Windows 11 infrastructure regressions

[claraberendsen]

Description

Windows 11 machines are not yet stable with the regressions with the transition to Windows Server 2025.

I've been testing a set of changes to try to stabilize it, and there are some interesting findings:

• On the first run of the job on a machine the number of test regressions is less than on subsequent runs. On this run there are 26 test regressions ->
• On a second run there are 62 test regressions, a lot of new ones having to do with networking and starting services.

This is a consistent behavior tested numerous times.

It's clear that there are cleanup issues between one run and the other, I haven't found a smoking gun on what in particular changed and is making these regressions.

System information

Docker networks on host afer 1+ runs:

docker network ls
NETWORK ID     NAME               DRIVER    SCOPE
e614b316055d   isolated_network   nat       local
841d37e5a6b4   nat                nat       local
21dc2c825cb0   none               null      local

isolated_network is the one created by the jobs and reused subsequently.