Max AIO

As part of the dataplane operation, StorageOS uses Linux AIO (Asynchronous Input Output) contexts to serve I/O requests without blocking. StorageOS requires 4 AIO contexts per deployment (i.e. a StorageOS volume deployment, whether master or replica).

Max AIO prerequisite.

By default there is a maximum number of AIO contexts that can be allocated at once.

The current and maximum number of AIO requests is visible in the virtual files /proc/sys/fs/aio-nr and /proc/sys/fs/aio-max-nr.

The default context limit has been set at 2^16 or 65536. This figure may vary so please check your /proc/sys/fs/aio-max-nr

When aio-nr reaches aio-max-nr the io_setup syscall will fail with EAGAIN. For more information please see the Linux kernel docs here.

Why is this relevant?

As StorageOS requires 4 AIO contexts per deployed volume, there is a limit to the number of volumes that can be deployed per node. Trying to provision additional deployments once the aio-max-nr has been reached will fail as the kernel will be unable to create enough new AIO contexts.

Increasing your AIO context cap.

If your nodes aio-max-nr is set too low you can either provision additional nodes to reduce the number of deployments per node, or increase the aio-max-nr kernel parameter.

You can do this by editing your /etc/sysctl.conf file with the following example line:

fs.aio-max-nr = 1048576

To activate the new settings, run the following command:

$ sysctl -p /etc/sysctl.conf