I. Maintenance

Maintenance can define a lot of actions:

Hardware upgrade
Security patch
Update and reboot

Obviously one day, everyone (or good sysadmin) needs to reboot a server.

Several setup and scenario:

VM disks are stored on a distributed filesystem (cephFS, glusterFS, NFS…). Easy to perform a maintenance.
VM disks are stored locally, this is the most difficul part.

II. Put a node into maintenance

There is already a maintenance mode, but the feature has been implemented only for Xen hypervisor, if you run KVM and you issue the following command, you will get an error, which is obvious:

$ nova host-update compute-01 --maintenance enable
ERROR: The server has either erred or is incapable of performing the requested operation. (HTTP 500) (Request-ID: req-5667d1ff-bgha-4veq-9drb-8c6'733f1s4b)

As a workaround you could:

Specify a different location for the new instances. But you can only do this if you manage your Cloud.
Disable a compute node

II.1. Private cloud approach

You can use the following command to force the run on a specific node:

$ nova boot bla bla bla --availability-zone <your-zone>:<compute-node>

R Note: admin permissions are required. and the default zone is `nova`.

II.2. Public Cloud approach

Even better I guess, you could disable the compute node from the service, thus nova-scheduler won’t attemps to run any instance on this compute node.

$ sudo nova-manage service disable --host=<host> --service=<service>

Hope it helps! If someone has another method, please share ;-).