Public-health build cache fills disk and civic nodes run out of space. What Docker remediation fits?

Public-health build cache filling civic disks is remediated by pruning Docker build cache and unused images to restore node capacity. BuildKit/cache layers and stale intermediate images accumulate during iterative model and container builds; pruning reclaims space while leaving running containers and tagged production images intact when done with appropriate filters. For municipal labs that iterate weekly on epidemiology containers, disciplined prune jobs are part of capacity hygiene—not an emergency format.

Formatting the OS disk without backing up cluster state fails by concept: it destroys node identity, credentials, and possibly local scratch needed for BCM/kubelet, turning a reclaimable cache problem into a rebuild. Disabling disk monitoring so full volumes look healthy fails observability: it hides the failure mode while pods still hit ENOSPC. Moving the entire BCM head-node database onto every worker fails architecture and multiplies risk; workers are not HA copies of the head-node monitoring store, and the copy would worsen disk pressure.

Exam caveat: prune carefully—blind docker system prune -a on a shared civic builder can delete images still needed for air-gapped restarts. Prefer timed cache prune, dangling filter strategies, and retention of pinned digests listed in the public-health image allowlist. Disk-full symptoms can also come from logs or datasets; the stem names build cache, so address that first.

Operational check: measure disk use before/after (docker system df), prune build cache and unused images per policy, confirm civic nodes regain headroom, re-enable or verify monitoring thresholds, and schedule recurring prune windows so the next training sprint does not page the NOC at 2 a.m.