Repository navigation
Description
Production deployment blocker — SSH unreachable from GitHub Actions
The Deploy to dev2 workflow is failing on multiple consecutive merges.
Evidence:
- Run 43685309859 (Scallion release PR chore(release): signed v0.3.77 Scallion foundation release on merge #387 merge), step
Deploy:
ssh: connect to host *** port 22: Connection timed out - Run 43685392929 (Scallion exit policy PR feat(scallion): fail-closed exit consent, private-IP blocking, and DNS pinning #388 merge): same error.
- Both failed before
deploy-app.shcould run. This is a transport-level inability to reach TCP/22, not a Rust build failure or SSH public-key rejection. - Production c0mpute.com health check is skipped when SSH fails. Therefore GitHub merge does not prove live deployment.
Required operator diagnosis (avoid posting secrets/IPs)
On dev2, over a known working admin console/session:
sudo systemctl status ssh || sudo systemctl status sshd
sudo ss -lntp | grep ':22'
sudo ufw status verbose
sudo journalctl -u ssh -n 100 --no-pager
sudo fail2ban-client status sshd 2>/dev/null || trueFrom an authorized external system, check the hostname configured as the DEV2_HOST secret resolves to the intended public or reachable overlay address and TCP/22 is reachable. Also inspect upstream firewall rules, ACLs, provider networking, SSH listen port, access restrictions and recent changes. A bad key normally results in Permission denied, not a TCP connection timeout.
Do not blindly disable the firewall, change host-key verification, paste SSH secrets, or expose SSH publicly to all GitHub runner IPs. Prefer a scoped reachable endpoint, a tightly controlled overlay connection (Tailscale/WireGuard with GitHub OIDC where provisioned), or an isolated self-hosted deployment runner with appropriate access.
Acceptance
- GitHub Actions deploy can establish an authenticated, host-key verified SSH connection to dev2.
- The remote
deploy-app.shcompletes on the exact expected master SHA. - The workflow verifies
https://c0mpute.com/and reports success. - Deployed build/version is verified; operator can roll back.
- Failure mode does not cause repeated ~130-second timeouts or claim deployment success.
Related: #386, #387, #388, #390. The signed v0.3.77 GitHub release is a separate pipeline; a published GitHub release does not mean dev2 deployment succeeded.
Activity
ralyodio commentedon Nov 15, 2026
Additional company-repo finding:
profullstack/threatcrushships.github/workflows/unban-dev2.yml, which explicitly documents auto-defence IP bans on dev2 causing SSH timeouts. It describes a GitHub runner SSH workflow as break-glass, but that workflow itself relies on TCP/22 reachability and may also fail if upstream port 22 is blocked. Check the threatcrushd blocklist/remediation rules and provider firewall from the dev2 console before making changes; do not indiscriminately disable protection or whitelist all runner IPs. The two c0mpute deployment attempts timing out are consistent with a firewall/ACL/drop, but the logs alone cannot establish whether ThreatCrush, fail2ban, cloud security groups, host downtime, or another cause is responsible.ralyodio commentedon Nov 17, 2026
Live-site verification note (2026-11-16): c0mpute.com/apps public page is readable, but external attempts to open
https://c0mpute.com/statusreturned HTTP 502 andhttps://c0mpute.com/apps/v1/appscould not be reached using the public fetcher. This does not conclusively prove a user-facing outage (fetcher/CDN restrictions are possible), but it does not validate the live gateway either. Combined with earlier GitHub Actions SSH TCP/22 timeouts, a trusted operator should check dev2 connectivity, edge proxy health and the actual app/gateway process before declaring distributed hosting launched. Do not change firewall/SSH rules blindly.