The canonical is last-GREEN, not CURRENT. Its promotion (WC5) deploys a
warm-<recipe> app and health-checks it; when that fails the canonical stays
pinned at an old release indefinitely — and the failure is invisible to the
upgrade tier, which keeps basing on it.
Live evidence (2026-08-10): gitea's canonical was 3.5.3+1.24.2-rootless from
2026-06-17 because warm-gitea crash-loops on a read-only app.ini (open fix:
recipe PR #4). Four other recipes are in the same state — the 2026-08-09 sweep
logged GREEN-BUT-PROMOTE-FAILED for bluesky-pds, drone, gitea, lasuite-drive and
lasuite-meet.
Consequence: the upgrade tier tested 3.5.3 -> head, a transition no deployment
performs, and MISSED the one they do. 3.5.3 -> head crosses an APP_INI_VERSION
change (v21 -> v22) so Swarm creates a fresh config object; the real
3.6.1 -> 3.6.2 upgrade keeps v22 and aborts on Swarm's immutable-config rule
('only updates to Labels are allowed'). That shipped in 3.6.2 and broke every
existing install taking the 1.27.1 security release.
Now: when the newest published release older than head is NEWER than the
canonical, use the release and say so in the log (naming the stale canonical so
the promote failure gets noticed). Resolution stays dynamic; UPGRADE_BASE_FLOOR
still filters structurally-invalid candidates; the canonical remains the base
whenever it is current.
Unit-verified on the real gitea tag set: base moves 3.5.3+1.24.2-rootless ->
3.6.1+1.26.2-rootless, exactly the transition that reproduces the break.
The strict recipe_meta key registry rejected the new declaration (caught by
verify run 1). str-typed, default None; full semantics documented on the key.
The 0.8.x->1.0.0 discourse recipe family switched app bitnami->official AND db
pgvector/pg17->discourse/postgres:pg18. That db-family change is a structural
break (bitnami cluster lacks the discourse role; pg_upgrade preserves-not-creates
roles) with NO supported in-place path. The dynamic base resolver's step-back
kept selecting 0.8.1+3.5.0 (newest tag below the unbumped 1.0.0+3.5.3 label) and
the upgrade tier red'd twice on this unsupported path (drone #1165, #1171 —
classified stale-test both times; recipe verified green on the real
official->official path).
Adds UPGRADE_BASE_FLOOR (phase basefloor) to resolve_upgrade_base: a recipe_meta
declaration naming the first post-break published version. Resolution stays
fully dynamic (this is NOT the removed sec2.G static pin): the floor only
EXCLUDES below-floor candidates (canonical, step-back, no-canonical fallback);
when no >=floor predecessor exists the tier records a DECLARED skip, never a
silent pass. main-tip fallback unaffected (post-break by construction).
tests/discourse/recipe_meta.py declares UPGRADE_BASE_FLOOR = 1.0.0+3.5.3 with
the full rationale. Unit-verified: head=1.0.0+3.5.3 -> declared skip (was:
0.8.1+3.5.0 wrong pick); post-release head=2026.x -> base 1.0.0+3.5.3 (the real
migration path). No assertion weakened - below-floor in-place upgrades were
never supported coverage.
A once-failed promote left swarm secrets (e.g. drone's gitea client_secret_v1) behind; the retry's
install_steps 'abra app secret insert' then FATAd 'already exists', so a recipe could never recover
its canonical. promote_canonical now teardown_app()s the warm domain when there is NO existing
canonical (fresh seed) — clearing leftover secrets/.env/partial volumes — while a re-promote
(canonical exists) still reattaches its retained known-good volume untouched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
drone (DEPS=[gitea], a COLD dep) deadlocked in promote: the cold test holds the gitea dep's
app-lock for the whole process lifetime, and promote's _provision_deps re-acquires the same lock
in the same process → blocks forever. By promote time the cold test + its deps are torn down
(dep teardown runs in the run finally, before promote), so the locks are stale. New
lifecycle.release_app_locks() frees them at promote start; the serial sweep guarantees no
concurrent run relies on them. lasuite-* (warm keycloak dep) were unaffected (no cold deploy).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
M2 finding (Adversary-flagged): promote_canonical did a bare `abra app deploy` that lacked the
cold install's wiring, so recipes that passed the cold test still failed to promote:
- ghost: `abra app new` FATA 'locally unstaged changes' — the CCCI_SKIP_FETCH per-run tree was
left dirty by the tier suite. Fix: force re-checkout the tag + `git clean -fd` before deploy.
- bluesky-pds: missing pds_plc_rotation_key (install_steps inserts it, #generate=false).
- custom-html-tiny: 404 (install_steps seeds index.html). Fix: run install_steps_hook in promote.
- OIDC recipes would miss their realm. Fix: provision DEPS in promote like the cold install.
promote_canonical now: clean tree → provision deps → deploy_app with install_steps_hook + overlay +
ready-probes, then snapshot. Also: sweep result label now derives from whether the canonical was
actually written (promote is non-fatal; rc==0 did not imply promoted) — fixes the misleading
'PASS (promoted)'.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Closes the head_version-vs-latest_version divergence: should_promote gates on head_version
(code under test) but promote_canonical recorded latest_version(recipe_tags). In a manual
RECIPE=<r> run whose main checkout sits on a tag OLDER than the newest published tag, the gate
would pass on the older tag yet promote the newer (never-tested) one. promote_canonical now
takes the tested `version` (head_version, guaranteed a release tag by the tagged-gate) and
records exactly that. Sweep path unaffected (head==tag by construction).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- should_promote_canonical gains a `tagged` requirement (canon §2.A): a green cold
latest run promotes only when the tested head version is a published release tag;
an untagged main commit never becomes a canonical.
- warm_reconcile.is_released_version(recipe, version): release-tag membership (exact or
by version_key). Caller computes `tagged` so the gate stays pure.
- unit tests: untagged -> no promote; is_released_version cases.
- drive-by (pre-existing reds, unrelated to canon, now green): test_warm_reconcile
traefik assertion was stale vs the phase-pxgate spec (probes /api/version, no
health_domain); meta.py UPGRADE_BASE_VERSION KEYS help synced to the prevb doc text.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
resolve_upgrade_base now reads the head's published version (abra.head_compose_version,
the coop-cloud.<stack>.version label) and, when the last-green warm-canonical version
equals it, steps back to the newest published version strictly older than head instead
of deploying a same-version no-op. warm_reconcile gains version_key + newest_older_version
(single coop-cloud ordering source; sort_versions refactored onto version_key, no behavior
change). Skip only when no older published predecessor exists. Step-back returns kind=version
so it inherits F1d-2 pinned-tag checkout. Extends tests/unit/test_upgrade_base.py (13 pass).
docker stack deploy doesn't prune services the head compose dropped (discourse PR#4 drops sidekiq),
leaving them orphaned on the base image. perform_upgrade now reconciles the live stack to the head
compose service set (lifecycle.prune_orphan_services). Makes the deployed stack faithfully reflect
the head — no test weakened. No-op when service sets match / compose unresolvable.
Blocker 4 fix: abra `secret generate --all` uses .env.sample for length hints; the
lfs-plain-gitea PR has SECRET_LFS_JWT_SECRET_VERSION=v1 COMMENTED OUT, so abra produces
a wrong-length secret. gitea requires exactly 43 chars (32 bytes base64 URL-safe); wrong
length → gitea fatals trying to save the JWT secret to the read-only Docker Config
app.ini → health check fails → swarm rolls back.
Fix: new UPGRADE_SECRET_PREP hook (meta.py) called before `abra secret generate --all`
in the upgrade path. abra's `--all` is idempotent (skips existing secrets), so the
correctly pre-inserted secret survives. gitea's recipe_meta.py implements the hook using
`docker secret create` directly to guarantee correct format regardless of .env.sample.
Also consumes machine-docs/BUILDER-INBOX.md (Adversary Blocker 4 digest).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Blocker 1 (LFS roundtrip fails on PR #1):
- Add UPGRADE_EXTRA_ENV to gitea recipe_meta.py — after PR-head checkout
(compose.lfs.yml now in ABRA_DIR), add compose.lfs.yml to COMPOSE_FILE
and set SECRET_LFS_JWT_SECRET_VERSION=v1 so the upgrade chaos redeploy
actually runs with LFS enabled. Without this, the base install checks out
the 3.5.x tag (compose.lfs.yml removed), EXTRA_ENV sees no LFS, and the
upgrade chaos redeploy inherits the no-LFS .env — so the LFS test runs
(compose.lfs.yml is restored by recipe_checkout_ref) but LFS is off.
- Add abra.secret_generate(domain) in generic.perform_upgrade when
upgrade_env is non-empty — generates lfs_jwt_secret before chaos redeploy.
Blocker 2 (REF=main upgrade fails HC1):
- Always use recipe_head_commit (git rev-parse HEAD) for head_ref instead
of using ref directly. When ref="main" (a branch name), the HC1 commit
check "head_ref.startswith(chaos_commit)" always fails since "main" ≠ SHA.
recipe_head_commit returns the actual SHA after the fetch/checkout.
Side-fix (stale creds — build #675):
- ops.py pre_install: delete the per-domain creds file before calling
_ensure_admin. A fresh install wipes gitea's DB; any creds file from a
prior run on the same domain is stale and causes 401s in all API calls.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When _enrich_deps_with_sso raises after deploy_deps succeeds (e.g., gitea API
call fails), deps_state stays {} and the finally block's `if deps_state:` guard
skips teardown, orphaning the dep at its deterministic domain.
Fix: add an `else` branch after the `if deps_state:` block that reads
$CCCI_DEPS_FILE (the legacy-list written by deploy_deps) and calls
teardown_deps on the cold entries so no dep is left running.
Unit tests: test_load_run_state_provides_fallback_for_enrichment_failure and
test_fallback_skips_warm_entries verify the data-flow that the fallback relies on.
19/19 unit tests pass.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The DG4.1 formula in run_recipe_ci.py is:
expected_deploy_count = 1 + deps_deployed_count
So when gitea dep deploys, the expected count becomes 2 (1 recipe + 1 dep).
The _count_deploy=False fix made dep deploys NOT count, giving actual=1 vs
expected=2 → DG4.1 violation even though the run was correct.
Original error "deploy-count 2 != 1" was because deps_state was empty when
the DG4.1 check ran (provisioning had failed), giving expected=1 while count
was already 2 from an early dep deploy. The proper fix is for _provision_deps
to succeed (which it now does), not to suppress counting.
Revert _count_deploy=False in deps.py; update docstrings for clarity.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
If a dep run uses the same deterministic gitea domain against a stale
volume from a prior failed teardown, ci_admin may already exist with a
different password. Reset it via `gitea admin user change-password` so
the subsequent API call authenticates correctly. This is idempotent and
does not affect clean (fresh-volume) runs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two issues found during first manual harness run:
1. gitea `--must-change-password false` (space form) leaves a pending
password-change for the ci_admin user, blocking the OAuth2 API call.
Fix: use `--must-change-password=false` (equals form, required by
gitea's BoolFlag with default=true).
2. dep deploy_app() calls incremented the DG4.1 "one deploy per run"
counter, causing a false violation when gitea dep + drone both deploy.
Fix: lifecycle.deploy_app gains _count_deploy=True param (default
backward-compat); deps_mod.deploy_deps passes _count_deploy=False so
only the recipe-under-test counts toward DG4.1.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
level.py: RUNGS += lint; statuses {pass,fail,skip,unver}; compute_level = max passed
rung with all below pass-or-skip (fail/unver block); cap_reason/capped DELETED.
harness/lint.py: lint executor — pristine scratch clone of the per-run tree at the
exact tested ref (mirror-origin + untracked-overlay pollution solved by context, no
rule filtered), PTY via script -qec, 60s hard budget, lint.txt artifact, table-parse
classifier (rc only signals FATA), unver on any non-run (never silent pass).
results.py: derive_rungs classifies every N/A source (structural/declared → skip,
else unver), lint rung + synthetic lint stage + lint block in results.json, schema 2,
cap fields removed. run_recipe_ci.py: lint call before tiers (double-wrapped,
verdict-neutral), badge = level only. card/dashboard: 0-5 ramp, cap line → 'level N
of {4|5}', unverified rows, badge number+colour only, lint.txt servable, old schema-1
artifacts render untouched. Unit suite rewritten: 245 passed on cc-ci venv.
Adversary heads-up (inbox 2026-06-10T19:06Z): meta values are repo-public by construction, but
the manifest lands on the dashboard — a field literally named SECRET_KEY_BASE showing a value
(plausible's committed CI dummy) is needless secret-scan noise. Mask values whose key NAME is
secret-shaped (SECRET|PASSWORD|TOKEN|CREDENTIAL|word-segment KEY), top-level and nested dict
keys; the key name stays visible. Unit test pins redacted vs passthrough (KEYCLOAK_URL).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
One block at run start answering "what does this recipe customize?" across every surface
(non-default recipe_meta keys, ops.py pre-ops, install_steps.sh, compose.ccci.yml, lifecycle
overlays by source, custom-test counts, active CCCI_SKIP_GENERIC* env overrides — !!-flagged when
riding a CI run, P2c), printed to the run log and embedded verbatim in results.json under
"customization". Pure presentation — building/printing it never influences a verdict; the
manifest honors the HC2 repo-local gate so it never advertises code the run will not execute.
Unit tests: synthetic recipe exercising every surface -> complete + deterministic + JSON-clean;
HC2 invisibility; env-override flagging; render golden lines; build_results threads the dict
verbatim (key always present, None when absent).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Placement RULE: discovery.custom_tests covers ONLY functional/ + playwright/ —
the top-level test_*.py glob for recipe dirs is removed (top level is reserved
for lifecycle overlays; zero in-repo users of top-level custom tests, verified
by sweep). Lifecycle-name exclusion inside the subdirs stays as the double-run
safety net. HC2 default-deny unchanged (repo-local custom now pinned via
functional/ in the gate test).
New conftest fixture op_state: parses $CCCI_OP_STATE_FILE (op context: versions,
artifact paths), skipping with a clear reason when unset/absent/unparseable —
overlay tests read op facts from the fixture instead of hand-parsing env (zero
existing hand-parsers found; the fixture is the documented path forward). deps
fixture landed in P2d.
Unit tests: placement-rule discovery tests (top-level custom NOT discovered;
functional/playwright are; misfiled lifecycle names excluded), op_state fixture
contract (reads file / skips without env / skips on missing file), deps fixture
attribute sugar.
Verified on cc-ci: cc-ci-run -m pytest tests/unit -q -> 184 passed; scripts/lint.sh -> PASS.
harness.meta.HookCtx (frozen): .domain, .base_url, .meta (RecipeMeta), .deps
(provisioned dep creds from $CCCI_DEPS_FILE or None), .op (current lifecycle op
or None); built via meta.hook_ctx() at each hook call site.
All recipe callables now take ctx: EXTRA_ENV(ctx), UPGRADE_EXTRA_ENV(ctx),
READY_PROBE(ctx), BACKUP_VERIFY(ctx), SCREENSHOT(page, ctx), ops.py pre_<op>(ctx).
Dict-valued EXTRA_ENV/UPGRADE_EXTRA_ENV unchanged (only the callable signature
moved). Call sites converted: deploy_app env shaping, perform_upgrade,
wait_ready_probes (gains op=), _perform_op BACKUP_VERIFY, screenshot.capture,
_run_pre_hook.
Legacy signatures fail FAST with a clear migration message: the registry carries
hook_params per hook key, enforced at meta.load() (MetaError names the old vs new
signature); ops.py pre-op hooks get the same check at the orchestrator call site
(meta.check_hook_signature) — no silent TypeError mid-run.
Migrated every in-repo user mechanically (17 ops.py files; cryptpad/lasuite-*/
mailu EXTRA_ENV; mumble+lasuite-drive READY_PROBE; ghost/discourse BACKUP_VERIFY)
— seeded values, probes and assertions byte-identical (domain -> ctx.domain;
keycloak pre_restore's meta arg -> ctx.meta).
Unit tests: hook_ctx field contract, ctx.deps from the run deps file, legacy-
signature MetaError (READY_PROBE/EXTRA_ENV/SCREENSHOT + pre-op checker), ctx
signatures accepted. Docs table regenerated (signature docs in key docs).
Verified on cc-ci: cc-ci-run -m pytest tests/unit -q -> 180 passed; scripts/lint.sh -> PASS.
a) compose.ccci.yml is FIRST-CLASS: the harness auto-copies tests/<recipe>/
compose.ccci.yml into the run's recipe checkout (ABRA_DIR-aware, lifecycle.
provide_ccci_overlay) and auto-chaoses the pinned base deploy on its presence
(kills the R7 implicit coupling). ghost/discourse install_steps.sh (copy-only
boilerplate) deleted; CHAOS_BASE_DEPLOY removed from both metas + the registry.
b) install-time deps wiring is the ONLY mode: deps with DEPS provision BEFORE the
single deploy; legacy post-deploy provisioning + the setup_custom_tests.sh
invocation machinery deleted. lasuite-docs migrated to install_steps.sh OIDC
wiring (same env names/values as the old hook — only the timing moved);
lasuite-drive's remaining post-deploy MinIO bucket one-shot moved to ops.py
pre_install; both setup_custom_tests.sh files deleted; OIDC_AT_INSTALL removed
from drive/meet metas + the registry.
c) SKIP_GENERIC meta key deleted (zero users). Env form CCCI_SKIP_GENERIC* stays
as the documented dev-only escape hatch; when active in a drone CI run the
orchestrator prints a loud !! warning (manifest embedding lands in P5).
d) conftest cleanup: dead pre-deploy-once fixtures deployed/deployed_app deleted
(zero users), app_domain + _short + _wait_healthy dropped (only users were the
deleted fixtures); deps_apps+deps_creds consolidated into ONE deps fixture
(entries expose .domain etc. as attributes; dict access intact); the 6 lasuite
test files renamed deps_creds->deps (fixture name only — assertions and flows
byte-identical). requires_deps marker + F2-11 skip-report plumbing unchanged.
Registry is now exactly the 14 final keys; docs §4 table regenerated. Stale
setup_custom_tests/OIDC_AT_INSTALL prose in docstrings/comments/assert MESSAGES
updated (no assert logic or expected value touched).
Verified on cc-ci: cc-ci-run -m pytest tests/unit -q -> 175 passed; scripts/lint.sh -> PASS.
One loader: runner/harness/meta.py::load(recipe) -> RecipeMeta (frozen dataclass,
attribute access), backed by the declarative KEYS registry (14 final keys + 3
P2-deprecated). The ONLY exec() of tests/<recipe>/recipe_meta.py. Validation per
the locked decision: unknown ALL-CAPS top-level name or type mismatch = MetaError
(hard error at load); underscore-prefixed names recipe-private; callables only on
hook-typed keys.
Migrated all six legacy loaders (spec §4 L1–L6):
- run_recipe_ci.py::_load_meta deleted; orchestrator loads once, passes meta down
- tests/conftest.py::_recipe_meta deleted; meta fixture returns full RecipeMeta (R3)
- lifecycle.py::_recipe_extra_env/_recipe_meta_flag deleted; deploy_app takes meta
- deps.py::declared_deps deleted; callers read meta.DEPS
- canonical.py::is_enrolled reads through meta.load()
- screenshot.py now actually receives SCREENSHOT through the orchestrator path (R2
fix; proven by unit test through the real load path)
Mumble private constants underscore-prefixed (_WELCOME_TEXT_MARKER/_MAX_USERS) +
importers fixed. New tests/unit/test_meta.py (all-recipes-load-clean typo gate,
MetaError cases, spec §2 baseline defaults, underscore exemption, doc sync). Docs
§4 key table now GENERATED from the registry (scripts/gen-meta-docs.py); drift
fails CI.
Verified on cc-ci: cc-ci-run -m pytest tests/unit -q -> 175 passed; scripts/lint.sh -> PASS.
The four CCCI state files (deploys countfile, opstate, deps, depskip) were keyed
by app domain in shared /tmp. A second run of the same domain executes its main()
preamble + deploy_app's pre-lock _record_deploy BEFORE blocking at the app lock,
so it reset/polluted the live first run's counter (false DG4.1 deploy-count=2,
build 279) and the first run's end-of-run os.remove crashed the second
(FileNotFoundError, build 281). Masked pre-restructure by the end-to-end recipe
flock. Now keyed by run id + harness pid via _run_state_path(); children receive
exact paths via the CCCI_*_FILE env vars, so domain keying was never load-bearing.
tests/concurrency/test_run_state.py: path-invariant cases + a real-process
regression (helpers.py deploy-count-run) reproducing the live interleaving —
verified to FAIL under simulated shared keying. docs/concurrency.md §3 updated.
- run_recipe_ci.setup_run_abra_dir(): builds <runs_dir>/<run-id>/abra with servers/ and
catalogue/ symlinked to the canonical ~/.abra (app .env files keep landing in the shared
canonical path, so janitor discovery and env-based teardown are unchanged; per-domain
filenames + the P2 app-domain lock prevent write conflicts) and a FRESH empty recipes/ —
each run clones + checkouts its own recipe trees. Exported as $ABRA_DIR (honored by the
abra CLI, verified on-host) before ANY abra call. Manual runs get manual-<pid> isolation.
- fetch_recipe(): plain clone into $ABRA_DIR/recipes/<recipe> — no shared-tree rm-rf, no lock.
CCCI_SKIP_FETCH=1 now copies the canonically-staged clone into the per-run tree (same staging
workflow, run reads staged state).
- abra.abra_dir()/recipe_dir(): single resolution rule ($ABRA_DIR else ~/.abra), used by
recipe_checkout, has_lightweight_version_tags, recipe_head_commit, recipe_versions,
generic._recipe_dir, lifecycle.prepull_images, snapshot_recipe_tests, and
warm_reconcile._recipe_dir (which keeps the canonical default for its own systemd runs but
follows the per-run tree when imported by promote_canonical inside a run).
- deleted: lifecycle.acquire_recipe_lock, RECIPE_LOCK_DIR, the main() call site and the
must-lock-before-fetch ordering rule.
- tests/{ghost,discourse}/install_steps.sh: RECIPE_DIR resolves ${ABRA_DIR:-$HOME/.abra} so the
compose.ccci.yml overlay lands in the tree the run actually deploys from (mechanical path fix
required by per-run trees; no assertion/gate touched — see DECISIONS.md).
- .drone.yml comments updated (HOME=/root rationale now via the servers symlink).
- acquire_app_lock(domain): exclusive flock on /run/lock/cc-ci-app-<domain>.lock, taken in
deploy_app exactly where register_run_app was (BEFORE app creation); blocks with a log line
when another run of the same domain is in flight (double-!testme serialisation). The file
object is retained in module-level _held_app_locks so GC can never close the fd and silently
release the lock. mtime is touched at acquisition (lock age for the long-held flag).
- janitor(): probes each candidate's lock (discovery unchanged: abra app ls + docker-service
sweep vs RUN_APP_RE). Acquirable -> orphan -> teardown_app(verify=False) WHILE HOLDING the
probe lock (a new same-domain run blocks until the reap finishes), then unlink before release.
Held -> live run -> leave it; held >120min (2x hard deadline) -> warn, never steal. Stale
unheld lockfiles with no app are unlinked on sight. Unreadable lockfile -> skip + log.
- unlink/recreate race guard (both sides): after ANY acquisition, verify the locked fd still is
the inode the path names (fstat vs stat); a waiter that won a just-unlinked inode retries on
the live path, and a probe that won one skips (unlinking now would hit a newer run's file).
- deleted: register_run_app, unregister_run_app, _run_owner_state, _registry_path,
ACTIVE_RUN_DIR, CCCI_JANITOR_MAX_AGE + age fallback, _stack_age_seconds, pid-reuse guard.
teardown_app no longer unregisters (release is process exit). janitor() takes no args now.
- post-reboot: /run/lock is tmpfs -> lockfiles gone -> probe trivially acquires -> immediate
reap (improvement over the old 2h age fallback).
- new harness/lifetime.py: install_lifetime_guards() arms PR_SET_PDEATHSIG(SIGTERM) (with
post-prctl ppid==1 orphan refusal), a SIGTERM handler raising SystemExit through the run's
finally: teardown funnel (exit 143), and signal.alarm(3600) funnelling SIGALRM the same way
with a distinct deadline log line (exit 142). Re-entrant signals during teardown are logged
and ignored (begin_teardown guard) so a second signal can't abort the running cleanup.
- run_recipe_ci.main(): guards installed first thing, before any abra call/lock; both teardown
finally: blocks (cold + quick) mark begin_teardown().
- .drone.yml recipe-ci step: harness runs under setsid in its own process group; a trap forwards
the step shell's TERM/EXIT to the whole group so drone cancel reaches the harness instead of
leaking it (docs/concurrency.md §8.1).
- PEP 446 note on the recipe-lock open(): the fd is non-inheritable, children never carry it.
68ef0f8 made services_converged() require UpdateStatus settled, treating
'paused' as in flight. But swarm's default update-failure-action pauses the
update on a single task flicker and the flag persists FOREVER (until the next
update): immich CI 241 had the app service 'paused' from a restart during
restore while the service was back at 1/1 and healthy — every subsequent wait
hung to its deadline and the run had to be killed.
Only 'updating' and 'rollback_started' now block convergence: those are the
states swarm is actively driving (the 238 stop-first race lives in 'updating').
'paused'/'rollback_paused' make no progress without intervention, so waiting on
them is pointless — N/N replicas is already required, and the HTTP-health and
tier assertions still gate whether the app actually works.
lint: PASS, unit tests: 138 passed.
services_converged() accepted N/N replicas as converged — but a chaos redeploy
that changes a non-app service image (immich PR #2 moves the db to the
vectorchord pin) registers a stop-first rolling update that swarm may not have
STARTED yet: the OLD task still shows 1/1, the wait passes, and the task dies
seconds later. Build 238: backupbot resolved the db hook container, the task
was killed in the gap, and the pre-hook exec crashed the whole backup with a
409 -> no dump in the snapshot -> restore had nothing -> RED.
- services_converged() now also requires every service's swarm UpdateStatus to
be settled ('', completed, rollback_completed) — updating/paused/rollback in
flight is NOT converged. Strictly stricter: no gate is weakened.
- backup_app() gains a bounded (300s) settle-wait before 'abra app backup
create' as defence in depth; on timeout the backup still runs and the tier's
assertion delivers the verdict.
lint: PASS, unit tests: 138 passed.
capacity=2 went live with three stale capacity=1-era assumptions that corrupted
concurrent runs (immich 229/230 '/pg_backup.sh: No such file'):
- ~/.abra/recipes/<recipe> is ONE shared working tree that fetch_recipe rm-rf's/
reclones and the upgrade tier git-checkouts mid-run. Same-recipe runs now
serialise on an exclusive flock (/run/lock/cc-ci-recipe-<recipe>.lock), taken
in main() BEFORE fetch_recipe and held for the whole run; the kernel releases
it on any process death, so there is no stale-lock failure mode. Different
recipes still run in parallel.
- CCCI_JANITOR_MAX_AGE=0 made a starting build reap ANY in-flight run app. Every
run now registers its app domain + pid in /run/cc-ci-active/<domain> before
app creation; the janitor checks the owner: alive (pid is a live run_recipe_ci
process) -> never reaped; dead -> reaped immediately; unknown (pre-registry or
post-reboot) -> age fallback (default 2h). The MAX_AGE=0 env override is gone
from .drone.yml.
- .drone.yml: concurrency.limit 1 -> 2 to match DRONE_RUNNER_CAPACITY=2; the
'safe because capacity=1' comments now describe the flock+registry model.
lint: PASS, unit tests: 138 passed.
Push builds have been RED on the lint step since ~build 209 from accumulated
formatting drift. This is the mechanical cleanup: ruff format + ruff --fix
(UP038 isinstance unions, SIM105 contextlib.suppress, UP031 f-strings, SIM115
tempfile context manager), shfmt -i 2 -ci, nixpkgs-fmt/statix/deadnix (merged
attrsets, dropped unused lib args), yamllint, and shell quoting fixes in
tests/lasuite-docs/setup_custom_tests.sh. No behaviour changes intended;
lint: PASS, unit tests: 138 passed.
autonomic-bot
