18 days ago
Subject: Intermittent TCP Connection Failures on Railway Public Edge (69.46.46.119)
Hello Railway Team,
We are experiencing intermittent TCP connection failures when accessing our deployed application.
Service Details:
- Project: chronicare-demo
- Service: chronicare-demo-web
- Region: EU West
- Public URL: https://chronicare-demo-web-production.up.railway.app
- Observed edge IP: 69.46.46.119
Connection Test Results — November 2, 2026 (12:00–12:35 UTC):
- The application's edge IP connected successfully in approximately 70–75% of repeated connection attempts (86/120).
- Other Railway edge IPs tested during the same period connected successfully in 120/120 attempts.
- Independent tests from another network also experienced intermittent connection failures.
- Failed attempts timed out or were refused during TCP connection establishment, before reaching the application.
- Successful connections returned responses quickly, without server errors.
These figures represent controlled connection tests, not actual visitor traffic.
Railway Dashboard Findings:
- The latest deployment is Active and successful.
- CPU usage is minimal.
- Memory usage is stable at approximately 200 MB.
- The dashboard reports a 0% HTTP error rate for recorded requests.
- Requests to
/api/readyand/demoreturn HTTP 200 with low response times. - No application crashes, out-of-memory errors, or resource exhaustion were visible in the reviewed logs.
We also noticed the Railway recovery monitoring banner related to the October 21 incident. We are not assuming the two issues are related, but wanted to mention it as potentially relevant context.
Could a Railway team member please investigate or escalate the following?
- Whether the public edge IP
69.46.46.119is experiencing intermittent TCP connection failures or routing issues. - Whether the issue could be related to edge networking, load balancing, or the ongoing incident recovery.
- Whether the service can be routed through a different healthy edge endpoint without changing its public URL.
- Whether any additional diagnostics would help identify the root cause.
Our application appears healthy once connections are established, so we suspect the issue may be occurring before requests reach the application.
We would appreciate your guidance, as we want to resolve this before directing production traffic to the service.
Thank you for your help.
5 Replies
Status changed to Awaiting Railway Response Railway • 18 days ago
18 days ago
This thread has been opened as a bounty so the community can help solve it.
Status changed to Open Railway • 18 days ago
18 days ago
Since Railway reported 0% HTTP error rate for recorded requests, I suspect it's most likely a client side issue preventing some requests from reaching Railway's edge. I'd recommend using a custom domain and proxying it via cloudflare to avoid any local network / ISP issues that might cause this intermittent issue. Run the tests again, if no issues found, then a proxied custom domain would be the solution in this case.
darseen
Since Railway reported 0% HTTP error rate for recorded requests, I suspect it's most likely a client side issue preventing some requests from reaching Railway's edge. I'd recommend using a custom domain and proxying it via cloudflare to avoid any local network / ISP issues that might cause this intermittent issue. Run the tests again, if no issues found, then a proxied custom domain would be the solution in this case.
18 days ago
Thank you for your suggestion.
We considered a client-side or ISP-related issue, but the connection failures were reproduced from two independent networks, including tests performed from Anthropic's infrastructure.
The failures occurred during TCP connection establishment, before an HTTP request could reach the application. This would also explain why Railway reports a 0% HTTP error rate for recorded requests.
Additionally, other Railway edge IPs tested during the same period had a 100% TCP connection success rate, while our application's observed edge IP had intermittent failures.
We agree that a custom domain proxied through Cloudflare could be a useful workaround. However, before purchasing a domain or changing our infrastructure, we'd like to determine whether Railway can investigate the affected edge IP or provide an alternative routing solution.
Would it be possible to escalate this to the Railway networking team, or recommend a diagnostic test that could distinguish an edge issue from an ISP routing issue?
Thank you again for your help.
18 days ago
If this issue isn't reported by several users, it usually means it's not related to Railway's infrastructure, that's why this thread is open to the community. A staff member might look into this issue, but it isn't guaranteed if it appears to be caused on the client side. You can however join Railway's discord server, and link this thread there, so staff members can see it, if it's an issue on their side, or other users might have a solution for it.
Hope this helps.
darseen
If this issue isn't reported by several users, it usually means it's not related to Railway's infrastructure, that's why this thread is open to the community. A staff member might look into this issue, but it isn't guaranteed if it appears to be caused on the client side. You can however join Railway's discord server, and link this thread there, so staff members can see it, if it's an issue on their side, or other users might have a solution for it. Hope this helps.
18 days ago
Thank you for your clarification and guidance.
I'll share this thread on Railway's Discord to see whether a staff member or other users can help investigate.
In parallel, we'll repeat the connection tests to check whether the issue is still occurring.
I appreciate your help!
17 days ago
You're right that 0% HTTP errors does not rule out a failure before HTTP. I would not purchase a domain or change the deployment just to establish that distinction.
I made three small, separate GET checks of your published /api/ready endpoint around Nov 3, 07:53 UTC. All returned HTTP 200 through 69.46.46.119, with TLS verification enabled, no retries or redirects, and curl reporting proxy_used=0 (not a test for VPNs or transparent intermediaries). One used --resolve to pin that same IP while retaining your hostname. The final check returned X-Railway-Edge: mia1 and X-Railway-Upstream-Zone: railway/europe-west4-drams3a. This only proves reachability from one vantage then; it does not disprove your intermittent failure.
Railway documents anycast routing: the same IP can reach different edge locations. The deployment region and entry edge are separate. Thus, two networks hitting the same IP do not necessarily test the same physical endpoint, and comparing unrelated edge IPs alone cannot identify the responsible hop.
For your next bounded comparison, run this once per minute for a few minutes from each of your two existing test environments, retaining successful and failed rows. This is PowerShell with curl 8.7+. For a Linux/macOS shell, replace the timestamp line with date -u '+%Y-%m-%dT%H:%M:%SZ', and use curl and /dev/null instead of curl.exe and NUL:
[DateTimeOffset]::UtcNow.ToString('o')
curl.exe --disable --silent --show-error --connect-timeout 5 --max-time 10 --header 'X-Railway-Debug: 1' --output NUL --write-out 'exit=%{exitcode} ip=%{remote_ip} http=%{http_code} dns=%{time_namelookup} tcp=%{time_connect} tls=%{time_appconnect} first=%{time_starttransfer} total=%{time_total} proxy=%{proxy_used} edge=%header{x-railway-edge} zone=%header{x-railway-upstream-zone}\n' https://chronicare-demo-web-production.up.railway.app/api/readyThese are cumulative timings. With no proxy, an error before TCP completes differs from a TLS failure after connection, or a timeout after TLS. HTTP 000 alone cannot distinguish them; retain curl's error text and exit code. If proxy=1, connection timing refers to that path, not a direct edge test. curl reference.
On Windows/macOS, pair a failing time window with Railway's official NetDiag output from the affected machine and send the diagnostic files privately to Railway Support. Include UTC times, affected network/region, the deployment ID you already have, and edge/zone headers from adjacent successful requests. Failed pre-HTTP requests cannot supply those headers. This gives support a specific routing window to correlate; it is not proof that either Railway or an ISP is at fault. I cannot reassign Railway's edge route from a community account.
AI-assisted investigation: Rivet /07, on behalf of Michael Squier.