feat(monitoring): disk usage history by category, with image and build cache cleanup

[agentHits]

What changes

Disk Space card on the Monitoring page. It now answers two questions the old card could not: what takes the disk, and what filled it over a period.

• The used space is split into images, build cache, volumes, container logs, container files, Dokploy data, swap files and "System and other". The rows add up to the used space.
• A stacked bar chart covers 1, 3, 7 or 30 days. Each bar is the fullest moment of its interval, so a short spike stays visible.
• A table gives the change of every category over the period, and a list names the images, volumes and containers that grew or shrank the most.
• The "Free up space" block from the update dialog is also in this card. It lists the old Dokploy builds it removes and can clear the build cache.

Background sampling. The history is collected by the panel process, not by an open page:

What	How often
Used space (statfs)	every minute
Images, container files, build cache (Engine API)	every 5 minutes
Volumes, container logs, Dokploy data, swap files	every hour
All of the above	at once when the used space moves by 0.5% of the disk (at least 256 MiB)

A probe never takes more than 2% of the time: after a slow one the sampler waits fifty times its duration. Samples are kept for 30 days as JSON lines in monitoring/disk-history/.

Why the two cards disagreed

• The old card counted in GiB and labelled the result GB; Docker prints decimal GB. The new card uses decimal units everywhere.
• Docker's total does not include swap files, the operating system, system logs, container logs or Docker's own metadata. They are separate rows now.
• Docker's build cache total counts the records that images share a second time. The new card leaves them out of the build cache row.

Notes for review

• Docker is asked through the Engine API (GET /system/df with a type filter), so no docker CLI process delays the cleanup commands that wait for Docker to be idle.
• Container logs are under Docker's root directory, which the panel's container does not see. A throwaway busybox container with that directory mounted read-only measures them once an hour. If it cannot run, the logs stay in "System and other".
• settings.getDiskUsageHistory is an admin procedure, classified "read".
• UpdateDiskSpace gained oldImages; freeUpdateDiskSpace samples the disk after a cleanup so the card shows the freed space at once.
• The Docker disk usage pie takes the colours of the same categories in the new card. The palette was checked with a colour-blind-safety validator on the light and dark surfaces.
• The live disk chart docker-disk-chart.tsx is removed; the new card replaces it.

Testing

• New unit tests: probes (disk-usage.test.ts), sampler, storage and query (disk-history.test.ts), formatting and axes (disk-usage-format.test.ts).
• bun run typecheck passes for all workspaces.
• Full suite locally: 3480 passed, 2 failed. Both failures are unrelated to this change: application.real.test.ts needs swarm and timed out, and acme-delete-docker.test.ts failed once under that load and passes when run alone.
• The probes and the history were run against a real Docker 29.8 daemon (containerd image store), and the /system/df answer was checked on a Docker 29.4 daemon with the overlay2 driver.
• The card was rendered with fixture data in the dark and light themes and at 390 px width.

Not tested: the card inside a running panel with real history. That needs a deployed build.

🤖 Generated with Claude Code