Fail to create agents with GPU attached from Machine Configuration in Jenkins settings UI. Instances with guest accelerators do not support live migration. about google-compute-engine-plugin HOT 10 OPEN

@rachely3n, yes, (terminate for migrate) should only apply for instances with GPU's.

from google-compute-engine-plugin.

For reference, this issue now tracks the specific work of changing the value of "onHostMaintenance" when the AcceleratorConfiguration is defined in the scheduling() method of InstanceConfiguration.java. This is a small scope to change but because there are workarounds, moving to backlog right now.

from google-compute-engine-plugin.

from google-compute-engine-plugin.

Hmm, it seems like your code change would resolve issues. Is there a reason you have not submitted a pull request?

from google-compute-engine-plugin.

@rachely3n it is workaround for GCP instances with GPUs because GCP instances without GPUs can migrate to other hardware without downtime.

from google-compute-engine-plugin.

I see, so that (terminate for migrate) should only apply for instances with GPU's?

from google-compute-engine-plugin.

Given that this can be accomplished with instance templates, closing.

from google-compute-engine-plugin.

@stephenashank, Will GPU support in Machine configuration UI be available only with instance templates? If yes, then will GPU controls in Machine configuration UI other than instance templates will be disabled? #68 (comment)

from google-compute-engine-plugin.

I'm revisiting this today, and realize that it is possible still to create instances with GPUs without using instance templates. By checking "preemptible", this also by definition means that it will terminate on host maintenance. The disadvantage of this is that your instance could possibly be terminated even without maintenance. The other workaround is to use an instance template, which provides the desired flexibility but requires you to use multiple interfaces for configuration.

I would prefer not to remove this feature. In my opinion the ideal solution, as mentioned here and in #68, is to change the value of "onHostMaintenance" if GPUs are configured.

from google-compute-engine-plugin.

What about adding another option for host maintenance policy?

Does it make sense to have better parity with the GCP console options?

from google-compute-engine-plugin.