Where the space goes, and how to reclaim the disposable half of it safely.
upkeep statussummarizes the cockpit, environments, and total tracked disk usage;
upkeep status --disk itemizes it per module, core version, and category
(project trees, docker volumes, materialized snapshots, base artifacts,
fixture dumps) with totals. The header line reports how many modules are
registered:
Cockpit: /home/you/upkeep-cockpit (2 registered module(s))
To count them, status — like base-artifacts:status and
base-artifacts:build — parses registry.yml rather than just checking that
it exists, so a malformed registry now fails these commands with exit 2 too,
instead of being discovered later by a command that reads it.
upkeep prune --allis a dry run — it lists deletion candidates (environment trees, docker
volumes, materialized fixture snapshots) with reclaimable sizes and deletes
nothing until you add --yes. Narrow it with --trees, --snapshots, or
--projects, age-gate with --older-than=30d, and keep the newest N
snapshots per project with --keep-latest=N. Everything pruned regenerates
on demand.
--older-than measures an environment's age from when it was last used,
not when it was created: each environment's .upkeep-env.yml carries a
last_used_at timestamp, stamped at provision and re-stamped every time the
environment is reused. An environment created months ago but checked against
this morning is not a deletion candidate. Environments provisioned before this
field existed parse fine — the key is optional — and fall back to their
creation date until the next time they are used.
Protected regardless of flags: base artifacts, committed fixture dumps
(.sql.gz), and keep-marked items — touch <project>/.keep keeps a whole
environment, and an <artifact>.keep sibling (e.g.
materialized/<name>.sql.keep) keeps a single snapshot. Environments are
always disposed through the engine adapter, so containers and named volumes
are released together with the tree.