Flux's add-on: everything Flux runs in one list, what's failing first, and a tab for each of its kinds.
In the sidebar
An add-on shows up once the cluster serves one of its tool’s kinds. Lumovi looks at what the cluster serves again every minute, so a tool you just installed appears on its own, and one you removed goes away.- Under Add-ons. Most add-ons sit in a section of their own, by name, between Pinned and Custom resources.
- With Kubernetes’ own kinds. Add-ons that extend what Kubernetes does sit at the end of that section: Cilium, Gateway API, Linkerd and Traefik under Network, Sealed Secrets under Configuration, and Longhorn, Rook Ceph and Volume Snapshots under Storage.
certificates offers cert-manager. (⌘ is Ctrl on Windows and Linux.)
The add-on’s page
Tabs
All comes first, then a tab for each of the tool’s kinds that the cluster serves, in the add-on’s order, with how many objects it has. A kind’s tab is its usual list, with its view’s columns and actions, and the add-on’s tabs stay above it. When there are more tabs than fit, they scroll sideways. Counts and lists follow the namespace you picked. Kinds without namespaces are always the whole cluster’s, and when all of a tool’s kinds are, like Karpenter’s, the namespace picker says Cluster-wide.Everything in one list
All lists every object of the tool’s kinds:
The list sorts by status, so what’s failing comes first: Failing, then Warning, In progress, Healthy and Inactive. Click any column to sort by it instead.
- Health. The chips above the list (Failing, Warning, In progress, Healthy, Inactive) show only objects in that state, with how many there are. Pick several to combine them.
- Filter. Press / and type. Filter Flux (named after the tool) matches names, namespaces, kinds, labels written as
key=value, and what the status says. - Label selector. Type a selector like
app=weband press ↵. The cluster filters every kind’s list with it.
When something’s missing
- Nothing yet. With no objects, the page says Nothing from Flux in shop (or in this cluster), and which kinds would show up here.
- A kind it couldn’t list. When you can’t list one of the kinds, a line above the list says Couldn’t list, which kind, and why. The other kinds still show.
- A long list. Lists that are too long to load whole are cut short, and the page says some objects may be missing. Choose a namespace, or use Filter by label.
- A tool the cluster doesn’t have. An add-on’s page opened from a link, in a cluster without the tool, says This cluster doesn’t have it: none of its kinds are served there.
Karpenter
Karpenter’s add-on opens on an overview of its own instead of a list: its node pools against their limits, the nodes they launched and the ones launching, the mix of instance types, capacity types and zones, what’s being disrupted, and the pods waiting for a node. Its first tab is Overview, and its kinds’ tabs (NodePools, NodeClaims, and the node classes) follow.
Karpenter's node pools against their limits, the nodes they launched, and what's being replaced or waiting for a node.
karpenter.sh/nodepool.
The four numbers
Each tile opens the list behind it. Node pools ready and Nodes ready open filtered to what isn’t ready when something isn’t, and Nodes ready opens Nodes with the label selector
karpenter.sh/nodepool. Launching opens the node claims, filtered to the ones in progress while some are launching.
Node pools
Each node pool, by name, with its status, how many nodes it has, and how much of its CPU and memory limits its nodes take, like 24 of 64 cores. A node pool without a limit says no limit, with a dashed track. Under its name, a node pool says what it launches: its capacity types and instance families, like Spot, on-demand · m6i, c6i. These come from its requirements onkarpenter.sh/capacity-type (on-demand when it has none), and on karpenter.k8s.aws/instance-family, karpenter.k8s.aws/instance-category, karpenter.azure.com/sku-family or node.kubernetes.io/instance-type, the first it has. A node pool that isn’t ready shows why instead.
Nodes
Karpenter’s nodes, grouped by node pool, each with its instance type, capacity type and zone. With metrics-server, each has meters for its CPU and memory use; without it, its status and how many cores it has. Node claims still on their way are listed with their node pool, as Launching, with how long ago they started.Node mix
How many nodes there are of each instance type, capacity type and zone, the most common first.Disruption
The node claims Karpenter is replacing, or could do without, with how long they’ve been that way:
When there are none, it says Nothing is being disrupted.
Waiting for a node
The pods the scheduler couldn’t place anywhere, which Karpenter launches nodes for, with the scheduler’s message. When there are none: No pods are waiting for a node. Nodes (for each node pool), Disruption and Waiting for a node show six rows, with Show all for the rest. Click anything to open it. The overview reads node pools, node claims and nodes, and pods in every namespace, so it needslist on all four. A card whose list you can’t read says why, and the others still show.
Your own add-ons
An add-on is a short YAML document of kindAddOn, kept with your views, in ~/.lumovi/views:
platform.yaml
kindsare the tabs, in order.kind: '*'with agroupis every kind of that group, each with a tab.iconis one of the view icons.categoryputs it at the end of one of Kubernetes’ sections instead:cluster,network,configorstorage.- Kinds Lumovi has a page of its own for, like Pods, Deployments or Secrets, can’t be in an add-on.
Replacing one of Lumovi’s
Give yours the samemetadata.name as Lumovi’s, like flux or cert-manager (Built-in views gives each one’s name). Yours replaces it whole: its label, icon, place and kinds. The kinds’ views stay as they are.
In your cluster: add-ons everyone sees go in the chart’s
views value, with the views. People can’t add their own from a browser. See Views for everyone.Built-in views
Every add-on Lumovi ships, and what its views add.
View format
Every field an add-on can have.