add docker-compose documentation for deploying on aws/gcp/digital ocean

Created by: ggilmore

Overview

This PR adds documentation for deploying docker-compose on each of the major cloud providers: AWS, GCP, and Digital Ocean. Each set of instructions gives recommendations on what machine size to choose, how much storage to give the machine, and documentation pointers about how to create backups of their Sourcegraph instance.

This PR also tells people what deployment option they should choose when they're starting out:

Deployment Type	Suggested for	Setup time	Multi-machine?	Auto healing?	Monitoring?
Single-container server	Local testing	60 seconds	Impossible	No	No
Docker-compose	Small & medium production deployments	5 minutes	Possible	No	Yes
Kubernetes	Medium & large highly-available cluster deployments	30 minutes	Easily	Yes	Yes

A later PR will give instructions for migrating off of sourcegraph/server to Docker Compose.

Notes:

Startup scripts

Each of the cloud provider docs provides a startup script to initialize the instance. Each of the scripts follows this general flow:

1. Install git
2. Clone https://github.com/sourcegraph/deploy-sourcegraph-docker to the installation directory (AWS: /home/ec2-user/deploy-sourcegraph-docker, GCP/DO: /root/deploy-sourcegraph-docker) and checkout the latest release
3. Prepare the external disk (used for storing all of Docker's data) and mount it into the file system
   • Using the external disk's device name provided in the startup script, check to see if the disk already has an existing file system
     • If it's unformatted, then format with xfs for AWS, and ext4 for GCP/DO
     • If the disk already has a filesystem, do nothing (this prevents data from being destroyed if the user is using a disk that had a backup restored onto it).
     • Notes:
       • The same init script can be used on both the initial install and when spinning up a VM that has existing sourcegraph data on it.
       • Digital Ocean's webUI doesn't allow you to attach an existing volume to a new droplet during the creation process - you can only do so after the instance is initialized. I'll need to investigate instructions for how to spin up a backup on DO (simply run the init script after you login for the first time)?
   • Add the external disk's UUID to /etc/fstab so that the disk is automatically mounted upon reboots.
     • A lot of cloud provider documentation said that you can't rely on the device name being stable between reboots. Some cloud providers allow you to give names to disks, but the resulting label in /dev/ wasn't predictable in my experience. I think using the device name is fine for now since it's only used for the initial format and the UUID is what actually gets stored in /etc/fstab.
4. Install docker
5. Install jq
6. Use jq modify /etc/docker/daemon.json to change the data-root setting to point to /mnt/docker-data.
   • All of docker-compose's data is stored via docker volumes. Telling Docker to store all of its data under /mnt/docker-data allows customer's to easily snapshot/backup all of the user data in a similar workflow to what they have with sourcegraph/server.
7. Restore the docker daemon to pickup these changes
8. Install docker-compose
9. Run docker-compose up -d to start the sourcegraph/services and restart them on reboots

Backups

Docker Compose doesn't offer any built-in tool to mass backup Docker volumes, and Docker's documentation only offers instructions for how to create/restore a tarball of a container on the individual container level. (I haven't tested this yet, but going down this route would involve custom scripting and would require managing these tarballs somewhere such as an S3 bucket, etc.)

As mentioned above, most of our customers currently solve this by mounting an external disk to the ~/.sourcegraph directory. I have replicated this workflow by telling Docker to store all of its data (including volumes) under /mnt/docker-data. The docs also contain pointers for how to take snapshots of the external disk.

In addition to the above method, I also call out that using an external Postgres service is an easy way to backup all of Sourcegraph's user data.

---

Notes:

I have personally tested these docs for AWS, GCP, and DO. I have also tested restoring backups on AWS and GCP (and not DO for reasons stated above).

I would still like someone to take an in depth look at this PR, the wording, and actually test out my instructions, but this PR needs to be merged already so that customers and take a look.