Skip to main content

User guide: kaironctl get --selector

kaironctl delete RESOURCE --selector k=v (see kaironctl-bulk-delete.md) added label-selector matching to the write side of kaironctl, but the read side was left behind: kaironctl get had no way to narrow its listing by label at all. Previewing exactly which Machines a selector reaches -- before trusting it to a bulk delete, or just to find every Machine one load test created -- meant piping kaironctl get's full output through grep, or reaching for kubectl get -l instead. This adds --selector to kaironctl get too, as the read-side counterpart bulk delete was always missing.

Usage​

kaironctl get [RESOURCE] [--selector k=v] [--selector k2=v2 ...] [-n NAMESPACE]

RESOURCE keeps its existing default of machine when omitted (unlike delete RESOURCE --selector, which requires an explicit kind -- see that guide's own reasoning for why bulk deletion is the one place a wrong default would be dangerous; a mere listing carries no such risk). --selector is optional and repeatable; every key=value pair given must match (logical AND, never OR), using the exact same model.LabelsMatch rule as delete --selector, MachineDisruptionBudget.spec.selector, and MachineNetworkPolicy.spec.selector. Omitting --selector entirely lists every resource of that kind, exactly as kaironctl get has always behaved.

$ kaironctl get machine
NAME         NODE     PHASE     CPU  MEMORY  IP
loadtest-1   node-a   Running   2    4Gi     10.0.1.5
loadtest-2   node-b   Running   2    4Gi     10.0.1.6
web-1        node-a   Running   4    8Gi     10.0.1.7

$ kaironctl get machine --selector purpose=loadtest
NAME         NODE     PHASE     CPU  MEMORY  IP
loadtest-1   node-a   Running   2    4Gi     10.0.1.5
loadtest-2   node-b   Running   2    4Gi     10.0.1.6

$ kaironctl get machine --selector purpose=loadtest --selector env=staging
NAME  NODE  PHASE  CPU  MEMORY  IP

That last example matches nothing: get, unlike delete, prints just its usual header row for an empty result (the same thing it has always printed for a namespace with no resources of that kind at all) rather than growing a special-cased "nothing matched" message of its own -- an empty list reads the same way here regardless of why it's empty.

Every kind kaironctl get already lists gets this for free: machine, migration, snapshot, restore, quota, budget, machineset, instancetype, migrationpolicy, snapshotschedule, networkpolicy, and securitygroup (same aliases as always) -- one generic helper (selectorFilter in internal/kaironctl/kaironctl.go), applied once per kind's existing list-and-print case, not a special case bolted onto one kind.

Why this is a generic helper, not one filter function per kind​

kaironctl get's dozen kinds each already have their own List*Namespace call and their own per-kind print loop (see the file's own long switch in cmdGet); the only thing all twelve share is that every item's labels live at .Metadata.Labels. Go's generics can't express "any struct with a Metadata field" structurally without reflection, so selectorFilter[T any](items []T, selector map[string]string, labels func(T) map[string]string) []T takes a small accessor closure instead -- one line added per case (items = selectorFilter(items, selector, func(m model.Machine) map[string]string { return m.Metadata.Labels })), reusing the exact same model.LabelsMatch bulk delete's matchingNames already calls, rather than either thirteen near-identical filter functions or a reflection-based one.

Real limits today (first cut)​

  • Selector matching, not a general query language. Same limit delete --selector already has: model.LabelsMatch's plain equality-per-key semantics only, no set-based in/notin, no !=.
  • No -o wide/-o json output mode. --selector only narrows which rows print, in the exact same tab-separated columns kaironctl get already printed for that kind; it does not add a new output format.
  • No new RBAC. This reuses exactly the same per-kind list calls kaironctl get already needed with no selector at all -- scripts/check_rbac_coverage.py was re-run to confirm nothing changed.