Kubernetes Dashboard Design Sprint (Łódź, Poland, 23 Mar 2016)

People

@bryk, @cheld, @digitalfishpond, @floreks, @maciaszczykm, @olekzabl, @zreigz

Goals

Sketch the directions of the development of the UI to achieve:

• scalability for the variety of types of K8s objects we want to handle (currently: replication controllers & pods; desired: replica sets, daemon sets, container, nodes, volumes, secrets, deployments etc.)
• scalability for the number of such objects (currently: cards are suitable for <20 objects; desired: show >100 pods in a nice way)
• good UX for various use cases (monitor, debug, deploy, explore)
• more aware about namespaces

Agenda

• agree on the meeting’s goals
• split into 3 groups - propose a general UX vision - discuss them together, unify
• split again - sketch the system in more detail - discuss, unify
• split once more - sketch a chosen page, in yet more detail - discuss
• decide when and how to publish the results

Decisions

General navigation

• constant, basic-purpose navigation menu on the left, containing:

  • items describing important classes (coarser than types) of objects e.g. “apps”, “services”, “config”
  • under them, sub-items describing types of objects
  • possibly, other special buttons, e.g. “resource explorer”
  • possibly, minimizable

• context-dependent action bar on the top (below the main “K8s” bar)

  • buttons for basic actions, e.g. “create”, “edit”, “delete”

• namespace switcher on the main “K8s” bar

  • single namespace / all namespaces / multi-choice / filter
  • shall it be visible also on the object details page?

Cards

• will be generally replaced by table rows
• possibly thicker to contain few more data items and/or some graphs
• much more clear sorting; sort triggered by clicking on column name
• easier pagination

Home page

• does no more list all objects (or RCs) in the cluster
• instead, it shows:
  • important messages (errors/warnings/suggestions)
  • a few selected objects (last visited/most visited/most important)

Details page of an object

• general requirements:

  • has possibly (but reasonably) unique design across various object types
  • allows convenient navigation to all directly related objects; by this we mean e.g. that a pod shows links to all volumes it uses, and conversely, a volume shows links to all pods using it
  • contains a number of sections (described below); some of them may be moved to other tabs if everything does not fit in one page (e.g. we will need several tabs for RCs, but not for secrets)

• section (main tab): a brief description of the object status

  • for a simple object, this contains all its properties
  • for a more complex one, this contains most important properties, while all properties are gathered in another tab of the page

• section (main tab): visual summaries for all directly related objects (aggregated by type)

  • example: for a RC, these could be activity graphs of all involved pods, volumes, services etc.

• section (main tab, bottom): navigation links to directly related objects

  • grouped by type
  • each type is headed by a user-friendly description of type, and of the relation to the current object (e.g. 'volumes mounted to this pod', 'pods mounting this volume', etc.)
  • limited to e.g. 10 top items, to make them all fit in one page (this has the form of a table, so clicking column names allows adjusting the meaning of “top”)
  • if there are more related objects of some type, we provide a link to the list of all of them (see below)

• other sections (main or additional tab): history, logs, (more?)

Summary page for objects of type X

• can be:
  • global (with namespace restriction applied), reachable by left menu panel
  • scoped (all objects of type X related to object O), reachable from O details page
• similar visual design to the object details page
• upper section: general metrics of involved objects
• lower section: list of all involved objects, tabularized

Supplementary ideas, questions etc.

• resource explorer

  • goal: give a larger-scale insight into cluster structure (beyond “directly related” links)
  • shall easily support filtering
  • two considered views:
    • “filebrowser view”: very compact but difficult to model all relations in a cluster (a volume may have many pods mounting it as “parents”; network communication between pods is not at all a “parent-child-sibling” relation)
    • “graph visualization”
  • in any view, exploring may be enhanced by showing an additional details preview side-by-side

• error messaging policy On the details page of a RC, do we want to show that 3 out of 100 its volumes are broken? Is this important enough? Generally, what is sufficiently important? How do we decide on this?

Images