[autoscaler] GCP support

[hartikainen]

It would be great if the autoscaler supported Google Cloud Platform.

I'm going to start implementing the GCP node provider soon. Thought I would open up this as a feature request to see if anyone else has done work towards this or possibly has any specific thoughts about the implementation (e.g. what should/could change from the aws node provider).

[AdamGleave]

I'd be happy to beta-test this, GCP has some attractive features for my workloads.

[hartikainen]

@AdamGleave Would be great to have some beta-testing help once this gets to master. I'm curious, what are the features that GCP has and AWS does not?

[AdamGleave]

The main one for me is the ability to specify custom instance types. This lets me get a higher CPU:GPU ratio than is possible on AWS, useful for my workloads given how expensive GPUs are on both AWS & GCP.

[richardliaw]

Hey Adam, You can try out the GCP autoscaler by running `ray create_or_update [cluster yaml]` off the current master, using this cluster yaml file: [ray dir]/python/ray/autoscaler/gcp/example-full.yaml I believe you should be able to specify custom instance types via the yaml. Let me know how it goes!

…

On Sat, Jun 2, 2018 at 10:59 AM Adam Gleave ***@***.***> wrote: The main one for me is the ability to specify custom instance types. This lets me get a higher CPU:GPU ratio than is possible on AWS, useful for my workloads given how expensive GPUs are on both AWS & GCP. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2049 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEUc5RSwyUwYr604a74Y6pJ0gcfAbopiks5t4tKSgaJpZM4T88Y6> .

[richardliaw]

This is merged in master, so closing this for now.