cc-ci/machine-docs/STATUS-dash.md

STATUS — phase dash (per-recipe run history fix)

SSOT: /srv/cc-ci/cc-ci-plan/plan-phase-dash-recipe-history.md Gates: M1 (fix implemented + locally verified) · M2 (deployed + verified live)

Gate: M1 CLAIMED, awaiting Adversary

WHAT — history_for(recipe) in dashboard/dashboard.py now sources the FULL per-recipe run history from the local run artifacts under /var/lib/cc-ci-runs (each run dir's results.json), newest-first by the finished timestamp, display-capped at HISTORY_CAP (default 30). It no longer reads the Drone …/builds?per_page=100 slice (the root cause: that window dropped a recipe's older runs out of view, so most recipes showed 1 run). Overview (/), /badge/<recipe>.svg, /runs/<id>/<file>, security guards, and stdlib-only constraint are unchanged.

WHERE —

• Commit: see git log on origin/main for the claim(M1) commit (this push).
• Changed files: dashboard/dashboard.py (new _run_status, _numeric_id, _local_history_row, _local_history; rewritten history_for; new HISTORY_CAP; new _LOCAL cache), and tests/unit/test_dashboard.py (new test_history_sourced_from_local_artifacts).
• Host artifacts the page reads: /var/lib/cc-ci-runs/<id>/results.json (bind-mounted read-only into the dashboard container, unchanged from before).

HOW to verify (cold, from a fresh clone) —

1. Unit suite (stdlib render + new local-sourcing test):

   nix-shell -p 'python3.withPackages(ps:[ps.pytest])' --run \
     'DRONE_TOKEN_FILE=$(mktemp) python3 -m pytest tests/unit/test_dashboard.py -q'
   

   EXPECTED: 13 passed.
2. Verify against the REAL host artifacts. Build a fixture of every results.json and run history_for against it (no Drone, no network):

   FIX=/tmp/advfix; rm -rf $FIX; mkdir -p $FIX
   ssh cc-ci 'cd /var/lib/cc-ci-runs && tar -cf - */results.json 2>/dev/null' | tar -xf - -C $FIX
   printf x > /tmp/t.tok
   DRONE_TOKEN_FILE=/tmp/t.tok CCCI_RUNS_DIR=$FIX python3 -c '
   import sys; sys.path.insert(0,"dashboard"); import dashboard as d
   r=d.history_for("bluesky-pds")
   print("count", len(r), [x["number"] for x in r])
   print("total parseable", sum(len(v) for v in d._local_history().values()))
   print("plausible cap", len(d.history_for("plausible")))'
   

   EXPECTED:
   • bluesky-pds count 8, order EXACTLY ['753','556','435','427','423','ab-bluesky-pds-oldmain','m2rr-bluesky-pds','m2r-bluesky-pds'] (newest-first by finished; note 423 sorts BELOW 427 though id 423<427, and named ids land in their timestamp positions — the mixed numeric+named id trap).
   • total parseable grouped rows 308 (matches host: 432 dirs, 308 with parseable results.json).
   • plausible capped at 30 (of 33), newest kept.

EXPECTED — invariants the Adversary's break-tests should confirm hold

• The 124 run dirs with no/malformed results.json are skipped (no 500, no garbage row): _results_for returns {} on miss/malformed/non-dir, _local_history skips any row with no recipe.
• Security preserved (untouched code paths): /recipe/<name> still gated by _RUN_ID_RE (^[A-Za-z0-9][A-Za-z0-9._-]*$ → rejects ../.., foo/.., spaces, ;); _results_for / serve_run_file still realpath-guarded against escaping /var/lib/cc-ci-runs.
• stdlib-only: no new imports (still html,json,os,re,sys,time,urllib,http.server).
• Overview (/) and /badge/<recipe>.svg still sourced from Drone latest-per-recipe (_custom_recipe_builds / latest_per_recipe unchanged) — only the history page changed source.
• Run-link resolution: numeric id → {DRONE_URL}/{CI_REPO}/<id>; named id (m2r-*, ab-*) → /runs/<id>/summary.html (local, since no Drone build number exists).
• Status pill derived from the per-stage results map (results.json has no top-level status): any fail/error → failure; all pass/skip → success; else unknown.

Gate: M2 CLAIMED, awaiting Adversary

WHAT — the dashboard service is rebuilt + redeployed with the M1 fix; the LIVE per-recipe history page now shows the full (display-capped) local-artifact history. Verified on bluesky-pds (8 runs) + plausible (30, capped from 33) + ghost (24); overview + badges + host health intact.

WHERE —

• Deployed image: cc-ci-dashboard:11ac2a1e6c07 (content hash of the M1 dashboard.py; rolled FROM 15addbc7bf45). Source built from commit 84ac65f+ (origin/main; this push adds the M2 status).
• Deploy: host flake clone /etc/cc-ci pulled, then nixos-rebuild switch from a path: flake of the synced working tree (path:/root/ccci-build#cc-ci) — a plain git-flake build drops the secrets/ submodule (gitlink), the path: copy includes the on-disk secrets/secrets.yaml. The deploy-dashboard reconcile rolled the swarm service on the new content-hash tag.
• Live: https://ci.commoninternet.net/recipe/<recipe>.

HOW to verify (cold) —

1. Deployed image + service health:

   ssh cc-ci 'docker service ls --filter name=ccci-dashboard --format "{{.Replicas}} {{.Image}}"'
   

   EXPECTED: 1/1 cc-ci-dashboard:11ac2a1e6c07.
2. Live full history (count rows = run count on host):

   for r in bluesky-pds plausible ghost; do
     echo -n "$r: "; curl -s https://ci.commoninternet.net/recipe/$r \
       | grep -coE '<tr><td><a href'; done
   

   EXPECTED: bluesky-pds 8, plausible 30 (capped from 33), ghost 24 — matching the host run counts (history_for cap = 30).
3. Live order matches host timestamp order (mixed numeric+named id trap):

   curl -s https://ci.commoninternet.net/recipe/bluesky-pds | grep -oE '>#[^<]+</a>' \
     | sed 's/[>#<]//g; s|/a||'
   

   EXPECTED exactly: 753 556 435 427 423 ab-bluesky-pds-oldmain m2rr-bluesky-pds m2r-bluesky-pds.
4. Other routes unaffected:

   curl -s -o /dev/null -w '%{http_code}\n' https://ci.commoninternet.net/            # 200 overview
   curl -s -o /dev/null -w '%{http_code}\n' https://ci.commoninternet.net/badge/bluesky-pds.svg  # 200
   

   EXPECTED: both 200; overview still latest-per-recipe (Drone-sourced, unchanged).

EXPECTED — retention confirmed adequate: no nix module/tmpfiles/cron trims /var/lib/cc-ci-runs (grep -rn cc-ci-runs nix/ shows no rm/find-delete/prune/maxage). Host: 439 run dirs spanning 2026-05-31 → 2026-06-17 (17 days). No growth cap needed now (recorded in DECISIONS).

Blocked

(none)