Sign in Subscribe

Production-Grade (ish) Rails deployment on Hetzner with Kamal

Production-Grade (ish) Rails deployment on Hetzner with Kamal
Photo by Dash Khatami / Unsplash

I've been toying around with Kamal for some time now, and I believe I have come up with a nice setup for a reasonably robust deployment. It includes two servers, one for the application and another for database and caching, a firewall to expose only the ports we need (e.g., the database server is not exposed at all), and automated backups. Let's get down to business.

Part 1: Setting up the servers

While experimenting with Kamal, I was frustrated with the manual labor for setting up and tearing down servers on Hetzner. It was a lot of steps, and since I wanted to customize a couple of things, I also needed to update and/or install things on the server before the deployment. I created a Terraform script to do that for me; you can copy it here.

Here is what the final architecture looks like:

Only the web server is exposed, accessories is blinded from outside access. They are on the same subnet, so they can communicate freely. The only way to ssh into accessories is by using web as a jump server. No root password is set; Login is possible using a private/public key.

Here is what my SSH configuration looks like:

Host web
  HostName <public IP>
  User root
  PreferredAuthentications publickey

Host accessories
  HostName 10.0.0.3
  User root
  PreferredAuthentications publickey
  ProxyJump web

~/.ssh/config

With this configuration, I don't need to use the IP address when configuring Kamal. I will show this later on.

I also used cloud-init to customize some things:

  1. upgrade packages
  2. update nameservers
  3. set the timezone to Berlin
  4. creates a file where our Let's Encrypt will be stored, with the correct permissions
  5. (for accessories, because of Redis) set vm.overcommit_memory=1. Here is why this is important

You can check both customizations under the cloudinit folder.

Part 2: Configuring Kamal

The app I'm deploying uses the following resources:

  1. PostgreSQL as database
  2. Redis for cache
  3. Good Job for background processing

Let's start with configuring the Rails application and the background jobs:

service: <your app name>

image: <username>/image

servers:
  web:
    hosts:
      - web
    labels:
      traefik.http.routers.sumiu.rule: Host(`sumiu.link`)
      traefik.http.routers.sumiu_secure.entrypoints: websecure
      traefik.http.routers.sumiu_secure.rule: Host(`sumiu.link`)
      traefik.http.routers.sumiu_secure.tls.certresolver: letsencrypt
    env:
      secret:
        - GOOD_JOB_USERNAME
        - GOOD_JOB_PASSWORD
  urgent_job:
    hosts:
      - web
    env:
      clear:
        CRON_ENABLED: 1
        GOOD_JOB_MAX_THREADS: 10 # Number of threads to process background jobs
        DB_POOL: <%= 10 + 1 + 2 %> # GOOD_JOB_MAX_THREADS + 1 + 2
        NEW_RELIC_APP_NAME: sumiu-urgent-job
    cmd: "bundle exec good_job --queues urgent"
  low_priority_job:
    hosts:
      - web
    env:
      clear:
        GOOD_JOB_MAX_THREADS: 3 # Number of threads to process background jobs
        DB_POOL: <%= 3 + 1 + 2 %> # GOOD_JOB_MAX_THREADS + 1 + 2
        NEW_RELIC_APP_NAME: sumiu-low-priority-job
    cmd: "bundle exec good_job --queues '+low_priority:2;*'"

There's not much happening in the web application (not to be confused with the web server name) other than some labels we set up for Traefik (more on that later).

On the background jobs part, though, some things are worth mentioning. I set up two distinct processes, one for urgent jobs (urgent_job) and another process for low-priority jobs (and everything else). The DB_POOL variable needs to match the GOOD_JOB_MAX_THREADS configuration as per the documentation: We need the pool size to be MAX_THREADS + 1 connection used by the job listener + 2 connections used by the cron scheduler and executor.

This is what my database configuration looks like

production:
  <<: *default
  url: <%= ENV["DATABASE_URL"] %>
  pool: <%= ENV["DB_POOL"] || ENV.fetch("RAILS_MAX_THREADS", 5).to_i %>

config/database.yml

For Good Job, it uses DB_POOL and for Rails, it uses RAILS_MAX_THREADS.

Part 2.1: Accessories

I'm just gonna dump the accessories part of the configuration and then break it down

accessories:
  db:
    image: postgres:16-alpine
    host: accessories
    port: 5432
    env:
      clear:
        POSTGRES_USER: postgres
        POSTGRES_DB: postgres
      secret:
        - POSTGRES_PASSWORD
    directories:
      - data/postgres:/var/lib/postgresql/data
    files:
      - infrastructure/postgres/postgresql.conf:/usr/local/share/postgresql/postgresql.conf.sample

  redis:
    image: redis:7-alpine
    host: accessories
    port: 6379
    directories:
      - data/redis:/data
    files:
      - infrastructure/redis/redis.conf:/etc/redis/redis.conf
      - infrastructure/redis/redis-sysctl.conf:/etc/sysctl.conf
    cmd: redis-server /etc/redis/redis.conf

  backups:
    image: eeshugerman/postgres-backup-s3:16
    host: accessories
    env:
      clear:
        SCHEDULE: "@daily" # optional
        BACKUP_KEEP_DAYS: 7 # optional
        S3_BUCKET: pg-sumiu-backups
        S3_PREFIX: backup
        POSTGRES_HOST: 10.0.0.3
        POSTGRES_DATABASE: sumiu_production
        POSTGRES_USER: postgres
      secret:
        - POSTGRES_PASSWORD
        - S3_ACCESS_KEY_ID
        - S3_SECRET_ACCESS_KEY

For PostgreSQL, I'm using version 16, an Alpine-based image with some custom configuration (I've used PgTune). What is worth noting here is that files section where I'm mounting my custom posgresql.conf onto postgresql.conf.sample. During the image setup, the .sample file gets copied within the image and becomes the postgesql.conf file used by Postgres. If you mount the file directly over /etc/postgres/postgresql.conf , you will see an error message, saying the database couldn't be set up because the db directory is not empty, so, pay attention to that.

I'm customizing Redis for cache since the default key eviction is to not evict but return an error. I mounted my redis.conf file onto /etc/redis/redis.conf with the following setup:

bind 0.0.0.0 ::
port 6379

protected-mode no

dir /data

tcp-backlog 511

maxmemory 128mb
maxmemory-policy allkeys-lru

redis.conf

I'm also backing up my database daily using Postgres Backup S3 image. It is pretty handy and it includes commands to manually trigger a deployment or a restoration.

Part 3: Traefik and SSL

This is my Traefik setup

traefik:
  image: traefik:v2.10.7
  options:
    publish:
      - "443:443"
    volume:
      - "/letsencrypt/acme.json:/letsencrypt/acme.json"
  args:
    entryPoints.web.address: ":80"
    entryPoints.websecure.address: ":443"
    entryPoints.web.http.redirections.entryPoint.to: websecure
    entryPoints.web.http.redirections.entryPoint.scheme: https
    entryPoints.web.http.redirections.entrypoint.permanent: true
    certificatesResolvers.letsencrypt.acme.email: "my@email.com"
    certificatesResolvers.letsencrypt.acme.storage: "/letsencrypt/acme.json"
    certificatesResolvers.letsencrypt.acme.httpchallenge: true
    certificatesResolvers.letsencrypt.acme.httpchallenge.entrypoint: web

Remember the labels on the web section? This is where the magic happens: Traefik will generate a certificate for us and redirect the traffic to the websecure entrypoint, which is our application.

Conclusion

With this configuration, deploying is a matter of running two commands

$ terraform apply
$ kamal setup

Thanks to Guillaume Briday and Adrien Poly for the inspiration!

One more thing...

I figure someone might like to see the whole config file I'm using. Here it is:

service: sumiu

image: <dockerhub_username>/<app_name>

servers:
  web:
    hosts:
      - web
    labels:
      traefik.http.routers.sumiu.rule: Host(`sumiu.link`)
      traefik.http.routers.sumiu_secure.entrypoints: websecure
      traefik.http.routers.sumiu_secure.rule: Host(`sumiu.link`)
      traefik.http.routers.sumiu_secure.tls.certresolver: letsencrypt
    env:
      clear:
        NEW_RELIC_APP_NAME: sumiu-web
      secret:
        - PGHERO_USERNAME
        - PGHERO_PASSWORD
        - GOOD_JOB_USERNAME
        - GOOD_JOB_PASSWORD
  urgent_job:
    hosts:
      - web
    env:
      clear:
        CRON_ENABLED: 1
        GOOD_JOB_MAX_THREADS: 10 # Number of threads to process background jobs
        DB_POOL: <%= 10 + 1 + 2 %> # GOOD_JOB_MAX_THREADS + 1 + 2
        NEW_RELIC_APP_NAME: sumiu-urgent-job
    cmd: "bundle exec good_job --queues urgent"
  low_priority_job:
    hosts:
      - web
    env:
      clear:
        GOOD_JOB_MAX_THREADS: 3 # Number of threads to process background jobs
        DB_POOL: <%= 3 + 1 + 2 %> # GOOD_JOB_MAX_THREADS + 1 + 2
        NEW_RELIC_APP_NAME: sumiu-low-priority-job
    cmd: "bundle exec good_job --queues '+low_priority:2;*'"

registry:
  username: luizkowalski
  password:
    - KAMAL_REGISTRY_PASSWORD

env:
  clear:
    RAILS_MAX_THREADS: 3
    WEB_CONCURRENCY: 2
    RAILS_ENV: production
  secret:
    - DATABASE_URL
    - RAILS_MASTER_KEY
    - SENTRY_DSN

builder:
  remote:
    arch: amd64
    host: web
  cache:
    type: registry
  args:
    GIT_REV: <%= `git rev-parse --short HEAD` %>
    BUILD_DATE: <%= `date -u +"%Y-%m-%dT%H:%M:%S %Z"` %>

accessories:
  db:
    image: postgres:16-alpine
    host: accessories
    port: 5432
    env:
      clear:
        POSTGRES_USER: postgres
        POSTGRES_DB: postgres
      secret:
        - POSTGRES_PASSWORD
    directories:
      - data/postgres:/var/lib/postgresql/data
    files:
      - infrastructure/postgres/postgresql.conf:/usr/local/share/postgresql/postgresql.conf.sample
  redis:
    image: redis:7-alpine
    host: accessories
    port: 6379
    directories:
      - data/redis:/data
    files:
      - infrastructure/redis/redis.conf:/etc/redis/redis.conf
      - infrastructure/redis/redis-sysctl.conf:/etc/sysctl.conf
    cmd: redis-server /etc/redis/redis.conf
  backups:
    image: eeshugerman/postgres-backup-s3:16
    host: accessories
    env:
      clear:
        SCHEDULE: "@daily" # optional
        BACKUP_KEEP_DAYS: 7 # optional
        S3_BUCKET: pg-sumiu-backups
        S3_PREFIX: backup
        POSTGRES_HOST: 10.0.0.3
        POSTGRES_DATABASE: sumiu_production
        POSTGRES_USER: postgres
      secret:
        - POSTGRES_PASSWORD
        - S3_ACCESS_KEY_ID
        - S3_SECRET_ACCESS_KEY

healthcheck:
  max_attempts: 20

traefik:
  image: traefik:v2.10.7
  options:
    publish:
      - "443:443"
    volume:
      - "/letsencrypt/acme.json:/letsencrypt/acme.json" # To save the configuration file.
  args:
    # accesslog: true
    # accesslog.format: json
    entryPoints.web.address: ":80"
    entryPoints.websecure.address: ":443"
    entryPoints.web.http.redirections.entryPoint.to: websecure
    entryPoints.web.http.redirections.entryPoint.scheme: https
    entryPoints.web.http.redirections.entrypoint.permanent: true
    certificatesResolvers.letsencrypt.acme.email: "luizeduardokowalski@gmail.com"
    certificatesResolvers.letsencrypt.acme.storage: "/letsencrypt/acme.json"
    certificatesResolvers.letsencrypt.acme.httpchallenge: true
    certificatesResolvers.letsencrypt.acme.httpchallenge.entrypoint: web

asset_path: /rails/public/assets

primary_role: web

Icons created by Freepik - Flaticon