Avoid unreachable Gitea runner cache

This commit is contained in:
Kevin
2026-09-04 02:09:46 -04:00
parent 6fa54a23d9
commit 052fa31d87
7 changed files with 9 additions and 15 deletions
+8 -3
View File
@@ -81,9 +81,14 @@ the pilot:
- The existing branch preview workflows remain in place. Local output checks
now prove preview robots rules and page metadata stay no-index and that the
admin response has a no-index header rule.
- Validation, preview, and production workflows cache npm's download store by
`package-lock.json` while retaining clean `npm ci` installs. Warm runner jobs
avoid repeatedly downloading the same dependency tarballs.
- Validation, preview, and production workflows use clean `npm ci` installs with
`--prefer-offline --no-audit --no-fund`. Do not enable `setup-node`'s npm cache
until the Actions runner cache service is reachable from job containers. The
September 4 pilot test timed out against the runner's auto-detected private
address during both cache restore and cleanup. Give `act_runner` a fixed
`cache.host` and `cache.port` reachable from its job containers, publish that
port (or attach runner and jobs to a shared network), restart the runner, and
then restore the lockfile-keyed `cache: npm` configuration.
The pilot dependencies are active: the public client ID, exact-origin CORS
policy, `cms-pilot` branch, and both protection rules have been verified.