Environment
- Operating system
- Ubuntu 24.04.5 LTS (container, kernel 6.18.44 built without IPv6)
- Versions
- Docker Engine 29.8.2Docker Compose 5.6.0n8n 2.42.6
A development environment, not a server: dockerd was started by hand inside the container, with cgroup v1.
Symptom
In the development runs of ANAROQ's Compose verifier on 2026-10-11, the n8n service restarted in a loop in each of n8n's eight runs (two variants, four exposure modes). The same message appeared in the logs of every run.
The TCP check on port 5678 failed with connection-closed and the HTTP check with connection-error, while n8n-postgres was healthy and n8n-runner was running.
Root cause
n8n listens on :: by default. The kernel of that environment was built without IPv6: there is no /proc/net/if_inet6, on the host or inside containers, so n8n cannot bind the address and exits.
The other apps in the same environment were not affected: PostgreSQL, Open WebUI+Ollama, WordPress+MariaDB, Uptime Kuma and Gitea passed 44 of 44 runs.
Fix
Run n8n on a host whose kernel has IPv6. On GitHub's ubuntu-24.04 runner, all eight n8n runs of the same template passed on 2026-10-12 (run 38256885780); the template was not changed.
Check the host before you deploy: if /proc/net/if_inet6 is missing, the kernel has no IPv6. ANAROQ has not tested changing n8n's listen address as a workaround, so this note does not recommend one.
Commands
test -e /proc/net/if_inet6 && echo "IPv6: yes" || echo "IPv6: no"docker compose logs n8n | grep "is not available"Evidence
n8n's address '::' is not availablen8n.tcp.5678 connection-closed
n8n.http connection-error
n8n-postgres healthy
n8n-runner runningn8n 8 runs (default, queue-mode-true × 4 exposure modes) 8/8 passed profileVersion 1.1.0Related tools and pages
Sources
How profiles and verified Compose files are built is described in the methodology. Read the methodology