oxygen914/kubeoncall/disk-capacity-triage

Disk Capacity Triage

Diagnostic objective

Distinguish byte capacity, inode exhaustion, read-only remount, persistent-volume, and growth-rate failures on the affected node.

Required live evidence

Confirm mount point, filesystem type, byte and inode usage, available capacity, growth rate, read-only state, affected workloads, and recent image or log growth. Exclude pseudo filesystems and compare against another node.

Decision branches

• Bytes high but inodes normal: identify the growing path or volume owner without deleting files.
• Inodes high but bytes normal: inspect small-file creation and rotation behavior.
• Filesystem read-only: collect storage and kernel evidence and treat cleanup as unsafe.
• Pressure is isolated to a persistent volume: hand off to the storage owner with volume identity and workload impact.

Safe next step

Use the metric-specific runbook and document the mount, owner, growth rate, and remaining time before proposing capacity or cleanup work.

Prohibited actions

Do not delete files, prune images, evict Pods, remount filesystems, or resize volumes from this Skill.

Recovery and escalation

Require usage and inode trends below policy thresholds, writable state, and stable workloads across two windows. Escalate for read-only remount, critical-volume exhaustion, or rapidly shrinking headroom.