一、含义

表示磁盘的I/O requests queue，即IO占用的时间，单位为毫秒。

参考：SystemDiskRequestQueuedWarning

二、表达式

（1）IO占比

1
rate(diskio_io_time{name=~"node-.*"}[1m]) / 1000

rate计算1分钟内的IO耗时占比，除以1000后单位为秒。意思为一秒内的IO占比。

（2）IO占比阈值

1
rate(diskio_io_time{name=~"node-.*"}[1m]) / 1000 > 0.9

I/O requests for the disk on the node spent 90% of the device time in queue during the last 10 minutes.

IO在1秒内的耗时占比大于0.9（即90%）的数据。

I/O requests queue 占用总的device_time的时间比例

三、作用

（1）间接反应磁盘的繁忙情况

若占比越高，则当前的IO request所占用的时间越长，响应速度则越慢。

参考：io-request-queueing

（2）blocking I/O call

Broadly, a blocking I/O call does this:

• (1) put a process/thread on a list of processes/threads which are waiting for I/O to complete
• (2) mark it as non-runnable
• (3) invoke a context switch.

Blocking： Since the thread which made the blocking call is now marked as non-runnable, the OS’ scheduler will never schedule it to run on the CPU, until the I/O operation completes, and the thread is marked as runnable again.

在IO操作完成后，线程才能继续执行。

（3）non-blocking I/O calls

• (1) Your program first makes a system call to create an “asynchronous I/O context”.

This is an object which includes a queue for notification events generated when an I/O operation completes.

创建一个系统调用，包含IO操作结束时生成的通知事件队列。

• (2) The kernel passes back an AIO context identifier which your program uses when making subsequent calls.

后续调用时，内核回传上步骤创建的AIO标识符

• (3) Then, when you want to do some I/O, you make another system call to “submit” an async I/O operation for execution. You can submit as many of these as you like.

当需要IO操作时，创建一个新的系统调用来“提交”一个异步操作，可以提交任意多个数量。

• (4) As they complete, notifications will accumulate in your AIO context’s event queue.

当它们完成时，通知将累积在您的 AIO 上下文的事件队列中。

• (5) Later, whenever you want, you make another syscall to retrieve the notifications from that queue.

之后，你可以在任何时候，通过另外的系统调用从队列中获取通知。