The suite allocated temp dirs at 68 sites across 21 files and removed none,
accumulating ~6700 directories and 189MB per machine-day, some holding age
keys. All 68 now go through a single `tmp()` helper allocating inside a
per-run root that vitest's globalSetup teardown removes wholesale, and a
class-guard test fails if `mkdtempSync` appears under test/ outside the
helpers.
The per-worker `process.once("exit")` reaper that suggests itself here does
not work under vitest and fails silently: the pool recycles workers by
killing them, so exit handlers registered in a test file never run. Measured
— a probe test writing from an exit hook produced no file, and a full run
with per-worker hooks still left 750 directories. globalSetup's teardown runs
in the main process, after every worker, and vitest awaits it.
Separately, and contrary to #117's framing that "cast itself does not leak":
resolveCheckout() mkdtemps an `infra-checkout-` dir, clones the infra repo
into it, and never removes it, so every `cast apply`/`diff`/`capture` without
--path leaked a full clone. The box that reported #117 was holding 602 such
directories, 73MB of real .git trees, from the same day. The leak fires on
the failure path too, since the dir is created before the clone runs.
Ephemeral checkouts are now reaped on process exit — the lifetime that fits,
since callers read the tree after resolveCheckout returns; a --path checkout
is the operator's own tree and is never registered.
Empirical: /tmp/cast-* + /tmp/infra-* count is 0 before and 0 after a full
`npm test`, against 750 with the exit-hook design. 626 tests green.
Refs #117
The greenfield manifest-first bootstrap was a chicken-and-egg with no
exit, found by the 2026-07-19 release drill: fresh Coolify instance,
registered project, a manifest declaring databases only and resolving
zero ${…} refs. apply refused with "no secret store", and capture — the
documented way to get a store — rightly refused a project absent on the
box, because apply is the verb that would create it. The drill unblocked
with a hand-rolled empty age store, documented nowhere.
Now diff/apply gate the refusal on the manifest actually referencing a
secret, asked via requiredSecrets — the same parser resolution uses, so
the two cannot disagree. Zero refs: an absent store is treated as empty,
a loud one-line note names the path it would live at, and the age key is
not demanded (nothing to decrypt, nothing to protect yet). One ref: the
refusal returns byte-identical to before. capture and destroy are
untouched.
Fixes#104
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every cast verb was single-project, so "do this to the whole instance" was a
shell loop the operator wrote from memory — and the project they forgot is the
one that drifted. `cast diff --env prod --all` and `cast apply --env prod --all`
iterate the registry (#25) instead.
The bulk of this is a refactor: the apply/diff block in cli.ts was one long
inline body, and it is now `runProject` — checkout → secrets → desired →
bindings → live → diff → optionally apply. Both the single-repo path and the
`--all` loop call it, so there is exactly ONE implementation of what a project
run is. A second, parallel fleet path is how the two would drift, and drift is
the subject of this tool. `openCoolify` and the team assert are hoisted out of
it: one --env means one instance and one team, so asserting once still lands
strictly before the FIRST project's first read — the read is already the lie.
Fails closed on the aggregate. A registered project cast cannot reach is an
ERROR, never a skip: the clone failing, no manifest block for this environment,
an absent or undecryptable store, an absent Coolify project/environment, any
HTTP error. A silently skipped project reads exactly like a clean one — #12/#18/
#22 at fleet scale — so the report leads with COVERAGE (registered / read /
clean / drifted / unreachable), and:
diff --all 0 every registered project was READ, and every one is clean
1 every one was read, and at least one has drift
2 a project could not be read — outranking drift, because an
unreadable project is not a diff result but the absence of one
apply --all 0 every registered project applied; non-zero otherwise
`diff --all` runs every project to completion (stopping hides the drift in the
projects it never reached); `apply --all` STOPS at the first failure and names
what it applied and what it did not touch (continuing to mutate a fleet after an
unexplained failure is not a thing cast gets to do).
Two refusals. An empty or absent registry refuses rather than printing
"0 projects, clean" — an empty fleet reading as a clean fleet is the whole
failure this is against; the message distinguishes an unmigrated state file from
a registry pointed elsewhere and prints the YAML to write. And `--all` is
mutually exclusive with the repo positional and with every single-project
coordinate (--path, --project, --environment, --resource, --hostname-overlay):
each names ONE project's checkout, ONE project's Coolify name, ONE box's
resource names, and `--project X` across a fleet would point every project at
the same Coolify project — a false report on diff, and on apply every manifest
in the fleet written into one project.
Also: `projectsIn`'s doc-comment guessed that `[]` made a fleet verb over an
unmigrated state file "a clean no-op rather than a crash". It is precisely
backwards, and now says so. And the --path/--env-prod refusal is hoisted to the
CLI's up-front flag validation (one rule, one string, two call sites in
resolve.ts) — it used to be caught only by accident of resolveCheckout running
before the bindings load.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
