blurringtool-ml-deployment/README.md

BlurringTool - Infra

Description

Infra-Deployment

CUDA Server deployment, specializing for the blurit-op software. While with Terraform the general infrastructure is deployed, information is gathered for use in configuration deployment afterward.

The infrastructure contains serveral linux hosts with GPUs for ML, STACKIT ObjectStorage Buckets for data transfer.

Configuration-Deployment

After the infrastrecture is set, files were written, that help in the following process of configurate the hosts:

• hosts.ini
• fstab - entry for s3fs mount
• passwd-s3fs

Final config

Access the host(s)

when the deployment is complete, the user set in the cloud.yml

users:
- name: blurringtool
  gecos: Blurring Tool User
  primary_group: blurringtool
  groups: ['sudo', 'users']
  shell: /bin/bash
  sudo: ALL=(ALL) NOPASSWD:ALL
  ssh_authorized_keys:
  - ssh-rsa SsH-RsA-KeY COMMENT
  - ssh-ed25519 sSh-Ed25519 COMMENT

can access the host if his client is sitting in the IP-range set variables.tf as default.

variable "ssh_source_ips" {
  type    = list(string)
  default = ["185.124.195.0/24"]
}

Install blurit-op

To install blurit-op follow the instructions in the docs: https://doc-op.blurit.io/blurit-on-premise/installation/run-blurit-op

Environment Variables

• in exmaple_env

USAGE local deployment

• git clone the branch/repo
• initialize the key flow https://github.com/stackitcloud/terraform-provider-stackit#key-flow
• set the credetials path in env (rename example_env)
• set all other necessary vars in env-file
• source the env-file
• comment out all unnecessary providers and backends in main.tf

DISCLAIMER

this is by no means an end-product, so if you find something to improve:

1. fix it and push the change
2. if this is not feasable, point it out in an issue