n8n · Help

n8n shows Connection lost while workflows keep running?

Blank n8n editor or repeated disconnects? Separate browser, connection and backend health without replaying active operations.

Source date: 5 November 2026Editorial team: KAN KAI WORKSEditorially updated:

Quick answer

Connection lost in the n8n editor does not necessarily mean an execution stopped. In the documented Cloud report, backend processing continued. First establish the destination state using business references, then compare a small workflow with the affected large one. Add browser observations and appropriate health or readiness signals; a successful /healthz response alone confirms neither database readiness nor complete editor functionality.

Source basis: Q1 · Q2

Applies to
The original case concerns n8n Cloud and a large workflow without a universally confirmed cause. Process, proxy and memory metrics depend on accessible hosting infrastructure.
How to verify the result
Both the small comparison workflow and the previously affected editor open reliably, while expected business results remain traceable without unintended repetition of external actions.
Your first check

Is only the editor connection missing, or also the intended destination result?

The n8n editor turns blank, loads slowly or shows “Connection lost”. Meanwhile, workflow results continue to appear at the destination. Restarting the workflow on a guess can create a new problem: the original execution may still be working.

Two different layers

A Cloud report describes repeated editor disconnects around a large workflow while backend execution continued. This does not establish one universal cause. Q1 n8n also explains that /healthz confirms reachability, not database health; readiness has a separate endpoint. Q2

Observe browser rendering, connectivity to the instance and actual processing independently. Memory pressure is one possible explanation that requires evidence. n8n's memory guidance discusses relevant loads, including large datasets and manual executions. Q3 The existence of a large workflow alone is not a diagnosis.

Narrow down the failure

  1. Establish run state outside the editor. Inspect destination results using business references and timestamps. Do not repeat a writing action merely because the editor failed to display its outcome.
  2. Open a small comparison. Does a small workflow work in the same instance while the large editor stalls? That suggests a different investigation from every page being unavailable. Treat it as a clue, not proof of a cause.
  3. Capture browser observations. Record browser version, timestamp and whether requests respond slowly, fail completely or finish before the display freezes. A redacted network trace can identify failed requests and repeated connection drops.
  4. Compare appropriate health signals. Examine the available health and readiness endpoints, including response time and status within the same period. An isolated 200 response is not an end-to-end editor check.
  5. Account for the hosting model. For self-hosting, add proxy and process logs plus available memory and restart data. On n8n Cloud, send the observed evidence to support. Access to worker or system metrics should not be assumed.

Verify the effect of a justified change

After a targeted correction, both a small workflow and the originally affected large workflow should open reliably. Compare actual business results at the same time, so that a recovered display does not conceal a processing problem elsewhere.

Prepare one short incident timeline connecting the browser failure, health responses and business outputs. It helps distinguish an intermittent network path from expensive editor rendering or a backend restart without exposing workflow credentials.

Frequently asked questions

Does Connection lost mean the execution stopped?

Not necessarily. Establish backend and destination state independently before repeating any operation.

Should I delete all browser data or the entire workflow?

No. Preserve the evidence first. Targeted comparisons with a small view and another permitted browser profile provide a more useful basis for deciding what to change.

Help with this case

We can help isolate the affected process and prepare a controlled next step. Discuss case A0147 with KAN KAI WORKS.

Support with this issue

We investigate n8n using your error message, versions and a narrowly scoped example.

Describe your case →Related service: AI & automation

Sources

Sources checked on 5 November 2026. Individual reports do not establish general prevalence.