diff --git a/.gitea/workflows/ci.yml b/.gitea/workflows/ci.yml index eff29c08f..e3d79cbab 100644 --- a/.gitea/workflows/ci.yml +++ b/.gitea/workflows/ci.yml @@ -30,14 +30,24 @@ jobs: uses: actions/setup-node@v4 with: node-version-file: '.node-version' + cache: npm + + # Cache node_modules keyed on the lockfile + Node version: an unchanged + # lockfile skips `npm ci` (extraction + postinstall folds patch) and just + # restores the tree. Save runs only on a miss AND only if install + # succeeded (`success()`), so a failed `npm ci` can't poison the cache. + # The act_runner cache server is reached over the dedicated `act-cache-net` + # docker network (runner cache.host 172.30.0.2) — jobs previously landed on + # isolated per-job networks and ETIMEDOUT on it (~5 min wasted per build). + - name: Restore node_modules + id: node-modules + uses: actions/cache/restore@v4 + with: + path: node_modules + key: node-modules-${{ runner.os }}-${{ hashFiles('package-lock.json', '.node-version') }} - # No npm / node_modules cache: the act_runner's internal cache server is - # unreachable from job containers (`getCacheEntry failed: connect ETIMEDOUT - # 172.17.0.2`), so every cache restore hangs ~5 min and then fails — pure - # cost, zero benefit. `cache: npm` was removed from Setup Node above for the - # same reason. Re-enable both (setup-node `cache: npm` + an actions/cache - # node_modules step) once the runner's cache server is reachable from jobs. - name: Install dependencies + if: steps.node-modules.outputs.cache-hit != 'true' # Harden against transient registry network failures (ECONNRESET etc.): # raise npm's built-in fetch retries/timeouts and retry `npm ci` up to # 3 times with backoff before failing the build. @@ -57,6 +67,13 @@ jobs: sleep $((attempt * 15)) done + - name: Save node_modules + if: steps.node-modules.outputs.cache-hit != 'true' && success() + uses: actions/cache/save@v4 + with: + path: node_modules + key: node-modules-${{ runner.os }}-${{ hashFiles('package-lock.json', '.node-version') }} + # ── Quality gates run BEFORE the slow build so a format/lint/type/test # error fails in seconds instead of after the ~minutes-long build. All are # hard gates — any failure fails the job and blocks the deploy. The tree is