Skip to content

v1.1.0

Latest

Choose a tag to compare

@narnat narnat released this 17 Nov 14:35

Adds support for r7i.metal-24xl and r7i.metal-48xl hosts, which have no local disks, by using Amazon FSx for NetApp ONTAP for the VCF management datastore. Also adds VCF 9.1.1, and the Amazon EVS Agent Skill, which enables an AI agent to assist with Amazon EVS questions and use the tooling in this GitHub repository.

EVS Deployment Orchestrator (EDO)

Added

  • r7i.metal-24xl and r7i.metal-48xl hosts, with Amazon FSx for NetApp ONTAP as the management cluster's datastore. Set it in the new, optional storage section of the blueprint: storage.principal.type: "fsxn", with the FSx settings in a storage.principal.fsxn block. Without storage, or with type: "vsan", the orchestrator deploys vSAN as before. Diskless instance types need type: "fsxn", because vSAN needs local NVMe. With FSx for NetApp ONTAP, a simple deployment needs only 2 hosts.
  • The EVS Deployment Orchestrator (EDO) can create the FSx for NetApp ONTAP file system for you. To have EDO create it, leave out fsxn.volume_id and set every field in the fsxn block: file_system_name, storage_capacity_gb, throughput_capacity_mbps, storage_efficiency and s3_access. The file system is Single-AZ, in the VPC the stack creates, and is created before any host is deployed.
  • Use an existing FSx for NetApp ONTAP volume by providing a valid fsxn.volume_id and an existing VPC (ExistingVpcId). The orchestrator checks the volume, its VPC, its Availability Zone, and its security group rules before creating any host. It will never change or delete the existing volume. Creation fields are ignored with a warning if the fsxn.volume_id is present.
  • Once a deployment succeeds, the stack that holds the orchestrator-created file system gets CloudFormation termination protection.
  • destroy.py and the runner's --destroy keep the orchestrator-created file system, which keeps billing. Add --delete-storage to delete it with the datastore and every VM on it. destroy.py --all needs --delete-storage when that file system exists. Teardown deletes the volume without a final backup, so no FSx backup is left billing, and waits up to 2 hours for the file system to delete. Backups you made yourself are never touched.
  • check-quotas.py checks FSx for ONTAP quotas when the orchestrator will create a file system.
  • VCF 9.1.1 (evs.vcf_version: "9.1.1"), with the 9.1.1 SDDC Manager OVA.
  • The README adds sections on monitoring a deployment, troubleshooting, and adding or removing hosts.

Fixed

  • The README's CLI launch uploads the template to your S3 bucket and creates the stack with --template-url. The template is larger than the 51,200 bytes CloudFormation accepts with --template-body, so the documented command failed.

Security

  • urllib3 2.8.0 in the VCF deployment stage, for the HTTPS proxy TLS and chunked-response issues fixed in that release.

Agentic experience

Added

  • Amazon EVS Agent Skill (skills/amazon-evs-agent-skill/) follows the Agent Skills specification. It works with agents that support Agent Skills.
    • Q&A: answers factual Amazon EVS questions from the Amazon EVS documentation and cites the page for each answer. When the documentation doesn't cover a question, it says so instead of answering from model knowledge.
    • EDO deployments: drives an EDO deployment end to end with the user's AWS credentials, in four phases:
      • Prepare: read-only discovery of the account, Region, EVS versions and instance types, VPCs, quotas, artifacts, and the depot secret. It then creates a validated blueprint, launch parameters, and a deployment plan for review.
      • Run: after the user approves the plan, performs the setup and stack launch the plan names.
      • Monitor: follows progress in the orchestrator log, confirms completion, and diagnoses failures with the README's documented recovery steps.
      • Operate: access to the environment, adding or removing hosts, and teardown.
    • Guardrails: the skill asks you to approve each change before it makes it in your AWS account, never reads secret values, and saves its progress in a local, git-ignored folder so a later session can continue.
    • Release pinning: the skill reads EDO's README at the release it deploys, so its commands and prerequisites match that release.
    • SKILL.md lists the read-only AWS CLI commands the skill runs in allowed-tools, so agents that honor the field can pre-approve them.
    • AGENTS.md routes Amazon EVS requests in this repository to the skill.
    • Requirements: Q&A needs a web fetch tool and access to docs.aws.amazon.com. Deployments need AWS CLI v2 or the AWS MCP server, git, and Python 3.10 or later with boto3 and pyyaml.
    • Install: copy skills/amazon-evs-agent-skill/ to your agent's skills folder. skills/amazon-evs-agent-skill/README.md has the details.

Full history: CHANGELOG.md