How to Migrate Sourcegraph from one cluster to another.

The instructions in this document will guide you on the steps you need to take to migrate your Sourcegraph deployment from one Cluster to another. This is a process that involves the backing up of the Sourcegraph database by generating dumps and restoring the dumps in the newly deployed Sourcegraph instance on another cluster.


  1. Backup Sourcegraph databases in the old cluster by generating dumps
  2. Restore the Dumps in the freshly deployed instance on another cluster.

Steps to backup Sourcegraph Databases

In this section of the tutorial, you would learn how to back up the primary sourcegraph database and the codeintel database.

  • Verify that the deployment is running.

    kubectl get pods -A

  • Stop all connections to the database by removing the frontend deployment.

    kubectl scale --replicas=0 deployment/sourcegraph-frontend

  • Generate the database dumps.

    • Dumps for the primary sourcegraph DB kubectl exec -it $pgsql_POD_NAME -- bash -c 'pg_dump -C --username sg sg' > sourcegraph_db.out

    • Dumps for the codeintel DB kubectl exec -it $codeintel-db_POD_NAME -- bash -c 'pg_dump -C --username sg sg' > codeintel_db.out

  • Ensure you copy/move the sourcegraph_db.out and codeintel_db.out files to the root of the deploy-sourcegraph directory of the new Sourcegraph deployment.

    These files are sensitive and should be stored/transported securely.

Steps to restore Sourcegraph Databases into a new environment

In this section of the tutorial, you would learn how to restore the primary sourcegraph database and the codeintel database from the generated dumps in the previous section.

  • Before restoring on the new deployment, It is very important you stop all connections to the Database from the frontend.

    • Scale down all the pods in the deployment
      kubectl scale deployment --all --replicas=0
    • Scale down all the statefulset
      kubectl scale sts --all --replicas=0
  • Then restart ONLY the pgsql and codeintel pods

    • Start the pgsql:
      kubectl scale deployment/pgsql --replicas=1
    • Start the codeintel-db:
      kubectl scale deployment/codeintel-db --replicas=1
  • Copy the database files into the pods by running the following command from the root of the deploy-sourcegraph directory

    kubectl cp sourcegraph_db.out $pgsql_POD_NAME:/tmp/sourcegraph_db.out

    kubectl cp codeintel_db.out $codeintel-db_POD_NAME:/tmp/codeintel_db.out

  • Exec into the pgsql pod and then the template1 database:

    kubectl exec -it <pgsql-pod-name> -- sh
    $ psql -U sg template1
  • Drop the sg database, recreate it, then exit the psql session (but still in the pod exec):

      # DROP DATABASE sg;
      # \q

    Note: This is a process that we have to do to work around how Sourcegraph sets up the databases as at the time of writing this documentation. Our database pods automatically create the sg database on startup if it isn’t present. Our liveness probes also check for it which means it not being present can cause k8s to kill the pod.

  • Restore the database dump:

  $ psql -U sg -f /tmp/sourcegraph_db.out sg  // exec into the pgsql-pod and run this command
  $ psql -U sg -f /tmp/codeintel_db.out sg   // exec into codeintel_db-pod and run this command

  • Exec into the pgsql and run SELECT * FROM REPO LIMIT 5; to be sure the restoration was complete.

  • Run kubectl apply --recursive -f base/ to apply the changes.

  • Run kubectl get pods -o wide to make sure your pods are running again.

  • Start the remaining Sourcegraph services by following the steps in applying manifests.

Additional Information

  • Since your new deployment was running on a port before the restore, there is a probability that you might lose connection. Run sudo lsof -i -P | grep LISTEN to see the services using the port and kill the PID with kill -9 <PID_number>.

  • You can now port forward your k8s deployment to localhost or wherever your deployment is hosted. For example, kubectl port-forward svc/sourcegraph-frontend 3080:30080 (note you may need to add -n $NAMESPACE to the command).