Docker/Docker-Compose Migrations

This is a crash course for migrating an existing Sourcegraph instance from docker to docker-compose for onboarding Application Engineers. This course assumes that you have already installed a local docker deployment, and a docker-compose deployment using the Google Cloud Platform.

While this crash course focuses on migrating to a GCP instance, by the end of the course you should have a good basis for migrating using other cloud services (AWS, Azure, etc.) as well.

Setup

To set yourself up for success in this crash course, make sure you’ve set up the following:.

  1. Single-container docker instance of Sourcegraph
  2. Docker-compose instance of Sourcegraph on GCP

Once your local instance is deployed, create an account, then add a codehost and a few small repos to your instance so there is something to migrate when the time comes.

NOTE: For your docker-compose instance on GCP, do not register a new user yet. User information from your single-container instance will be added automatically when we migrate the database later.

Prepare the Migration

  1. First, find your local instance’s CONTAINER_ID by running your local instance, then executing the command docker ps. You will see the following output:

    > docker ps
    CONTAINER ID        IMAGE
    ...                 sourcegraph/server
    

    Take the value from the CONTAINER_ID column and export it as a variable in your terminal:

    export CONTAINER_ID=<"CONTAINER ID" value from previous command>
    
  2. Generate database dumps from codeintel-db and psql and save them to the /tmp folder. For codeintel-db:

    docker exec -it "$CONTAINER_ID" sh -c 'pg_dump -C --username=postgres sourcegraph-codeintel' > /tmp/codeintel_db.out
    

    and for the postgres database:

    docker exec -it "$CONTAINER_ID" sh -c 'pg_dump -C --username=postgres sourcegraph' > /tmp/sourcegraph_db.out
    

    There should now be two files in the /tmp folder on your system: sourcegraph_db.out and codeintel_db.out. Navigate to your /tmp folder and make sure this is the case.

    cd /tmp
    
  3. Check the contents of the files and make sure you see information you recognize by using the less command.

    less /tmp/sourcegraph_db.out
    less /tmp/codeintel_db.out
    

    Use the grep command and search for a repo you are certain was indexed in your single-container instance to be sure:

    grep <repo or organization name> sourcegraph_db.out
    

    If it returns expected results, you’re all good to move on to the next step.

Migrating

NOTE: you will need to have access to the gcloud SDK to complete some of the steps in this section. Install instructions here.

  1. First, we need to navigate to the directory containing the docker-compose definition. To do so, SSH into the GCP instance, then sudo -i then cd /deploy-sourcegraph-docker/docker-compose.

  2. Kill the existing Docker Compose containers (and associated volumes) so that we avoid conflicting transactions while modifying the database:

    docker-compose down --volumes
    
  3. Start the postgres instance by itself:

    docker-compose -f db-only-migrate.docker-compose.yaml up -d
    
  4. The next thing we need to do is copy the local dumps into our gcp instance. This step is tricky, and may require some trial and error depending on your environment. We will be using the gcloud compute scp command to do so. Think of scp as the cp command in your terminal. The s just stands for secure. scp is a command that allows you to securely copy files on a local volume to a remote volume and vise versa. It is important to note that if you’re following this crash course using another cloud service like AWS or Azure, this is where you might have to do your own research or use the standard scp command. You can read about it here. I highly recommend familiarizing yourself with the gcp command we’ll be using here. The command is structured like this:

    gcloud compute scp --project=<name of project> --zone=<project time zone> <files you want to move> <gcp instance>:<location within the gcp instance>
    

    For our purposes, let’s say our project name is “my-project”, the time zone is “us-central1-a”, and our gcp instance is “sourcegraph-dc-instance”. If we want to move our local database dumps from the local /tmp folder to the /tmp folder in our remote instance, the command will look like this*:

    gcloud compute scp --project="my-project" --zone="us-central1-a" /tmp/*_db.out sourcegraph-dc-instance:/tmp/
    

    Execute the command with the correct information from your instances. If the command succeeds, you’ll be greeted with output that looks something like this:

    Updating project ssh metadata...done.
    Waiting for SSH key to propagate.
    Warning: Permanently added 'compute.<some long number>' (ECDSA) to the list of known hosts.
    codeintel_db.out                                          100%  101KB 752.5KB/s   00:00
    sourcegraph_db.out                                        100%  313KB   2.9MB/s   00:00
    

    _NOTE: the __db.out marker will indicate that all files ending in _db.out should be moved.

  5. Make sure the files really did copy by navigating to the /tmp file in your gcp instance and finding the files. You can also use grep to search familiar repos to be sure. If you see them, congrats! You’ve completed the trickiest step in this process.

  6. SSH into your gcp instance. Copy the dumps from the /tmp folder into the /tmp folder in the pgsql and codeintel-db containers using the docker specific docker cp command, respectively:

    > docker cp /tmp/sourcegraph_db.out pgsql:/tmp/
    > docker cp /tmp/codeintel_db.out codeintel-db:/tmp/
    
  7. Create a shell session inside the pgsql container in your gcp instance

    docker exec -it pgsql /bin/sh
    

    Restore the Database dump:

    psql --username=sg -f /tmp/sourcegraph_db.out postgres
    

    Open up a psql session inside the pgsql container:

    psql --username=sg postgres
    
    

    Apply the following alteration to the database:

    DROP DATABASE sg;
    ALTER DATABASE sourcegraph RENAME TO sg;
    ALTER DATABASE sg OWNER TO sg;
    

    End the psql session with \q, then exit the pgsql shell session by executing the exit command.

  8. Create a shell session inside the codeintel-db container

    docker exec -it codeintel-db /bin/sh
    

    Restore the Database dump:

    psql --username=sg -f /tmp/codeintel_db.out postgres
    

    Open up a psql session inside the codeintel-db container:

    psql --username=sg postgres
    
    

    Apply the following alteration to the database:

    DROP DATABASE sg;
    ALTER DATABASE "sourcegraph-codeintel" RENAME TO sg;
    ALTER DATABASE sg OWNER TO sg;
    

    End the psql session with \q, then exit the codeintel-db shell session by executing the exit command.

  9. Restart the rest of the containers. Make sure you’re in the directory that contains the docker-compose definition before doing so:

    docker-compose -f docker-compose.yaml up -d
    
  10. Navigate to the external url associated with your gcp instance and attempt to login using the same credentials you used with your local instance. If you’ve done everything correctly, it should successfully validate your credentials and all your repos should’ve started cloning. NOTE: The url may not immediately be available. It can take up to 15 minutes for the url to start working.

Getting Help

If you run into any issues during this crash course, please reach out to Jason Harris.