From b9f07e84ad7239519a624bbcbacb26e5092172fd Mon Sep 17 00:00:00 2001 From: Joep Meindertsma Date: Thu, 24 Sep 2026 22:19:28 +0000 Subject: [PATCH 1/3] Stop dispatching atomic-saas staging deploys on every develop push atomic-saas deploys staging only when started by hand since its PR 148, so deploy_staging.yml no longer listens for the atomic-server-develop event and this workflow's dispatch reached nothing. To test an atomic-server branch there, start Deploy to staging in atomic-saas with atomic_server_ref set to it. ATOMIC_SAAS_DISPATCH_TOKEN stays: main.yml's Downstream job still uses it to read atomic-saas. --- .github/workflows/notify-saas-staging.yml | 42 ----------------------- 1 file changed, 42 deletions(-) delete mode 100644 .github/workflows/notify-saas-staging.yml diff --git a/.github/workflows/notify-saas-staging.yml b/.github/workflows/notify-saas-staging.yml deleted file mode 100644 index 1f9c7544fd..0000000000 --- a/.github/workflows/notify-saas-staging.yml +++ /dev/null @@ -1,42 +0,0 @@ -name: Notify atomic-saas staging - -# Every push to develop tells ontola/atomic-saas to redeploy its staging -# (staging.atomicserver.eu) against exactly this commit. Its -# deploy_staging.yml listens for the `atomic-server-develop` event and reads -# `client_payload.sha`. Until 2026-09-03 the same thing took a pin-bump PR -# over there plus a manual dispatch per change. -# -# Fires on the push, not on the main pipeline passing: PRs are green before -# they merge, the deploy verifies itself (health check, smoke test, rollback -# on the box), and waiting on the pipeline would tie staging to Mancave being -# awake. deploy_staging.yml in this repo (staging.atomicdata.dev) still waits -# for the pipeline; that is a different box with a different bar. -# -# ATOMIC_SAAS_DISPATCH_TOKEN: a fine-grained PAT for ontola/atomic-saas with -# Contents: read and write (repository_dispatch needs write). GITHUB_TOKEN -# cannot cross repositories. Missing token → the step fails loudly rather -# than staging silently going stale. - -on: - push: - branches: [develop] - -jobs: - dispatch: - runs-on: ubuntu-latest - timeout-minutes: 5 - steps: - - name: Dispatch a staging deploy in atomic-saas - env: - GH_TOKEN: ${{ secrets.ATOMIC_SAAS_DISPATCH_TOKEN }} - run: | - set -euo pipefail - if [ -z "$GH_TOKEN" ]; then - echo "::error::ATOMIC_SAAS_DISPATCH_TOKEN is not set; atomic-saas staging will not follow this push." - exit 1 - fi - gh api repos/ontola/atomic-saas/dispatches \ - -f event_type=atomic-server-develop \ - -f "client_payload[sha]=${{ github.sha }}" \ - -f "client_payload[ref]=${{ github.ref }}" - echo "Dispatched: atomic-saas staging will build against ${{ github.sha }}." From 467f1829f2a11588a5832d4961d4342dca4652e2 Mon Sep 17 00:00:00 2001 From: Joep Meindertsma Date: Fri, 25 Sep 2026 01:41:56 +0000 Subject: [PATCH 2/3] Give the CI job a 150-minute timeout The ci job had no timeout-minutes, so GitHub's 6-hour default applied. On 2026-09-25 both Mancave CI jobs (develop run 4587 and batch run 4591) stopped logging at the same minute, mid e2e, and would hold both runners until that default released them. A full run takes 45 to 75 minutes; 150 leaves room for a cold per-runner cache. The images job already has 120. --- .github/workflows/main-ci.yml | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/.github/workflows/main-ci.yml b/.github/workflows/main-ci.yml index d5121b0a99..6dbfde251e 100644 --- a/.github/workflows/main-ci.yml +++ b/.github/workflows/main-ci.yml @@ -32,6 +32,10 @@ jobs: ci: name: CI runs-on: ${{ fromJSON(inputs.runner) }} + # A full run takes 45 to 75 minutes. Without a limit, a wedged Dagger + # engine held both Mancave runners for GitHub's 6-hour default + # (2026-09-25, runs 4587 and 4591). 150 leaves room for a cold cache. + timeout-minutes: 150 env: NETLIFY_TOKEN: ${{ secrets.NETLIFY_TOKEN }} steps: From f7b03bd9e177dc66fd4a153136b4ecd160dd87d8 Mon Sep 17 00:00:00 2001 From: Joep Meindertsma Date: Fri, 25 Sep 2026 07:21:30 +0000 Subject: [PATCH 3/3] Install pnpm into each runner's own temp dir pnpm/action-setup defaults to ~/setup-pnpm, which every runner on a host shares. On 2026-09-25 develop run 4596 failed at Install pnpm with ENOTEMPTY because the batch's Downstream job on the other Mancave runner was installing into the same directory at the same second. Point dest at runner.temp in the two jobs that run on Mancave. --- .github/workflows/main-ci.yml | 4 ++++ .github/workflows/main.yml | 2 ++ 2 files changed, 6 insertions(+) diff --git a/.github/workflows/main-ci.yml b/.github/workflows/main-ci.yml index 6dbfde251e..075be0ddec 100644 --- a/.github/workflows/main-ci.yml +++ b/.github/workflows/main-ci.yml @@ -71,6 +71,10 @@ jobs: with: version: 10.15.1 run_install: false + # The default, ~/setup-pnpm, is shared by every runner on the host. + # Two jobs installing at once failed with ENOTEMPTY (2026-09-25, + # run 4596), so each job installs into its own runner's temp dir. + dest: ${{ runner.temp }}/setup-pnpm - name: Install Dagger regression test dependencies run: pnpm --dir .dagger install --frozen-lockfile - name: Dagger regression tests diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml index cd6153d3a1..4e3fa3a903 100644 --- a/.github/workflows/main.yml +++ b/.github/workflows/main.yml @@ -180,6 +180,8 @@ jobs: uses: pnpm/action-setup@v4 with: version: 10.15.1 + # Per-runner, not the shared ~/setup-pnpm; see main-ci.yml. + dest: ${{ runner.temp }}/setup-pnpm - name: Install Node uses: actions/setup-node@v4