Upgrading Azure Machine Learning Compute Instance (virtual machine)

Question

Hi,

How can I safely upgrade the default OS and softwares of a compute instance (Azure Machine Learning)?

The default compute instance comes with pre-default versions for OS and packages.

For example, the following are a few which I’ve inspected after spinning on a compute instance:

1. OS is currently an Ubuntu LTS 16.04 (release date February 28, 2019)
2. Python version 3.6.9 (release date July 2, 2019)
3. R version 3.6.3 (release date February 29, 2020)

The latest stable versions of the above (as of today June 4,2020) are:

1. Ubuntu 20.04 LTS (release date April 23,2020)
2. Python version 3.7.7 (release date March 10,2020)
3. R version 4.0 (release date April 24,2020)

I actually tried upgrading the OS through the command line but it doesn't seem to be stable after the upgrade to Ubuntu LTS 18.04.
Upgrading RStudio server also fails.

So I'm not entirely confident whether the usual commands to upgrade work.
Is there anything I need to do with the repository list to ensure smooth upgrades and updates?

Thanks

Answer 1

Hello,

From Azure portal, go to the VM page. Under Operations, Select 'Update Management'. Select 'Enable for this VM'. It will take couple of minutes to finish apply settings. Once finished, you may be able to see the missing updates at the same page. Click 'Schedule Update Deployment' > Select the options needed and schedule it as per your convenience

Please mark as "Accept the answer" if the above steps helps you. Others with similar issues can also follow the solution as per your suggestion

Regards,

Manu

Answer 2

@AfiqCheJohari-9488 Are there any specific steps you followed to upgrade your VM version?
From the documentation of RStudio 20.04 is not in the list of supported operating systems as mentioned here.

Answer 3

It would be great to have a reliable way to upgrade the Azure ML SDK and its dependencies on such a compute instance (or any existing conda environment for that matter) without having to abandon and recreate it from scratch, also redoing all the extra setup necessary by the user...

In my experience, just following the standard documentation pip upgrade instruction almost always messes up the conda environment and introduces multiple dependency errors...
So it would be great to have an upgrade option that just works.
At least to have a valid end-to-end script example in the documentation.