KNIME Hub Inspect shows “Connection lost” while workflow execution is successful

KNIME Business Hub
1

Description:

We are experiencing an issue with the Inspect functionality in KNIME Hub deployed on a Kubernetes environment.

The observed behavior is as follows:

  • Workflows execute successfully without any issues
  • System resources (CPU / memory) are sufficient, with no noticeable bottlenecks
  • However, when clicking “Inspect”, the following error appears:

Connection lost. Try again later.

Observed Behavior:

  • The issue is consistently reproducible (not intermittent)
  • Even when executing a workflow with only a single node, the problem still occurs
  • The issue persists during low-load periods (e.g., testing at 22:00)
  • All users are affected (not related to user permissions)

Environment:

  • KNIME Hub deployment: Kubernetes (K8s)
  • Deployment size: Single-node server (no multi-node cluster)
  • No Nginx or external reverse proxy in use
  • Server resources are sufficient and workflow execution is stable

Investigation So Far:

We have confirmed:

  • Workflow execution (Executor) is functioning correctly
  • This is not a resource limitation issue
  • The problem is not specific to any particular workflow or node

However, the Inspect functionality consistently fails.

Suspected Causes:

We suspect the issue may be related to:

  • Communication between Hub frontend and backend (e.g., WebSocket / persistent connections)
  • Kubernetes networking configuration (Service / Ingress)
  • Internal Hub services (e.g., execution service / job manager) communication
  • Browser connection handling or session timeout mechanisms

Questions:

We would like to understand:

  1. What services or communication mechanisms does the Inspect feature rely on (e.g., WebSocket)?
  2. Are there any specific networking or port requirements for Inspect when deploying on Kubernetes?
  3. Are there any known issues or similar cases related to this behavior?
  4. What logs (e.g., Hub logs, executor logs) would be most helpful for further troubleshooting?

Additional Information:

We will attach:

01
011809×1034 56.7 KB

02
021496×1057 32.8 KB

  • Screenshot of the error message
  • System/runtime screenshots
  • Deployment architecture overview
1 Like
2

We have the same issue.

In our case, after clicking a few times “Inspect” it eventually works. I suspect that when the job is big, it needs some time to be loaded to the executor. During this time, if shows “Connection lost”.

The reason, as far as I can tell, is that when you click on “Inspect”, if the job has already been swapped into disk, it needs to be loaded again into the executor, and this can take some time depending on the job size.

Would be great to have a better way to deal with it, since it can be confusing to users.

3

Thank you very much for your reply.To rule out resource-related issues, I tested with a minimal workflow containing only one single node, but the same issue (Connection lost) still occurs consistently. This behavior seems quite unusual.Thanks again for sharing your insights.

4

Hi @DanielHua ,

I recommend that you open a support ticket and upload a support bundle from your KBH. There are a couple different possible causes for this (that I know of) so it will be easiest to troubleshoot with a support bundle.
Ideally, you can open the ticket from the Customer Portal, and guidelines for opening tickets are available in this article: Best practice to submit a ticket

1 Like
5

Noted, thank you very much for your advice.
I need to first find out how to create a support ticket and upload the support bundle from KBH as soon as possible.
I will follow the best practice guidelines.
Thanks again for your support.