Add more instances and node groups to cover all templates by default #41
Conversation
sridharkonduri-99
requested review from
brucez-anyscale and
chrisfellowes-anyscale
github-actions
bot
added
documentation
chrisfellowes-anyscale
left a comment
chrisfellowes-anyscale
left a comment
There was a problem hiding this comment.
lgtm, just to confirm all of these additional pools are autoscaled with default min count 0 right?
Yes, I have tried out AWS stack for now, min counts are from 0. Have seen some issues on pod start up side (insufficient CPU and all) investigating those, will confirm for GCP as well |
sridharkonduri-99
changed the title
- Making exception on gitignore for gpu_instances.tfvars
.gitignore
Outdated
| *.tfvars.json | ||
|
|
||
| # Allow gpu_instances.tfvars files (these contain example GPU configurations, not secrets) | ||
| !**/gpu_instances.tfvars |
There was a problem hiding this comment.
Let's change the gpu_instances.tfvars to gpu_instances.tfvars.example
examples/aws/eks-private/eks.tf
Outdated
| } | ||
| } | ||
| # Additional GPU types can be added via gpu_instances.tfvars | ||
| gpu_types = merge( |
There was a problem hiding this comment.
Let's not do merge(default, additional).
Instead, just specify everything from the variable and this default into variable as well.
| default = ["T4"] | ||
| } | ||
|
|
||
| variable "additional_gpu_types" { |
There was a problem hiding this comment.
Just call it gpu_instance_types.
Add this to the default value:
{
"T4" = {
product_name = "Tesla-T4"
instance_types = ["g4dn.4xlarge"]
}
"A10G" = {
product_name = "NVIDIA-A10G"
instance_types = ["g5.4xlarge"]
}
},
- Remove merge on gpu_instances, use full replace
Pull request checklist
Please check if your PR fulfills the following requirements:
Pull Request Type
https://github.com/anyscale/product/blob/master/backend/workspace-templates.yaml(on_gallery=true)Does this introduce a breaking change?
Terraform is working for both GKE and EKS (Both new clusters public/private)
Screenshots:
GKE Nodegroup (default):
GKE Nodegroup (After additional instances):
EKS Nodegroup (After additional instance):
