Gitpod Self-Hosted Instructions

Since Gitpod enabled GitHub Actions by default it appears that the container images are no longer published to a publicly accessible registry.

This means that the dream of maintaining Gitpod self-hosted is over. I still really hope that this decision is reversed, but I can't see it ever happening now. If you do want to maintain a self-hosted development environment, there are some strong projects in this space - in alphabetical order, some to investigate are:

Coder

Hocus

Thanks to everyone who helped with this project, especially Tom Barber. This is a very sad day indeed.

Simon Emms, April 2023

Resources for managing your own Gitpod installation

About
Infrastructure
Cluster Sizes
Supported Cloud Providers
- Provider Interfaces
  - Input Variables
  - Output Variables
Deploying Your Own Gitpod Instance
Versioning
Contributing
Roadmap
Support
Referral Links

This is a community-maintained project containing instructions on how to run Gitpod in your own infrastructure. This is original work, not a fork of Gitpod and is not affialiated to Gitpod in any way. Gitpod is licenced under AGPL-3.0 and all Gitpod components are used under that licence.

About

Railway trains are impartial too. But if you lay down the lines for them, that's the way they go.

Bernard Woolley, Yes Minister

From August 2021, I provided the technical leadership of the Gitpod Self-Hosted team until my redundancy in January 2023. Even though Self-Hosted is no longer part of Gitpod's business strategy, I've done too much good work on the project that I wasn't prepared to just let it die. Also, I still maintain my own Gitpod instance that I enjoy hacking on.

The general aim of this project is to be railway tracks. The major issue that Gitpod had when we supported self-hosting was that we aimed to support every cloud provider's managed Kubernetes service which resulted in limiting what features Gitpod could have. This project will have an intentionally limited number of things that you can configure so that your installation works reliably.

In fact, by design this project will only allow you to configure five things:

The cloud you run it on
The domain name it's run on
The location it runs in
The size of the installation - small (for PoCs), medium and large (see Cluster Sizes)
The SSH key used to access the virtual machines

A more detailed explanation and reasoning on why I'm still passionate about Gitpod Self-Hosted is on my blog.

Infrastructure

This builds an infrastructure similar to the one that is used for Gitpod.io, namely a k3s cluster running on Ubuntu 20.04 with all the required dependencies and packages. If the cloud provider supports it, this will also use container registry, database and object storage to connect to your Gitpod instance, or the in-cluster dependencies if not.

This is not designed to run on the managed Kubernetes service of a cloud provider. That proved too difficult to do so reliably. Gitpod.io abandoned using GKE in mid-2022 because it was too difficult to get the full feature-set and our team made the mistake of supporting all the managed Kubernetes providers. This project corrects that mistake by using k3s-only.

IMPORTANT this is NOT a good project for a Kubernetes beginner. Whilst I laud any effort for self-improvement, Gitpod is not the project to do it on. I won't be providing any general Kubernetes support in this repository.

Cluster Sizes

There is a standard approach provided for the size of clusters. As this project is designed to be railways tracks for installing Gitpod, the cluster size is the main method of configuration. Any user-defined data is configured in the size_data variable - there is limited validation on this parameter and will be ignored if an unsupported size is usedl

Name	Purpose	Manager Highly-Available	Manager Nodes	Worker Nodes	Worker Clusters
`small`	PoCs and personal developer environments	No	1 x 4vCPU, 16GB RAM	1 x 4vCPU, 16GB RAM	1
`medium`	Small development teams	Yes	3 x smallest usable machine	3-100 x 4vCPU, 16GB RAM	1
`large`	Large development teams	Yes	3 x user-defined	Auto-scaled, user-defined	User-defined

The large cluster may not be supported in all cloud providers. This is dependent upon the cloud provider having some form of auto-scaling ability for virtual machines.

In future, there may be support for extra-large clusters with support for a multi-region installation for international development teams.

Supported Cloud Providers

Provider	Supported Sizes	Notes
Hetzner	small, medium
Microsoft Azure	small	Requires 2 x 4CPU, 16GB VMs - this is above the free trial quota so will need a paid Azure account Azure Storage is not S3-compatible, so does not work with Gitpod storage Further work will require access to a paid Azure account
Scaleway	small

Provider Interfaces

The interface of each provider will differ slightly between each implementation. However, there is the attempt made to provide a certain amount of consistency between each one. Whilst it's unlikely that many users will have the desire to switch between providers, consistency is good practice and allows for better testability.

Input Variables

Name	Description	Type	Default	Required
domain_name	The domain name Gitpod will be accessed on	`string`	n/a	yes
kubecontext	Kubecontext name to use	`string`	`"gitpod-self-hosted"`	no
location	Data centre location	`string`	<provider-specific>	no
size	Deployment size such as `small`, `medium` or `large` - not all of these sizes are guaranteed to exist	`string`	`"small"`	no
size_data	Additional data for deployment size - this will be ignored if using an invalid `size`	`object`	<provider-specific>	no
ssh_private_key_path	Path to private key, used for logging in to VM - passphrases are not supported	`string`	`"~/.ssh/id_rsa"`	no
ssh_public_key_path	Path to public key, used for logging in to VM - passphrases are not supported	`string`	`"~/.ssh/id_rsa.pub"`	no

Output Variables

Name	Type	Description
dns_records	list(object({ type = string name = string subdomain = string value = string }))	List of DNS records generated by the provider
k3s_token	`string`	Join token for k3s
kubeconfig	`string`	Kubernetes config YAML file
kubecontext	`string`	Kubecontext name to use

Deploying Your Own Gitpod Instance

You can check out my Gitpod App repo, which is the configuration for my own Gitpod installation.

Create Your Terraform

This project expects you to have an understanding of how to use Terraform.

Create a new repository
Configure your desired remote state
Import the Gitpod Module for your desired cloud provider. This is in the format github.com/mrsimonemms/gitpod-self-hosted/infrastructure/providers/<provider>
Add any additional Terraform resources you want. This will typically be your DNS configuration, but may include container registry, database and object storage if you don't wish to use the default configuration (or if your desired provide doesn't have these services)
Configure your cert_manager, domain_name, gitpod_config, gitpod_secrets and kubeconfig outputs.

Next Steps

Once your Terraform resources have been applied, you will need to configure your k3s instance and install Gitpod.

Container Storage Interface (optional)

This is an optional step and only recommended if you are going to use the in-cluster database or object storage services. Gitpod creates no other persistent volume claims.

In-cluster services are provided as a convenience. These can be difficult to scale and backup, so should only be used if there is no alternative.

A Container Storage Interface (CSI) driver is used to configure how the Kubernetes cluster saves persistent data. k3s uses local-path storage by default, which saves the data to the node. That's fine, but is not necessarily performant and, if the node is evicted, the data is lost.

Where possible, a CSI driver is given to save data to the cloud's volume implementation. They may not exist for every provider, especially if the provider maintains a managed database and storage option.

Here's how to install the Hetzner CSI. Notice the HCLOUD_TOKEN, which is required to manage volumes.

curl -sfSL https://raw.githubusercontent.com/mrsimonemms/gitpod-self-hosted/main/csi.sh | \
  CMD=hetzner \
  HCLOUD_TOKEN="${HCLOUD_TOKEN}" \
  bash -

A Word On The Container Registry

The container registry is not a critical part of the system. It stores no data aside from images generated for a workspace. If the workspace no longer exists, it will be rebuilt.

There is no reason to put effort into backing up the container registry because the resources will be rebuilt to the desired state. The reason the container registry exists is to avoid having to build workspace images needlessly.

Cert-Manager

This installs cert-manager with Helm, creates any secrets and configures the ClusterIssuer you've defined in the Terraform output.

NB - the SECRETS and CLUSTER_ISSUER arguments must be encoded as Base64.

curl -sfSL https://raw.githubusercontent.com/mrsimonemms/gitpod-self-hosted/main/install.sh | \
  CMD=cert_manager \
  DOMAIN_NAME="$(terraform output -raw domain_name)" \
  SECRETS="$(terraform output -json cert_manager | jq -cr '.secrets | @base64')" \
  CLUSTER_ISSUER="$(terraform output -json cert_manager | jq -cr '.cluster_issuer | @base64')" \
  bash -

DNS01 Webhooks

If you want to use the DNS01 webhook provider to generate your TLS certs, you will need to define a script to install any additional dependencies required. The script needs to be defined with the environment variable WEBHOOKS_SCRIPT. This can be any valid command that your terminal supports - for readability, it is recommended that you execute a file.

This script is run immediately after cert-manager is installed/upgraded.

Here's a worked example using the Scaleway webhook. The script would be saved to a file called ./install-scaleway-webhook.sh and WEBHOOKS_SCRIPT would be set to bash ./install-scaleway-webhook.sh.

#!/usr/bin/env bash

set -eo pipefail

git clone https://github.com/scaleway/cert-manager-webhook-scaleway.git ~/.cert-manager-webhook-scaleway || true

helm upgrade \
  --atomic \
  --cleanup-on-fail \
  --create-namespace \
  --install \
  --namespace="cert-manager" \
  --reset-values \
  --set secret.accessKey="$ACCESS_KEY" \
  --set secret.secretKey="$SECRET_KEY" \
  --wait \
  scaleway-webhook \
  ~/.cert-manager-webhook-scaleway/deploy/scaleway-webhook

This example sets the secret values. If you prefer, these can still be set in the secrets parameter in the cert_manager output from your Terraform scripts.

Install Gitpod

Installs or upgrades Gitpod.

NB - the GITPOD_CONFIG and GITPOD_SECRETS arguments must be encoded as Base64.

curl -sfSL https://raw.githubusercontent.com/mrsimonemms/gitpod-self-hosted/main/install.sh | \
  CMD=install_gitpod \
  GITPOD_CONFIG="$(terraform output -json gitpod_config | jq -cr '@base64')" \
  GITPOD_SECRETS="$(terraform output -json gitpod_secrets | jq -cr '@base64')" \
  bash -

Monitoring

This will also install the MrSimonEmms/gitpod-monitoring Helm chart to the monitoring namespace. This is designed to provide some observability on your Gitpod instance. If you do not wish to install this chart, please set the environment variable MONITORING_INSTALL to false.

Outputs

`cert_manager`

Gitpod requires cert-manager as there are various internal certificates generated. The install.sh script also configures cert-manager to be your TLS issuer for your instance's public certificates. Whilst Gitpod works fine with other issuers, this project will only support cert-manager as the issuer. As Gitpod requires a wildcard certificate, you must configure a DNS01 ClusterIssuer.

The cert_manager output must have two keys - secrets and cluster_issuer. The secrets are the secrets required for the ClusterIssuer to work. Each key is the secret name and the value is key/value pairs of the secret name and secret value.

These secrets will be created in the cert-manager namespace.

The cluster_issuer is the Kubernetes manifest's spec. Check the cert-manager documentation for your desired provider. Typically, this will be configuring the DNS01 challenge provider to use the secret detailed in secrets.

This is an example for Cloudflare configuration.

output "cert_manager" {
  sensitive   = true
  value = {
    secrets = {
      cloudflare-solver = {
        apiToken = var.cloudflare_api_token
      }
    }
    cluster_issuer = {
      spec = {
        acme = {
          solvers = [
            {
              dns01 = {
                cloudflare = {
                  apiTokenSecretRef = {
                    name = "cloudflare-solver" # This is the secret referenced above
                    key  = "apiToken"
                  }
                }
              }
            }
          ]
        }
      }
    }
  }
}

`domain_name`

Output the domain_name that the instance lives behind. This is in the format gitpod.domain.com.

output "domain_name" {
  value       = var.domain_name
  sensitive   = true
}

`gitpod_config`

Output the Gitpod Installer configuration values in JSON format. This will be used to override the default values.

At a minimum, this should set the domain, but can set other things. See the Gitpod documentation for more details.

output "gitpod_config" {
  value = {
    blockNewUsers = {
      enabled  = length(var.domain_passlist) > 0
      passlist = var.domain_passlist
    }
    domain = var.domain_name
  }
  sensitive = true
}

`gitpod_secrets`

These will be secrets that are required by the gitpod_config. Typically, these will be credentials for your database, registry, storage and the like.

These secrets will be created in the gitpod namespace.

Like the cert_manager output, these will have the secret name as the top-level key and then any secrets as key/value pairs.

This example is how one might define the database secret for an Azure MySQL database.

output "gitpod_secrets" {
  value = {
    database = {
      # If unsure, use this key
      encryptionKeys = "[{\"name\":\"general\",\"version\":1,\"primary\":true,\"material\":\"4uGh1q8y2DYryJwrVMHs0kWXJlqvHWWt/KJuNi04edI=\"}]"
      host           = "${azurerm_mysql_server.db.name}.mysql.database.azure.com"
      password       = azurerm_mysql_server.db.administrator_login_password
      port           = 3306
      username       = "${azurerm_mysql_server.db.administrator_login}@${azurerm_mysql_server.db.name}"
    }
  }
  sensitive = true
}

`kubeconfig`

Pass the kubeconfig value from the provider module.

output "kubeconfig" {
  value       = module.hetzner.kubeconfig
  sensitive   = true
}

Example Gitpod Module - Hetzner

module "hetzner" {
  source = "github.com/mrsimonemms/gitpod-self-hosted/infrastructure/providers/hetzner"

  domain_name          = var.domain_name
  location             = var.location
  size                 = var.size
  ssh_private_key_path = var.ssh_private_key_path
  ssh_public_key_path  = var.ssh_public_key_path
}

Versioning

Gitpod currently uses Werft as their CI/CD system and publishes images to werft.gitpod-dev.com. When I left Gitpod, there was an RFC about switching from Werft to GitHub Actions. Watch this space for more information.

This project maintains two tagged versions of the Installer:

nightly: a daily version, generated from the most recent version at midnight UTC. This is also tagged with a date tag, in format YYYY-MM-DD.
latest: a monthly supported version. This is also tagged with a version tag, in a semver-like format YYYY.MM.<patch>. It is recommended you use this.

To define your version, set the GITPOD_INSTALLER_VERSION environment variable. If unset, it will be latest.

Contributing

Please see the contributing guide for more details.

Roadmap

The roadmap will largely depend upon the community response to this repository. Ideally, I would like to see support for:

Amazon Web Services
Civo Cloud
DigitalOcean
Google Cloud Platform
Scaleway

Work on these platforms will cost money to get resources spun up. If you want me to work on these for your own Gitpod installation, I will only be able to do so if you can provide me an account to cover the cost of creating these resources for testing purposes.

I'm happy to receive pull requests for these platforms from contributors. As above, I will need to be able to test these before they're merged to main - pull requests will only be accepted if they come with an account for the platform in question so I can verify the work.

Please contact me to discuss further.

Support

I'm happy to discuss options for commercial support contracts. Please contact me to discuss further.

Referral Links

Using these links will help support the project and provide you with additional credits for the cloud provider.

Hetzner