Skip to main content
For a target in autonomous mode, the Autonomous scaling history panel on the target’s page lists every change Thoras made to the workload: each pod it resized or evicted, each restart and replica change, and each new pod it sized. Every entry shows the result of the request and why Thoras made it. An accepted request does not mean the pod has finished resizing or the rollout has completed. Today’s activity refreshes every 15 seconds while the page is visible; use the date selector to look back up to 30 days. New pods are discovered once per minute, so their entries can take roughly 75 seconds plus execution time to appear. For a target in recommendation mode, the panel shows forecast history in the same timeline and table layout. Entries show current and suggested values and explain why changes were not applied. They describe suggestions, not pod changes.

​
What’s recorded

Each change has an outcome:
  • Accepted: Kubernetes accepted the request. Completion depends on the workload and cluster state.
  • Failed: Kubernetes could not apply the change. When an admission policy prevents it, the entry identifies that cause without exposing internal policy names, expressions or cluster diagnostics.
  • Blocked: a PodDisruptionBudget refused an eviction, or a rollout was paused. Thoras retries a blocked eviction on its next pass, and each attempt is listed.
Resizes and evictions from the same disruption cycle appear as one entry with a pod count. Accepted, blocked and failed requests remain separate. Expand a group to browse its individual pod requests and resource changes. Expand a change to see why it was made (a new forecast, resources behind the latest forecast, a stalled in-place resize, an out-of-memory kill, or a mode or bounds change), the previous and requested resources per container, and the forecast reasoning when a retained forecast is linked to the request. Creation-time sizes appear only when captured by the webhook; they are not inferred from later pod sizes.

​
What isn’t recorded

  • Changes made while scaling is paused: Thoras makes none, so none are listed.
  • Replica counts Thoras adjusts as an HPA scales the workload, and changes made by a DaemonSetAutoscaler.
  • Pods removed before the discovery job can observe them, and admissions older than the job’s 10-minute lookback.
Scaling activity is shown in the in-cluster dashboard only, not in the Thoras console.