Postgres

Keep in mind providing postgresql.conf, pg_hba.conf and recovery.signal to data directory if necessary. Examples are in this repo.

Redis

Possible parameters to adjust (using redis.conf, overrides.conf or args ):

• maxmemory 60gb
• maxmemory-policy noeviction
• tcp-backlog 511
• tcp-keepalive 60
• timeout 300

RedisGraph

TODO: Switch to redis-stack (new binary from redislabs). But it seems built-in algos are a bit different.

Cron

Business logic is handled by Clockwork inside of Carbo, but PostgreSQL maintenance is scheduled using simple cron.

# backup
0 4 * * 1 curl -X POST "https://captain.dev.viktorvsk.com/api/v2/user/apps/webhooks/triggerbuild?namespace=captain&token=$TOKEN"

# backup cleanup
0 4 * * 2 curl -X POST "https://captain.dev.viktorvsk.com/api/v2/user/apps/webhooks/triggerbuild?namespace=captain&token=$TOKEN"

# restore and check
0 3 * * 7 curl -X POST "https://captain.dev.viktorvsk.com/api/v2/user/apps/webhooks/triggerbuild?namespace=captain&token=$TOKEN"

# wal cleanup
0 4 1,5,10,15,20,25,28 * *  curl -X POST "https://captain.dev.viktorvsk.com/api/v2/user/apps/webhooks/triggerbuild?namespace=captain&token=$TOKEN"

Caprover

Currently Caprover is used to deploy recar.io. Its a self-hosted PaaS on top of docker swarm, nginx and letsencrypt

Pre-requisits

• Install docker (e.g. https://www.digitalocean.com/community/tutorials/how-to-install-and-use-docker-on-ubuntu-22-04)
• Setup UFW (don't forget SSH port)
• Ensure SMTP ports are not blocked by provider

Custom commands

If you want to run different workers from a single Dockerfile:

TaskTemplate:
  ContainerSpec:
    Command: ["bundle", "exec", "sidekiq"]

One-time commands (e.g. rails db:migrate)

TaskTemplate:
  RestartPolicy:
    Condition: none

They can also be combined together (More details here):

TaskTemplate:
  ContainerSpec:
    Command: ["rails", "db:migrate"]
  RestartPolicy:
    Condition: none

Limitations

• Shared ENV variables do not exist here. For server/worker/migrate/rollback/etc ENV variables should be managed separately.
• Callbacks after success/fail deploy (migrate/rollback) are not implemented so migrations are executed at the same time as the main deploy. It makes develop backward-compatible migrations and adjust database in multiple steps. It would be much better to avoid db rollbacks completely in such setup.
• Labels override. Caprover updates docker service in a "stateless" way. So if you assign LABEL_1=A and then you assign LABEL_2=B — service will have 2 labels assigned. It brings some difficulties for ofelia, traefik and other labels-based services.

Redis HA

Redis Sentinel is configured already in Carbo and Borax and only requeris Sentinel and Master to connect to (with optional Replicas).

PostgreSQL HA

Currently single instance is used but HA setup with Pgpool-II, Patroni and manual replica with streaming replication was configured and tested in case its necesasry.