Azure Linux 3 VMs not provisioning correctly

[ilyas1974]

As reported in teams, our customers are having issues standing up workloads in the Azure Linux 3 queue.

Error they are seeing

Console log: 'Microsoft.Extensions.Logging.Generators.Roslyn3.11.Tests' from job 67363cb9-a077-426f-b406-8ef5c7ec2ad6 (azurelinux.3.amd64.open.rt) using docker image mcr.microsoft.com/dotnet-buildtools/prereqs:centos-stream8-helix on a00002Krunning $HELIX_CORRELATION_PAYLOAD/scripts/57b07292d7f64d4591ad8e370868206b/execute.sh in /mnt/work/A5D80980/w/A67909B4/e max 900 seconds

Output: Unable to clean up running docker containers
Exit Code:-6

Release Note Category

• Feature changes/additions
• Bug fixes
• Internal Infrastructure Improvements

Release Note Description

restarted generating the azurelinux.3.amd64 image. this will affect all azurelinux.3.amd64* queues. changes address deployment problems (due to a missing Python package) and adds full Docker support.

[dougbu]

fixed in !46122 but reverted in !47331 and reintroduced in !47334

[richlander]

Thanks! I'm looking forward to using these VMs.

[richlander]

Is there any testing we should do ahead of time with real pipelines?

[dougbu]

Is there any testing we should do ahead of time with real pipelines?

dotnet-helix-machines now tests simple Assert.True(true) xUnit scenarios using the Helix SDK. that includes a number of Docker containers w/ the azurelinux.3.amd64 host. I also plan to run somewhat more complicated arcade-validation scenarios on our staging environment today

going further w/ your real tests may help us get started on additional work we need to do (if any). from my perspective, starting early won't make a huge difference but it's up to you

let's chat offline if you want to test during the few days b/4 our production rollout. using our staging environment requires a different token than most pipelines have available

[richlander]

We've had to roll this deployment back twice (and last time broke dotnet/runtime CI all-up). That points to needing a different strategy. Certainly, if it fails a third time, that suggests we're not adapting to signal.

This is very similar to dotnet/dotnet-buildtools-prereqs-docker#1369 (comment). If we merge that PR w/o a test run, then we are at the mercy of the devops gods. We've been successful doing test runs on new prereqs images by pushing them to ACR and creating a PR in dotnet/runtime that consumed those images. I don't have enough context to propose what an analog of that would be for our VMs queues.