ci: run all test workflows weekly on master (issue #78)

.github/workflows/test-e2e-ddev.yml

@@ -6,9 +6,14 @@ name: test-e2e-ddev
 # ddev project lifecycle — too heavy per PR, but master is the staging
 # area for the 'stable' release mirror and must be fully green before a
 # release is cut (scripts/release.sh includes these runs in its CI gate).
-# Push runs use the defaults (no camino site tier, latest ddev); the
-# inputs only apply to manual dispatch. CI rebuilds the golden image warm
-# every run (the cache is a local-workspace concept, design doc §6 R6).
+# A weekly schedule (Mon 03:27 UTC, issue #78) re-runs the suite on
+# master even when nothing was pushed — the design doc's planned
+# burn-in (§8/R6): the suite installs the LATEST ddev and rebuilds the
+# golden image warm from scratch, catching ddev/opencode release drift
+# between pushes. Push and scheduled runs use the defaults (no camino
+# site tier, latest ddev); the inputs only apply to manual dispatch.
+# CI rebuilds the golden image warm every run (the cache is a
+# local-workspace concept, design doc §6 R6).
 on:
   push:
     branches:
@@ -25,12 +30,20 @@ on:
         required: false
         default: ''
         type: string
+  schedule:
+    - cron: '27 3 * * 1'  # Mondays 03:27 UTC (after test-e2e's verdict)
 
 permissions:
   contents: read
 
+# Cancel a superseded run when a new commit lands on the same ref; also
+# dedupes if two events somehow fire for the same push. Scheduled runs
+# get their own group: a weekly run must never cancel an in-flight
+# push run of the same ref (a cancelled run on a master sha reads as
+# red to scripts/release.sh's CI gate) — and conversely stays alive to
+# deliver its drift verdict even when master moves under it.
 concurrency:
-  group: e2e-ddev-${{ github.ref }}
+  group: e2e-ddev-${{ github.event_name == 'schedule' && 'weekly' || github.ref }}
   cancel-in-progress: true
 
 jobs:

.github/workflows/test-e2e.yml

@@ -7,6 +7,12 @@ name: test-e2e
 # staging area for the 'stable' release mirror — it must be green before
 # a release is cut (scripts/release.sh checks exactly this run). A
 # branch push with an open PR fires only `pull_request` — no double-run.
+# A weekly schedule (Mon 03:17 UTC, issue #78) re-runs both suites on
+# master even when nothing was pushed: the suites resolve the LATEST
+# opencode release at runtime, so a new opencode version can break a
+# "green" master between pushes — the cron catches that drift before a
+# release is cut. Scheduled runs use the defaults (E2E_STRICT=0, like
+# push runs; the strict input only applies to manual dispatch).
 on:
   workflow_dispatch:
     inputs:
@@ -21,6 +27,8 @@ on:
   push:
     branches:
       - master
+  schedule:
+    - cron: '17 3 * * 1'  # Mondays 03:17 UTC (after test-unit's verdict)
 
 # Least-privilege token (CodeQL actions/missing-workflow-permissions):
 # read-only access is enough (checkout + cache only — no artifacts,
@@ -29,9 +37,13 @@ permissions:
   contents: read
 
 # Cancel a superseded run when a new commit lands on the same ref; also
-# dedupes if two events somehow fire for the same push.
+# dedupes if two events somehow fire for the same push. Scheduled runs
+# get their own group: a weekly run must never cancel an in-flight
+# push/PR run of the same ref (a cancelled run on a master sha reads as
+# red to scripts/release.sh's CI gate) — and conversely stays alive to
+# deliver its drift verdict even when master moves under it.
 concurrency:
-  group: e2e-${{ github.ref }}
+  group: e2e-${{ github.event_name == 'schedule' && 'weekly' || github.ref }}
   cancel-in-progress: true
 
 jobs:

.github/workflows/test-unit.yml

@@ -7,6 +7,10 @@ name: test-unit
 # 'stable' release mirror — it must be green before a release is cut
 # (scripts/release.sh checks exactly this run). A branch push with an
 # open PR fires only `pull_request`, avoiding the classic double-run.
+# A weekly schedule (Mon 03:07 UTC, issue #78) re-runs the suite on
+# master even when nothing was pushed — runner image updates
+# (ubuntu-latest) and environment drift can break a "green" master
+# between releases. Scheduled runs always target the default branch.
 on:
   pull_request:
     branches:
@@ -15,16 +19,22 @@ on:
     branches:
       - master
   workflow_dispatch:
+  schedule:
+    - cron: '7 3 * * 1'  # Mondays 03:07 UTC (off-peak, off the hour)
 
 # Least-privilege token (CodeQL actions/missing-workflow-permissions):
 # read-only access to the repository is all this workflow needs.
 permissions:
   contents: read
 
 # Cancel a superseded run when a new commit lands on the same ref; also
-# dedupes if two events somehow fire for the same push.
+# dedupes if two events somehow fire for the same push. Scheduled runs
+# get their own group: a weekly run must never cancel an in-flight
+# push/PR run of the same ref (a cancelled run on a master sha reads as
+# red to scripts/release.sh's CI gate) — and conversely stays alive to
+# deliver its drift verdict even when master moves under it.
 concurrency:
-  group: tests-${{ github.ref }}
+  group: tests-${{ github.event_name == 'schedule' && 'weekly' || github.ref }}
   cancel-in-progress: true
 
 jobs:

CONTRIBUTING.md

@@ -41,6 +41,12 @@ make e2e-rootless      # docker-rootless daemon suite (needs systemd-in-containe
 - When adding a new executable under `files/` or a new test script under
   `tests/`, add it to the `chmod +x` list in **both**
   `.github/workflows/test-unit.yml` and `.github/workflows/test-e2e.yml`.
+- Besides PRs and `master` pushes, CI runs a **weekly scheduled** burn-in
+  on `master` (Mondays ~03:00 UTC, issue #78): the e2e suites install the
+  *latest* opencode/ddev releases at runtime, so the schedule catches
+  environment drift on a "green" master even when nothing was pushed.
+  Scheduled runs use their own concurrency group — they never cancel
+  push/PR runs, and a red weekly run blocks the next release by design.
 
 ## Testing a branch on a real machine
 

docs/design/ddev-e2e-test.md

@@ -15,6 +15,8 @@
 > **master push** (plus `workflow_dispatch` on demand) — master is the
 > staging area for the `stable` mirror and must be fully green before a
 > release is cut; PRs stay excluded (test-e2e.yml keeps the PR gate).
+> The planned weekly `schedule` itself landed 2026-09-11 (issue #78)
+> on all three workflows, including this one.
 >
 > Original framing, kept for context: a third e2e suite answering the
 > recurring burn-in problem — most ddev issues (#18, #20, #21, #25, the
@@ -423,7 +425,10 @@ bare-origin flow (DD12).
   **First green GitHub run (full camino tier): 2026-08-23** — after two
   plain-Linux fixes the workspace could not catch (R9 hosts seeding, R10
   fixture group baseline). Both prior failing runs are pinned in the
-  issue tracker. Weekly `schedule` is now unblocked.
+  issue tracker. Weekly `schedule` shipped 2026-09-11 (issue #78):
+  Mondays 03:27 UTC, defaults only (no camino tier, latest ddev), in a
+  dedicated concurrency group so a scheduled run can never cancel an
+  in-flight push run (release.sh reads a cancelled run as red).
 - **Docs:** this record + a MANUAL.md "troubleshooting with e2e-ddev" note +
   README testing mention in the same PR as the runner (repo rule: docs move
   with code).