infrastructure/op_guide/update_matrix/README.md

3.5 KiB

How to update Matrix?

1. Build the new containers

Often, I update Riot Web and Synapse at the same time.

  • Open app/docker-compose.yml and locate riot (the Element Web service) and synapse (the Matrix Synapse server). There are two things you need to do for each service:

    • Set the VERSION argument to the target service version (e.g. 1.26.0 for Synapse). This argument is then used to template the Dockerfile.

    The VERSION value should match a github release, the link to the corresponding release page is put as a comment next to the variable in the compose file;

    • Tag the image with a new incremented version tag. For example: superboum/amd64_riotweb:v17 will become superboum/amd64_riotweb:v18.

    We use the docker hub to store our images. So, if you are not superboum you must change the name with your own handle, eg. john/amd64_riotweb:v18. This requires that you registered an account (named john) on https://hub.docker.com.

So, from now we expect you have:

  • changed the VERSION value and image name/tag of riot
  • changed the VERSION value and image name/tag of synapse

From the /app folder, you can now simply build and push the new images:

docker-compose build riot synapse

And then send them to the docker hub:

docker-compose push riot synapse

Don't forget to commit and push your changes before doing anything else!

2. Deploy the new containers

Now, we will edit the deployment file app/im/deploy/im.hcl.

Find where the image is defined in the file, for example Element-web will look like that:

  group "riotweb" {
    count = 1

    task "server" {
      driver = "docker"
      config {
        image = "superboum/amd64_riotweb:v17"
        port_map {
          web_port = 8043
        }

And replace the image = entry with its new version created above. Do the same thing for the synapse service.

Now, you need a way to access the cluster to deploy this file. To do this, you must bind nomad on your machine through a SSH tunnel. Check the end of the parent README.md to do it. If you have access to the Nomad web UI when entering http://127.0.0.1:4646 you are ready to go.

You must have installed the Nomad command line tool on your machine (also explained in the parent README.md).

Now, on your machine and from the app/im/deploy folder, you must be able to run:

nomad plan im.hcl

Check that the proposed diff corresponds to what you have in mind. If it seems OK, just copy paste the nomad job run ... im.hcl command proposed as part of the output of the nomad plan command.

From now, it will take around ~2 minutes to deploy the new images. You can follow the deployment from the Nomad UI. Bear in mind that, once the deployment is done on Nomad, you may still need to wait some minutes that Traefik refreshes its configuration.

If everythings worked as intended, you can commit and push your deployment file.

If something went wrong, you must rollback your deployment.

  1. First, find a working deployment with nomad job history
  2. Revert to this deployment with nomad job revert

Now, if the deployment failed, you should probably investigate what went wrong offline. I built a test stack with docker-compose in app/<service>/integration that should help you out (for now, test suites are only written for plume and jitsi).