Skip to content

Latest commit

 

History

History
231 lines (162 loc) · 13.4 KB

README.md

File metadata and controls

231 lines (162 loc) · 13.4 KB
Raw

Deploy Ollama and Open WebUI with GitHub Actions 🌐🚀

Welcome to the easiest way to get your own hosted private language model running swiftly with Ollama and Open WebUI

alt text

Why use this action?

With this action, you can easily have your very own Large Language Model (LLM) like OpenAI's GPTChat or Anthropic's Claude. Except that it's entirely yours! You can tune it with your own data, and it's hosted on your own AWS account.

This action is perfect for anyone who wants to try out the latest models, ask questions about documents, or even build Retrieval Augmented Generation (RAG) apps against an LLM you own!

🖥️✨ Get started below, and you'll be talking with your own hosted LLM in no time!

Supported Cloud Providers:

  • AWS

Note: This action is currently in beta. Please report any issues you find by creating an Issue or a Pull Requests

alt

Need help or have questions?

This project is supported by Bitovi DevOps and a proud supporter of Open Source software.

You can get help or ask questions on our Discord channel! Come hang out with us; We love discussing solutions!

Or, you can hire us for training, consulting, development, and deployments (including LLM deployments ;). Set up a free consultation.

Requirements

This is a list of requirements you'll need to meet in order to use this action.

  1. An AWS account (yep, that's it!)

1. An AWS account

You'll need Access Keys from an AWS account

Example usage

Create .github/workflow/deploy.yaml with the following to build on push.

Basic example

name: Basic deploy
on:
  push:
    branches: [ main ]

jobs:
  EC2-Deploy:
    runs-on: ubuntu-latest
    steps:
    - id: deploy
      name: Deploy
      uses: bitovi/[email protected]
      with:
        aws_access_key_id: ${{ secrets.AWS_ACCESS_KEY_ID}}
        aws_secret_access_key: ${{ secrets.AWS_SECRET_ACCESS_KEY}}

Once deployed, visit the url provided in the workflow output summary.

Advanced example

(todo)

Customizing

Inputs

  1. Action Defaults
  2. AWS Configuration
  3. Secrets and Environment Variables
  4. EC2
  5. Prometheus
  6. Stack Management
  7. Domains
  8. VPC

The following inputs can be used as step.with keys

Action defaults Inputs

Name Type Description Required Default
checkout Boolean Set to false if the code is already checked out. false true

AWS Configuration Inputs

Name Type Description Required Default
aws_access_key_id String AWS access key ID. true
aws_secret_access_key String AWS secret access key. true
aws_session_token String AWS session token, if you're using temporary credentials. false
aws_default_region String AWS default region. true us-east-1
aws_resource_identifier String Auto-generated by default so it's unique for org/repo/branch. Set to override with custom naming the unique AWS resource identifier for the deployment. Defaults to ${org}-${repo}-${branch}. false ${GITHUB_ORG_NAME}-${GITHUB_REPO_NAME}-${GITHUB_BRANCH_NAME}
aws_extra_tags JSON A list of additional tags that will be included on created resources. Example: {"key1": "value1", "key2": "value2"} false {}

Secrets and Environment Variables Inputs

Name Type Description - Check note about environment variables.
env_aws_secret String Secret name to pull env variables from AWS Secret Manager, could be a comma separated list, read in order. Expected JSON content.
env_repo String File containing environment variables to be used with the app.
env_ghs String .env file to be used with the app from Github secrets.
env_ghv String .env file to be used with the app from Github variables.

EC2 Inputs

Name Type Description
aws_ec2_instance_type String The AWS EC2 instance type. Default is inf1.xlarge.
aws_ec2_instance_profile String The AWS IAM instance profile to use for the EC2 instance. Use if you want to pass an AWS role with specific permissions granted to the instance.
aws_ec2_create_keypair_sm Boolean Creates a Secret in AWS secret manager to store a kypair. Default is false.
aws_ec2_instance_vol_size String Root disk size for the EC2 instance. Default is 10.
aws_ec2_additional_tags JSON A JSON object of additional tags that will be included on created resources. Example: {"key1": "value1", "key2": "value2"}
aws_ec2_ami_filter String AMI filter to use when searching for an AMI to use for the EC2 instance. Defaults to ubuntu/images/hvm-ssd/ubuntu-focal-20.04-amd64-server-*.
infrastructure_only Boolean Set to true to provision infrastructure (with Terraform) but skip the app deployment (with ansible). Default is false.

Stack Management

Name Type Description
tf_stack_destroy Boolean Set to true to destroy the created AWS infrastructure for this instance. Default is false.
tf_state_file_name String Change this to be anything you want to. Carefull to be consistent here. A missing file could trigger recreation, or stepping over destruction of non-defined objects.
tf_state_file_name_append String Append a string to the tf-state-file. Setting this to unique will generate tf-state-aws-unique. Can co-exist with the tf_state_file_name variable.
tf_state_bucket String AWS S3 bucket to use for Terraform state. Defaults to ${org}-${repo}-{branch}-tf-state-aws.
tf_state_bucket_destroy Boolean Force purge and deletion of S3 tf_state_bucket defined. Any file contained there will be destroyed. tf_stack_destroy must also be true.

Domains

Name Type Description
aws_domain_name String Define the root domain name for the application. e.g. bitovi.com. If empty, ELB URL will be provided.
aws_sub_domain String Define the sub-domain part of the URL. Defaults to ${org}-${repo}-{branch}.
aws_root_domain Boolean Deploy application to root domain. Will create root and www DNS records. Domain must exist in Route53.
aws_cert_arn String Existing certificate ARN to be used in the ELB. Use if you manage a certificate outside of this action. See https://docs.aws.amazon.com/acm/latest/userguide/gs-acm-list.html for how to find the certificate ARN.
aws_create_root_cert Boolean Generates and manage the root certificate for the application to be used in the ELB.
aws_create_sub_cert Boolean Generates and manage the sub-domain certificate for the application to be used in the ELB.
aws_no_cert Boolean Set this to true if you want not to use a certificate in the ELB. Default is false.

VPC Inputs

Name Type Description
aws_vpc_create Boolean Define if a VPC should be created. Default is true.
aws_vpc_name String Set a specific name for the VPC.
aws_vpc_cidr_block String Define Base CIDR block which is divided into subnet CIDR blocks. Defaults to 10.0.0.0/16.
aws_vpc_public_subnets String Comma separated list of public subnets. Defaults to 10.10.110.0/24.
aws_vpc_private_subnets String Comma separated list of private subnets. If none, none will be created.
aws_vpc_availability_zones String Comma separated list of availability zones. Defaults to aws_default_region.
aws_vpc_id String AWS VPC ID. Accepts vpc-### values.
aws_vpc_subnet_id String Specify a Subnet to be used with the instance. If none provided, one will be picked.
aws_vpc_additional_tags JSON A JSON object of additional tags that will be included on created resources. Example: {"key1": "value1", "key2": "value2"}

Environment variables

For envirnoment variables in your app, you can provide:

  • env_repo - A file in your repo that contains env vars
  • env_ghv - An entry in Github actions variables
  • env_ghs - An entry in Github secrets
  • env_aws_secret - The path to a JSON format secret in AWS

These environment variables are merged (in the following order) to the .env file and provided to both the Prometheus and Grafana services:

  • Terraform passed env vars ( This is not optional nor customizable )
  • Repository checked-in env vars - repo_env file as default. (KEY=VALUE style)
  • Github Secret - Create a secret named DOT_ENV - (KEY=VALUE style)
  • AWS Secret - JSON style like '{"key":"value"}'

Note about resource identifiers

Most resources will contain the tag ${GITHUB_ORG_NAME}-${GITHUB_REPO_NAME}-${GITHUB_BRANCH_NAME}, some of them, even the resource name after. We limit this to a 60 characters string because some AWS resources have a length limit and short it if needed.

We use the kubernetes style for this. For example, kubernetes -> k(# of characters)s -> k8s. And so you might see some compressions are made.

For some specific resources, we have a 32 characters limit. If the identifier length exceeds this number after compression, we remove the middle part and replace it for a hash made up from the string itself.

S3 buckets naming

Buckets names can be made of up to 63 characters. If the length allows us to add -tf-state, we will do so. If not, a simple -tf will be added.

CERTIFICATES - Only for AWS Managed domains with Route53

As a default, the application will be deployed and the ELB public URL will be displayed.

If domain_name is defined, we will look up for a certificate with the name of that domain (eg. example.com). We expect that certificate to contain both example.com and *.example.com.

If you wish to set up domain_name and disable the certificate lookup, set up no_cert to true.

Setting create_root_cert to true will create this certificate with both example.com and *.example.com for you, and validate them. (DNS validation).

Setting create_sub_cert to true will create a certificate just for the subdomain, and validate it.

⚠️ Be very careful here! Created certificates are fully managed by Terraform. Therefor they will be destroyed upon stack destruction.

To change a certificate (root_cert, sub_cert, ARN or pre-existing root cert), you must first set the no_cert flag to true, run the action, then set the no_cert flag to false, add the desired settings and excecute the action again. (This will destroy the first certificate.)

This is necessary due to a limitation that prevents certificates from being changed while in use by certain resources.

Made with BitOps

BitOps allows you to define Infrastructure-as-Code for multiple tools in a central place. This action uses a BitOps Operations Repository to set up the necessary Terraform and Ansible to create infrastructure and deploy to it.

Contributing

We would love for you to contribute to bitovi/github-actions-deploy-ollama and help make it even better than it is today!

Would you like to see additional features? Create an issue or a Pull Requests.

License

The scripts and documentation in this project are released under the MIT License.