Appearance
When something breaks
Start broad and narrow down.
1. Look
- the ops page
infra status- ask the Telegram bot "what's wrong?"
2. One app is down
In Claude Code: "why is <app> down?". It checks, in this order:
- the latest deployment log
- the app's runtime logs
- env var names (it never prints the values)
- the database's status
- server memory and disk
The usual causes:
| Symptom | Likely cause |
|---|---|
| Build fails early | the base image has no arm64 build |
| Running but unhealthy | wrong exposed port, /health is missing, or there's no curl/wget in the image |
| Crash loop | a missing env var, or the memory limit is too low |
| 502 from Cloudflare | the container isn't listening on 0.0.0.0:$PORT |
| 403 or a login page where you didn't expect one | the hostname matches a Cloudflare Access app |
| Times out, but only from some places | the DNS record isn't proxied, so the firewall drops the traffic |
3. A bad deploy
Roll back to the previous image (Claude can list them), or revert the commit and push.
4. Bad data
Restore the database from its R2 backup. See Databases and backups.
5. The server itself
Disk full, an update gone wrong, a firewall mistake:
infra ssh 'df -h /; free -h; uptime'- clear out Docker through Coolify's docker cleanup (ask Claude), rather than running
docker system prune - as a last resort,
infra snapshot revert <name>
6. Can't get in at all
Known gotchas
- The Coolify root signup window is only 10 minutes long.
docker restart coolifyreopens it. - Resetting Coolify's root password revokes every API token.
- Apps without a domain get a public
sslip.ioaddress unless you clear it. - The S3 keys for R2 are derived from the main Cloudflare token, so rolling that token breaks backups.
- The netcup server panel password isn't the customer panel password. Use SCP Auto-Login instead.