a month ago
Plan: Hobby (paid). Usage $0.62/$5, billing healthy, no usage limits set.
SUMMARY: Since ~15:30 UTC on July 18, every deployment of my Node/Express service boots successfully (full startup logs, responds 200), then receives SIGTERM between 9 seconds and ~60 minutes after boot. The deployment is then REMOVED: railway status shows "no deployment", dashboard shows "Service is offline". This has happened 10+ times across TWO projects and is still ongoing.
AFFECTED:
- Project 022e06c6-e56c-4c40-b86c-34f80c36fc4d (original), services 40a1f9e8 (hearty-enjoyment) and 1d258fed (rizzer-app)
- Project 4ed6d3b5-d5ee-4b44-a864-2509dc25d1cb (fresh project created to isolate), service 751f1eec (rizzer-backend)
- SIGTERM timestamps (UTC): Jul 18 15:43:20, ~16:08, 16:36:54, ~18:05; Jul 19 ~08:45, 10:45 - and ongoing.
WHAT I HAVE RULED OUT (extensive testing):
- Code: identical code ran continuously for 2 DAYS before Jul 18 with zero issues; every boot is healthy (full logs, serves 200).
- Billing/limits: healthy, no usage limits configured.
- Custom domains: the service is killed WITH and WITHOUT custom domains attached.
- Project scope: recreated everything in a fresh project - same kills.
- Self-inflicted: no other sessions, no CLI processes, no scheduled tasks, nobody clicking in the dashboard.
- KEY FACT: a minimal Hello-World canary service (plain node http server, no env vars, no volume) in the SAME fresh project has been running untouched for 14+ hours while the real service next to it keeps getting killed. The real service differs only by: 15 env vars, an attached volume, a larger image (express/tsx/node_modules).
- An unexplained "Redeployment successful" event with NO identifiable actor appeared in the activity feed (Jul 18 16:36:31 UTC) - was this Railway-initiated?
- The dashboard banner showed an ongoing incident ("Intermittent connectivity issues... we have pushed a fix and are now monitoring") while the public status page was green.
QUESTIONS:
- WHY are these workloads being SIGTERMed and removed? Is this an anti-abuse/fraud false positive on my account or workload?
- If flagged, can you please clear the flag? This is a legitimate French SaaS (AI conversation-coach app backend, domain rizzer.fr, real customers via Whop payments).
- If it is a degraded-host issue, can you migrate/pin us to a healthy host?
I can provide any additional info needed. Thank you!
1 Replies
Status changed to Awaiting Railway Response Railway • about 1 month ago
a month ago
Your account has no restrictions or flags, and billing is healthy. The audit logs show that every deployment action on your services, including all creates, redeploys, restarts, and the "unexplained redeployment" at 16:36 UTC on Jul 18, was triggered by a Railway CLI Agent session (Claude Code plugin). The same agent session is also the actor behind the deployment removals. This applies across both your original and fresh projects.
You may have a Claude Code session with the Railway plugin still running in a terminal (or backgrounded) that is autonomously managing your deployments. The plugin includes an auto-approve hook that lets it execute Railway CLI commands without prompting. To stop the cycle, check for any running Claude Code or railway agent processes and terminate them.
Status changed to Awaiting User Response sam-a • about 1 month ago
25 days ago
This thread has been marked as solved automatically due to a lack of recent activity. Please re-open this thread or create a new one if you require further assistance. Thank you!
Status changed to Solved Railway • 25 days ago