Make container restart-policy configurable (currently hardcoded RESTART_MODE = 'unless-stopped')

[netholics]

Summary

Coolify hardcodes RESTART_MODE = 'unless-stopped' in bootstrap/helpers/constants.php:18 and applies it to every managed service on (re)deploy. There's no way (UI, env var, per-service override) to use a different restart policy without editing the source.

Use case: our coolify-proxy needs restart: always, not unless-stopped. Under certain host-shutdown signals (exit 255), Docker treats unless-stopped containers as user-stopped and skips auto-restart after the host comes back — leaving the proxy down post-recovery. All other Coolify core containers (coolify, coolify-db, coolify-realtime, coolify-redis) ship with always. Only coolify-proxy is on unless-stopped, and that's the practical difference.

We can docker update --restart=always coolify-proxy as a runtime workaround, but Coolify regenerates the proxy compose on every proxy redeploy and the workaround is silently reverted. We currently run an hourly cron just to alert on the drift.

Current behavior

RESTART_MODE is referenced from 18 sites in the codebase:

• bootstrap/helpers/constants.php:18 — the constant definition
• bootstrap/helpers/proxy.php:277, 362 — proxy compose generation
• bootstrap/helpers/shared.php:2571, 3366 — service helpers
• bootstrap/helpers/parsers.php:653, 1985 — service parsers
• app/Jobs/ApplicationDeploymentJob.php:2616 — app deployments
• app/Actions/Database/Start{Mariadb,Postgresql,DatabaseProxy,Mysql,Redis,Clickhouse,Mongodb,Keydb,Dragonfly}.php — database deployments (9 files)
• app/Actions/Server/ConfigureCloudflared.php:22 — Cloudflared

In proxy.php specifically, data_forget($config, 'services.traefik.restart') is called before the restart key is re-injected from RESTART_MODE, so any user edit to the regenerated compose file is stripped on next deploy.

Proposal

Two options, ordered by smallness of change:

Option A — env-overridable constant. Single line in constants.php:

const RESTART_MODE = env('COOLIFY_RESTART_MODE', 'unless-stopped');

Lets operators set COOLIFY_RESTART_MODE=always in Coolify's own .env and have it apply system-wide. Backward compatible: default unchanged.

Option B — per-service override. A config map keyed by service:

restart_policies:
  proxy: always               # override default
  postgresql: unless-stopped  # explicit default

More flexible, but a larger change touching ~14 call sites.

Either closes our use case. Option A is the smallest possible fix; happy to PR it if you're open.

Environment

• Coolify: 4.0.0 (ghcr.io/coollabsio/coolify:4.0.0)
• Docker: 28.1.1
• Ubuntu 24.04.4 LTS

Receipts

Bug surfaced via a kernel soft-lockup incident on our VPS (2026-05-17). All four always-policy core containers auto-restarted on host recovery; only coolify-proxy (the unless-stopped one) required manual docker start. Recovery delay: ~30 minutes from host-up to proxy-up.

[Cinzya]

The coolify-proxy restart policy can already be edited from the UI when navigating to Servers > [YourServer] > Proxy > Configuration. Changes made in there will also persist when you press save. As Coolify saves the complete docker compose configuration into it's own database, which is then re-used when regenerating the proxy.