docker: insufficient disk space

Docker disk space errors in CI are dramatically worsened by layer cache accumulation on self-hosted runners. Managed runners (GitHub-hosted, CircleCI cloud) are ephemeral — they start clean and are discarded after each job. Self-hosted runners persist between jobs, accumulating Docker images, volumes, and build caches that are never automatically cleaned. A self-hosted runner active for 3–6 months can accumulate 50–100 GB of Docker data: old base images, intermediate build layers, dangling images from failed builds, and BuildKit's build cache. docker system df shows exactly how much space each category uses, broken down by images, containers, volumes, and build cache. The recommended maintenance approach: a weekly docker system prune -af --volumes cron job that removes all unused images, stopped containers, and unused volumes. Note that -a removes all unused images including ones that may be needed for upcoming builds — the first build after cleanup is slower as images are re-pulled from registries. A more targeted approach: docker image prune -f --filter "until=168h" removes images not used in the past 7 days, preserving recently used ones. BuildKit cache clears separately with docker builder prune --filter "until=72h". GitHub Actions supports native Docker layer caching through docker/build-push-action's cache-to: type=gha parameter, which stores build layers in GitHub's artifact cache and restores them between runs, eliminating the self-hosted accumulation problem.