devices and Special Files
       5      File Formats and Conventions
       6      Games et. Al.
       7      Miscellanea
       8      System Administration tools and Deamons

       Distributions customize the manual section to their specifics, which often include additional sections.

所以有的文章会用 SELECT(2) 和 POLL(2) 这样的写法来强调 select() 和 poll() 是 System Calls。

SELECT(2) 和 POLL(2)

man select:

select() examines the I/O descriptor sets whose addresses are passed in readfds, writefds, and errorfds to see if some of their descriptors are ready for reading, are ready for writing, or have an exceptional condition pending, respectively.

man poll:

poll() examines a set of file descriptors to see if some of them are ready for I/O or if certain events have occurred on them.

• 这个 man poll 感觉没有抓住实质，我们暂时忽略它

其实记住最典型的应用场景就可以了：假设我是一个 server，我 accept 了很多的 client sockets (which IS-A type of file descriptor)，我怎么确定哪个 client socket 给我发了 message？我如何挑选出这个 client socket 来 reply 它？

Select is fundamentally broken 这篇精彩的 blog 就提到了 The Design and Implementation of the FreeBSD Operating System 这本书针对这个应用场景在 OS 级别的处理策略：

There are four possible alternatives that avoid the blocking problem:

1. Set all the descriptors in nonblocking mode. […]
2. Enable all descriptors of interest to [send] signal when I/O can be done. […]
3. Have the system provide a method for asking which descriptors are capable of performing the I/O […]
4. Have the process register with the [operating] system all the events including I/O on descriptors that it is interested in tracking.

SELECT(2) 和 POLL(2) 实现的是策略 (3):

• POLL(2) 负责收集 descriptors
• SELECT(2) 负责 “asking which descriptors are capable of performing the I/O” from the poll

Variants: EPOLL(7) / devpoll / KQUEUE(2)

SELECT(2) 和 POLL(2) 源于 Unix，后续的变种有:

• Linux 用的是 EPOLL(7)
• Solaris 用的是 poll(7d)，由于它的 path 是 /dev/poll，所以也被叫做 devpoll
• BSD 用的是 KQUEUE(2)
  • 严格来说 KQUEUE(2) 应该不算 POLL(2) 的变种，因为它配合 SELECT(2) 实现的是上面的策略 (4)
  • 但现阶段你理解成它和 POLL(2) 的目的是类似的就可以了

注意：虽然这些变种都仍然使用 SELECT(2)，但目前我不知道它们的 SELECT(2) 是不是有啥区别……

SELECT(2) + POLL(2)/EPOLL(7) 即是 “轮询”

这个 “asking which descriptors are capable of performing the I/O” 并没有多么的 fancy，我甚至觉得它可以简单理解成：

def select(poll, timeout):
    while not timeout:
        for fd in poll:
            if fd is ready:
                return fd
    raise TimeoutError()

“轮询” 这个翻译还是 OK 的，poll 做 verb 本身的意思就有：

• poll sb. => to record the opinion or vote of sb.
• poll sth. => (in Computer Science/Telecommunications) to check the status of sth., especially as part of a repeated cycle.

Select is fundamentally broken 介绍了 POLL(2) 相比 EPOLL(7) 的缺点，但它们在 complexity 上并没有本质的区别，你可以理解成 POLL(2) 是 $O(42 \times n)$，然后 EPOLL(7) 是 $O(7 \times n)$，(从 heavy-weight 变成了 light-weight)，但无法突破 $O(n)$。

关于 EPOLL(7) 可以阅读 Async IO on Linux: select, poll, and epoll，这篇也简略 cover 了很多相关的基础概念。

EPOLL(7) 也有它自己的问题，可以参见 (还是那个作者的)：

• Epoll is fundamentally broken 1/2
• Epoll is fundamentally broken 2/2

这两篇非常的细节，但我觉得主要的 takeaway 应该是：EPOLL(7) 的使用有它的 best practice，当你真正需要直接使用 EPOLL(7) 的时候，再研究也不迟。

Python: select module

我铺了这么多，就是为了引出 select — Waiting for I/O completion 这个 module，现在看它的代码就能很清楚地理解它们的层级关系了。

Python: selectors module

selectors — High-level I/O multiplexing 是建立在 select 之上的一个高一级别的 module，无非就是屏蔽了 select 过于琐碎的技术细节 (毕竟是 OS-specific 的)。docs 也说了：

Users are encouraged to use this module instead, unless they want precise control over the OS-level primitives used.