Adds first-class support for creating and deploying Braintrust data planes on module-managed EKS clusters.#245
Adds first-class support for creating and deploying Braintrust data planes on module-managed EKS clusters.#245JPurcell-Braintrust wants to merge 5 commits into
Conversation
… and public CIDR expectations
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7649ef9801
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "Codex (@codex) review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "Codex (@codex) address that feedback".
| variable "brainstore_reader_node_group_instance_type" { | ||
| type = string | ||
| description = "Instance type for the brainstore reader node group. Must be an instance with local NVMe storage. c8gd (Graviton4) instances are the recommended default." | ||
| default = "c8gd.8xlarge" |
There was a problem hiding this comment.
The new EKS Brainstore node-group inputs only describe the NVMe requirement (brainstore_reader_node_group_instance_type/writer) but never validate it, so users can pass non-NVMe types and still get a successful plan. In this module that leads to Brainstore nodes being provisioned without the required local instance storage, which breaks the safety invariant already enforced in the EC2 Brainstore path and can cause Brainstore scheduling/runtime failures after apply. Add plan-time validation (for both on-demand and spot inputs) against EC2 instance storage support.
Useful? React with 👍 / 👎.
|
Moving to draft for now. Lets wait until we fully move to ECS before we merge this in. Right now our module has a lot of complexity as we juggle lambda and ECS. |
2 participants
What changed
create_eks_clustermodules/eks-deployflow for Kubernetes bootstrap resources and Braintrust Helm deploymentexamples/eks-auto-modefor a one-click Auto Mode deploymentexamples/eks-terraform-managedfor customer-managed Helm/ingress workflowsNotes
The Auto Mode example is intended as the full one-click path: Terraform creates the AWS infrastructure, EKS cluster, Kubernetes bootstrap resources, and Braintrust Helm release.
The Terraform-managed example is intended as the handoff path: Terraform creates the AWS data plane and EKS cluster/node groups, then customers install the Braintrust chart through the official helm chart and choose their own ingress and domain approach.