KEDA

Every KEDA ScaledObject in plain words, HPAs that fight them, scale-to-zero workloads, unused TriggerAuthentications, and a wizard for new ScaledObjects.

KEDA in the sidebar’s Tools lists the cluster’s KEDA ScaledObject objects. If KEDA isn’t installed, the page says so.

ScaledObjects

Each one is explained in plain words, such as scales 1→20 on cron Mon–Fri 09:00, with its target workload linked. A trigger type Kubebay doesn’t know is shown by its raw name rather than dropped.

Each ScaledObject shows a Paused or Active pill. KEDA marks a paused ScaledObject with the autoscaling.keda.sh/paused: "true" annotation, or with autoscaling.keda.sh/paused-replicas, which pins the workload to that replica count. Kubebay also treats a Paused status condition of True as paused.

To pause a ScaledObject, choose Pause on its row. Kubebay asks Replicas to hold, which starts at the current replica count and can be 0, then sets autoscaling.keda.sh/paused-replicas to that number (and removes autoscaling.keda.sh/paused, which would otherwise hold the current count instead). KEDA then stops autoscaling and keeps the workload at that count. Unpause removes both annotations so KEDA resumes scaling. Both changes are audit-logged and use your own permissions.

Above the list, Search KEDA resources… filters by kind, name or namespace, and a row of buttons picks which kind to show:

  • All: ScaledObjects, ScaledJobs, TriggerAuthentications and ClusterTriggerAuthentications together, each linked to its detail view;
  • ScaledObjects (the default), ScaledJobs, TriggerAuthentications and ClusterTriggerAuthentications: one kind each.

In the ScaledObjects view, filter further with the buttons above the list:

  • All;
  • Conflicts: ScaledObjects whose workload also has its own HorizontalPodAutoscaler. The two fight over the replica count, and this is the most common way KEDA setups go wrong;
  • scale-to-zero: ScaledObjects with minReplicaCount: 0, whose workload can go idle between triggers.

Orphaned TriggerAuthentications

TriggerAuthentication objects that no ScaledObject references, listed under Orphaned trigger authentications. They’re usually left over from a ScaledObject that was removed.

A workload’s Autoscaling tab

A Deployment’s or StatefulSet’s drawer has an Autoscaling tab. It shows every HPA, VPA and ScaledObject that targets the workload, with min, max and current replicas. A KEDA-managed workload is marked Managed by KEDA.

New ScaledObject opens a wizard:

  1. Name it and set Min replicas and Max replicas.
  2. Pick a Trigger:
    • CPU utilization or Memory utilization: set a target percentage;
    • Cron schedule: a start and end in cron syntax (such as 0 9 * * 1-5 and 0 18 * * 1-5), a timezone, and the replicas to run in that window;
    • Prometheus query: the in-cluster server address, a PromQL query and a threshold. Kubebay suggests Prometheus Services it finds in the cluster.
  3. Review the YAML, Dry-run it, then Apply.