Cloud-init

[miguel-abulencia]

Hello.

We’re deploying the project using a Bicep to ARM template, and we’re trying to set up cloud-init as part of that process. We noticed there’s some cluster_init in the code, but we’re only looking to use cloud-init for now.

Do you have any tips or working example configs for getting cloud-init running with the Bicep/ARM deployment?

Really appreciate any help you can share .

Thanks!

[xpillons]

you want to use cloud-init for which machines ?

[ossonts]

A cloud init for each partition. There is often a few tasks we need to run / install on each node before it becomes available.

[ossonts]

[Bumping this issue]

What is the recommended way to run a script on each node, before a job is started.

Apps/code/files get added to the OS image
But there is often a case where some change is needed on the node, e.g a path set or permissions or something in the home folder, starting a service or setting env vars.

What is the best mechanism to do this?

[xpillons]

@ossonts there are two approach to consider :

1. running a task on each node before a job starts
2. running a task on each node upon node startup

Based on the earlier discussion, you are probably referring to 2. If so the way of doing it is to create a cyclecloud cluster-init project and store it in github. You can find an example here https://github.com/xpillons/ccsw-projects/tree/main/projects/cvmfs-eessi which will mount CVMFS and initialize it so you can stream applications.
Then you must do the following :

• on the CycleCloud VM, git clone the project, and use this command to upload it into the CycleCloud locker. "cyclecloud project upload azure-storage"
• from the CycleCloud portal, edit the cluster template, in the advanced settings, add a your cluster-init project to the list of the existing ones. Save your cluster.
• These new settings will take effect on the next node startup.

In a future release we will make this easier to set as part of the deployment, but today this is unfortunately a manual action. You have the option to not start the cluster so that you can add these customization before starting it.

Let me knows if you have any additional questions.

[ossonts]

Thanks, very helpful.

1. Any tips on which folder I can run the "cyclecloud project upload azure-storage" ?

I see that 2 default projects are installed already and that the main template has a clusterInitSpecs variable. Adding my project here works.

• It is a bit annoying to have to do a release to get an update -- is there another method?
• Also the git repo has to be public, does the code support a PAT token -- I could not see it , but there is a type type: 'gitHubReleaseURL' -- what else is supported
• the url needs a /tag but that is then removed.... seems a bit odd.

How do I upload the files on cycle... i tried --cyclecloud project upload azure-storage

I guess I am in the wrong place, but cant find it.....

[ossonts]

Also I see the scripts use apt rather than apt-get this is better

    apt-get update
    apt-get install -y moreutils

And probably worth setting this env var

DEBIAN_FRONTEND noninteractive

[xpillons]

Also I see the scripts use apt rather than apt-get this is better

    apt-get update
    apt-get install -y moreutils

And probably worth setting this env var

DEBIAN_FRONTEND noninteractive

good catch, however the CycleCloud VM is AlmaLinux, so this code is never called ;-)

[xpillons]

The customClusterInitSpec feature is still WIP so I'm not sure it will work end to end.
What works today is the following :

• open a shell window on the CC VM under root
• git clone your repos
• cd in each repo folder and run "cyclecloud project upload azure-storage"
• open the CycleCloud UI portal, edit the cluster tempate and update the projects for the targeted node and node arrays.
• save the cluster and eventually start or terminate/start the cluster

as said before we are working on automating this in a future release.
thank you