Feedback on Cluster Explorer (2.6.x) productivity

[adaliszk]

I’m sorry for the long post beforehand.

I’ve recently upgraded to 2.6 as a part of upgrading to the newest EKS version (Add Kubernetes 1.21 for Amazon EKS · Issue #33723 · rancher/rancher · GitHub), and I was experiencing severe problems with the Cluster Explorer as I’m no longer able to use the good old Cluster Manager.

The areas I have the most trouble with:

• Cluster Nodes view
• Deployment view
• Navigation tool
• Overall slowness

To go through point by point:

Cluster Nodes view

• Lost the ability to add a custom name
• Added labels are no longer displayed

Both of these makes it difficult to overview the nodes because with a Cloud Provider provisioning you will end up having cryptic names that tell you nothing. In the previous UI, you could have changed the logical names so that it remains overviewable across the whole system.

Additionally, the labels are extremely useful to see on the main node list view when you do use node selectors. I have no desire to go into each node and then check one-by-one what labels they have, scan through a bunch of auto-generated ones and find the few I care about.

Deployment view

• Lost the ability to quickly check status & logs
• Lost the ability to quickly scale deployments
• Pod restarts are no longer displayed

One feature that we have been using regularly is to open a workload, see if the containers are healthy and enter their logs directly from the list view. This now requires you to open each workload individually, wait for the system to wake up its slumber (usually 3-8s) and then be able to see the logs. However, if you wanted to see what each containers states are, now you have to go into each pod, then wait for another slumber to see anything.

While the logs and status checks are still possible, scaling a deployment is no longer a thing on the UI. I have to edit each workload and change the replicas count there instead of a quick + and - sign.

On the list view, the Pod Restarts are no longer displayed. While this is not the end of the world, it would be great if we could see them as a clear indication that something is not stable.

Navigation tool

• Selection of a Project (or Namespace) no longer redirects to the Deployments/Apps view
• The selection often changes the sidebar

When using the new navigation tool, it’s nice for having more options, however, it no longer switches the views. If you are on the Deployments or Installed Apps view, it does it correctly, but if you trying to switch from a Workload detail or Pod detail view, then it just freaks out and does not navigate you back to the most common top-level section.

Another thing is that depending on what you filter for the sidebar options change. Like, when you select a Project or a Namespace the Cluster menu disappears, so I have no way to quickly go back to the top-level overview.

Overall slowness

• Switching between views are taking 3-8s or more
• Entering detail views quickly load the header information, but you wait for 3-8s for any details
• Browser blocking loading

The most amount of time wasted for me on the new UI is that everything takes a lot of loading time, often without even displaying a loader to indicate that the app is not just frozen, but working on something. I consistently experience 3 or more seconds for each screen I switch.

Additionally, it seems that the loading of the data blocks the Browser to do anything, this makes Rancher look like it’s frozen and broken while it’s only just loading.

Things I wish were addressed

• Cluster Nodes view

  • Add the ability to custom name nodes, or at least display the description
  • Show the custom labels added for each node, maybe make it toggleable

• Deployment view

  • Add the Pod Restarts to the displayed fields, maybe make it configurable
  • Add the opening of the Workloads back to review Pod states quickly
  • Add the quick scale buttons back

• Navigation tool

  • Set the top-level options (Cluster Home, Project/Namespace) to always visible
  • Add auto-navigate to the most common top-level element when changing the context

• Overall slowness

  • Fix it. Set a rule that every page MUST load under 1 second

Additional context

• Rancher I’m using is 2.6.0-rc9
• Rancher is deployed in AWS, on a c4.xlarge EC2 instance
  • CPU usage is usually around 10% mark
  • Memory usage is usually around 30% mark
  • Deployed using Docker (version 20.10.8, build 3967b7d)
• Two clusters are hooked up with this Rancher
  • EKS cluster in the same region (v1.21.2-eks-0389ca3)
  • AKS cluster in the same city region-wise (v1.18.10)
• My internet bandwidth is 1000Mbit
• Mainly using Firefox 91.0.1 (64-bit)

[Rihan]

I agree with a lot of the points shown here, but I have more to add.

The reason I started using Rancher in the first place is the complete ease of managing everything. As a “beginner” to Kubernetes, Rancher made it dead simple for me to manage the cluster and run applications through the Cluster Manager, and separate Cluster Explorer allowed me to grasp more advanced Kubernetes concepts when I needed to modify more things without being thrown in the deep end or compromising on the ease the Cluster Manager gave. This update completely stopped that.

I’ve spent a few hours trying to grasp the new interface but it’s basically just Kubernetes visualised, kind of unnecessarily remaking the Kubernetes Dashboard rather than a tool to make managing a cluster easier. While the slowness is something that can be repaired easily, I think more thought needs to be put into the UX to be more similar to the Cluster Manager rather than folders and folders of stuff. I have to dig through folders and a bunch of unnecessary stuff when I want to edit deployments and manage ingress, where before it was nicely laid out in the UI.

Here’s a fine example: (ignore the red box, stole this off of the old docs)

I’ll be downgrading and wishing that 2.7 fixes this poor UI.

[adaliszk]

Yeah, there are many other things that are just plain annoying now. Like, previously I could just provide the absolute minimum for Helm charts, a sort of values.yaml that only just overwrites the default values, now if it’s not hand-tailored to Rancher you must edit a multiple hundred lines of file and find that 5 lines you are interested in.

While I do value the ability to manage clusters without the concern of what provider I’m using, having just another Kubernetes Dashboard but worst in terms of performance makes me think that I might be better off just using Lens or Kubernetic, and manually configure the clusters in their respective providers.

Sadly, this is partially my own fault too, as the whole Cluster Explorer was in the works for many months if not a year or so already but have not bothered to try to give any feedback nor a decent go at it as the Cluster Admin was the primary tool I was needed.

[jgallucci32]

This thread I believe underscores the importance of how a UI can be used as a training tool for learning a completely new concept or set of technologies. Cluster Manager is what got our business into Kubernetes in the first place and what allowed our business to transition traditional admins over to understanding this entirely new concept of container orchestration. Cluster Explorer (new UI) is a graphical representation of vanilla Kubernetes from a more experienced point of view and tailored toward developers who already have a strong grasp on the concepts of Kubernetes and container orchestration.

The only drawback I ever saw in Cluster Manager was the heavy reliance on websockets and constant refresh and Chrome cache clearing required to prevent long-standing sessions from being bogged down. Other than that, the layout perfectly captured the concepts of Kubernetes for less experienced administrators and developers to be able tackle the concepts of Kubernetes without having to go through long periods of training and hands-on experience in order to effectively manage Kubernetes clusters and applications. I hope as well in v2.7 they bring back the dual-UI experience.

[stefb69]

I dislike the new UI and everyone at my team agrees with that. We have decided not to upgrade production rancher servers to 2.6 at all. But I found a way to have the old ui displayed in version 2.6.1 of rancher: just remove the dashboard and explorer words from the url and you will see the old ui, somewhat modified though

:
Example:
https://rancher/dashboard/c/c-xxaa/explorer becomes https://rancher/c/c-xxaa

[adaliszk]

What a nice workaround!
I just hope that the code there won’t suddenly be removed with the next patch.