I found an answer and I figured it should be here for future Googling.
The issue is as I first expected IIS proxies the request to kestrel and that means the process is in use as far as Windows is concerned. There are three solutions.
The Good Solution
Have two (or more) VMs on azure behind a load balancer. Have a script which turns off the sites one at a time, does what it needs to do and turns them back on. Do this right and no downtime!
Intermission
Before I talk about the other solutions a little explanation. I have not been working with .NET for a long time but apparently there was this thing you could do where you add a app_offline.htm
and it will temporarily take down the site for you.
In the context of IIS and .Net Core it also releases the process, which is really useful as it solves my problem! Although I had to visit the web page first for it to take effect, unless I'm mistaken.
The Bad Solution
Use an automated script to rename _app_offline.htm
to app_offline.htm
. Do the upgrade and then revert that change. Takes your site down, kind of ugly but scripting is always better than...
The Ugly Solution
You only have access to FTP, no remote admin or proper deployment process because... reasons.
Upload an app_offline.htm
, upload as little as possible and hope it doesn't break anything before deleting or renaming app_offline.htm
.
Also you would have to perform any DB migrations by using EnableAutomaticMigrations = true
because you have no server access or scripting methods.