Migration Docs for Kustomize

The old method of deploying Sourcegraph with custom scripts has been deprecated. Instead, the new setup uses Kustomize, a Kubernetes-native tool, for configurations. This guide explains how to migrate from the old setup (deploy-sourcegraph) to the new one (deploy-sourcegraph-k8s).

NOTE: Both the old custom scripts and Kustomize only create manifests for deployment and don’t change any existing resources in an active cluster.

Why migrate?

Here are the benefits of the new base cluster with the new Kustomize setup compared to the old one:

• Improved security defaults:
  • Runs in non-privileged mode
  • Uses non-root users
  • Does not require RBAC resources
• Streamlined resource allocation process:
  • Allocates resources based on the size of the instance
  • Optimized through load testing
  • The searcher and symbols use StatefulSets and do not require ephemeral storage
• Utilizes the Kubernetes-native tool Kustomize:
  • Built into kubectl
  • No additional scripting required
  • More extensible and composable
  • Highly reusable that enables creation of multiple instances with the same base resources and components
• Effortless configurations:
  • A comprehensive list of components pre-configured for different use cases
  • Designed to work seamlessly with Sourcegraph’s design and functionality
  • Prevents merge conflicts during upgrades

Migration process

The migration process for transitioning from the Kustomize setup in deploy-sourcegraph to deploy-sourcegraph-k8s involves the steps shown below.

The goal of this migration process is to create a new overlay that will generate similiar resources as the current cluster, ensuring a smooth deployment process without disrupting existing resources.

Step 1: Upgrade current instance with the old repository

Upgrade your current instance to the latest version of Sourcegraph (must be 4.5.0 or above) following the standard upgrade process for the repository (deploy-sourcegraph) your instance was deployed with.

From Privileged to Non-privileged

Sourcegraph's deployment mode changed from privileged (containers run as root) to non-privileged (containers run as non-root) as the default in the new Kustomize setup. If your instance is currently running in privileged mode and you want to upgrade to non-privileged mode, use the migrate-to-nonprivileged overlay from the Sourcegraph deploy-sourcegraph repository when following the standard upgrade process to perform your upgrade.

NOTE: Applying the migrate-to-nonprivileged overlay will convert your deployment to run in non-privileged mode

Step 2: Set up a release branch for the new repository

Set up a release branch from the latest version branch in your local fork of the deploy-sourcegraph-k8s repository.

BASH
  # Recommended: replace the URL with your private fork
  $ git clone https://github.com/sourcegraph/deploy-sourcegraph-k8s.git
  $ cd deploy-sourcegraph-k8s
  $ git checkout v4.5.1 && git checkout -b release

Step 3: Set up a directory for your instance

Create a copy of the instances/template directory and rename it to instances/my-sourcegraph:

BASH
  $ cp -R instances/template instances/my-sourcegraph

NOTE: In Kustomize, this directory is referred to as an overlay.

Step 4: Set up the configuration files

kustomization.yaml

The kustomization.yaml file is used to configure your Sourcegraph instance.

1. Rename the kustomization.template.yaml file in instances/my-sourcegraph to kustomization.yaml.

• The kustomization.yaml file is used to configure your Sourcegraph instance.

BASH
  $ mv instances/my-sourcegraph/kustomization.template.yaml instances/my-sourcegraph/kustomization.yaml

buildConfig.yaml

2. Rename the buildConfig.template.yaml file in instances/my-sourcegraph to buildConfig.yaml.

• The buildConfig.yaml file is used to configure components included in your kustomization file when required.

BASH
  $ mv instances/my-sourcegraph/buildConfig.template.yaml instances/my-sourcegraph/buildConfig.yaml

Step 5: Set namespace

Replace ns-sourcegraph with a namespace that matches the existing namespace for your current instance.

You may set namespace: default to deploy to the default namespace.

YAML
# instances/my-sourcegraph/kustomization.yaml
  namespace: ns-sourcegraph

Step 6: Set storage class

To add the storage class name that your current instance is using for all associated resources:

6.1. Include the storage-class/name-update component under the components list.

YAML
# instances/my-sourcegraph/kustomization.yaml
  components:
    # This updates storageClassName to
    # the STORAGECLASS_NAME value from buildConfig.yaml
    - ../../components/storage-class/name-update

6.2. Input the storage class name by setting the value of STORAGECLASS_NAME in buildConfig.yaml.

For example, set STORAGECLASS_NAME=sourcegraph if sourcegraph is the name of an existing storage class:

YAML
# instances/my-sourcegraph/buildConfig.yaml
  kind: ConfigMap
  metadata:
    name: sourcegraph-kustomize-build-config
  data:
    STORAGECLASS_NAME: sourcegraph # -- [ACTION] Update storage class name here

Step 7: Recreate overlay (OPTIONAL)

NOTE: You may skip this step if your instance was not deployed using overlays from the deploy-sourcegraph repository.

If your running instance was deployed using an existing overlay from the deploy-sourcegraph repository (except migrate-to-nonprivileged):

1. Copy and paste everything (except kustomization.yaml) from the old overlay directory to instances/my-sourcegraph.
2. Manually merge the contexts from the old kustomization.yaml file into instances/my-sourcegraph/kustomization.yaml.
3. In the new kustomization.yaml file, replace the old base resources (bases/deployments, bases/pvcs) with the new base resources that include:
   • buildConfig.yaml
   • base/sourcegraph
   • base/monitoring

For example:

DIFF
# instances/my-sourcegraph/kustomization.yaml
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
namespace: default
resources:
-  - ../bases/deployments
-  - ../bases/pvcs
+  - buildConfig.yaml
+  - ../../base/sourcegraph
+  - ../../base/monitoring

Step 8: Recreate instance resources

Follow our configuration guide to recreate your running instance with the provided components.

It is recommended to refrain from introducing any changes to the characteristics of a running Kubernetes cluster during a migration. For example, if the cluster is currently running in privileged mode with root user access, deploying the instance in non-privileized mode could cause permission errors.

Ensure the following configurations are present/ consistent for your Sourcegraph instance during migrations:

• Storage size for all services
• Permission settings (privileged or non-privileged mode)
• Networking
• Ingress
• Storage class
• Storage class name
• Namespace
• cAdvisor
• Tracing services (OpenTelemetry and Jaeger for example)

If you have previously made changes directly to the files inside the base directory, please convert these changes into patches before adding them to your kustomization.yaml file as patches.

Privileged

If your Sourcegraph instance is currently running in privileged mode, start building your overlay with the clusters/old-base component, which generates resources similar to the base cluster in deploy-sourcegraph.

YAML
# instances/my-sourcegraph/kustomization.yaml
  components:
    - ../../components/clusters/old-base

Non-privileged

The default cluster now runs in non-privileged mode.

If your instance was deployed using the non-privileged overlay, you can follow the configuration guide without adding the clusters/old-base component.

Step 9: Build and review new manifests

pgsql, codeinsights-db, searcher, symbols, and codeintel-db have been changed from Deployments to StatefulSets. However, redeploying these services as StatefulSets should not affect your existing deployment as they are all configured to use the same PVCs.

From Deployment to StatefulSet

searcher and symbols are now StatefulSet that run as headless services. If your current searcher and symbols are running as Deployment, you will need to remove their services before re-deploying them as StatefulSet:

BASH
  $ kubectl delete service/searcher
  $ kubectl delete service/symbols

1. Generate new manifests with the overlay:

BASH
  $ kubectl kustomize my-sourcegraph -o cluster.yaml

2. Review the changes to ensure that the manifests generated by your new overlay are similar to the ones currently being used by your active cluster.

Compare the manifests generated by your new overlay with the ones in your running cluster using the command below:

BASH
  $ kubectl diff -l deploy=sourcegraph -f cluster.yaml

Step 10: Deploy new manifests

Once you are satisfied with the overlay output, you can now deploy the new overlay using these commands:

BASH
  # Build manifests again with overlay
  $ kubectl kustomize $PATH_TO_OVERLAY -o cluster.yaml
  # Apply manifests to cluster
  $ kubectl apply --prune -l deploy=sourcegraph -f cluster.yaml

WARNING: Make sure to test the new overlay and the migration process in a non-production environment before applying it to your production cluster.