I’ve been seeing a constant “Connection lost” issue in the n8n Cloud editor and spent some time isolating it.
The WebSocket connection to the push endpoint connects successfully (HTTP 101), but then drops after about 2–4 seconds and immediately reconnects. This repeats continuously. In the console, it shows:
[WebSocketClient] Connection lost, code=1005
So this isn’t a failed connection — it’s being terminated without a proper close frame.
I went through a pretty thorough set of tests to rule out local issues. The behavior is identical across Chrome (clean profile) and Brave, across multiple machines, and across different networks including Verizon Fios, a Verizon mobile hotspot, and even a VPN connection routed through London. I also disabled QUIC/HTTP3, Secure DNS, and ran without extensions. None of that changed the behavior.
A couple of things stand out. The WebSocket handshake always succeeds, but the connection is never maintained. The exact duration varies slightly depending on region, but it always drops quickly and reconnects. I’m also occasionally seeing 403 responses from an ingest endpoint, which suggests there may be some edge or routing inconsistency.
At this point it doesn’t look client-side. It feels like the WebSocket (push) connection is being dropped somewhere upstream — possibly at the Cloudflare edge or within the n8n Cloud push backend itself.
Workflows still execute fine, but the editor UX is degraded because it can’t maintain a stable real-time connection.
Curious if anyone else on n8n Cloud is seeing this, or if this might be regional or account-specific.
@DPAM does it still happens when you use an incognito window? If that does, then it is clearly a n8n cloud side problem, although as you have said the workflow still executes and works fine, but in case if the problem escalates, feel free to reach out at support channels:
I have the same Lost connection issue. Seems like it only happens when I open two n8n tabs in the same browser window. The problem persists even if I drag one tab to another window. But it doesn’t happen if I open a new window and then navigate to n8n.
To isolate this problem, I tried,
fresh install of n8n with fresh database
install n8n on a different server
many permutations of http vs https in environment variables and while testing access
@DPAM this is a known cloudflare edge thing with n8n cloud, the 1005 close code means no close frame was sent which points to an idle timeout on the proxy layer. since your workflows run fine you can work around the editor flakiness by polling execution status instead of relying on the websocket push. here’s a quick workflow that checks execution status via the n8n API on a loop:
set up an Header Auth credential with your n8n API key, swap the url to https://your-instance.app.n8n.cloud/api/v1 and you can monitor executions without the websocket dropping on you. also worth emailing help@n8n.io about the 403s on the ingest endpoint, that part is definitely on their infra side.
Same problem over here. n8n self hosted, 2.16.2, behind Caddy. If I open 2 workflows on different tabs, it starts the cycle of “connection lost” - all good, on all workflows.
Same issue here, multiple tabs can cause this issue on the latest build. Now for fun, incognito mode seems to do something a little different and it can work.