Skip to content

Commit 0081f65

Browse files
docs: the promote round's live findings — SITE_URL is set, autoDeploy is not
Two corrections from CD run 33266359801, the first run of the new promote path, plus the second push item 13's acceptance asks for (the first only creates `release`; this one fast-forwards an EXISTING ref, which is the case a depth-1 checkout fails). DIVERGENCES.md 4's live status was stale: SITE_URL IS set now. Measured from inside the workflow rather than assumed — the "Say plainly that the live half is dormant" step SKIPPED, which is `env.SITE_URL != ''`. DIVERGENCES.md gains 7, a record of the host rather than a divergence from the template's code: this Render service does not auto-deploy from a branch. The deploy hook item 13 removed was its ONLY trigger. 5c73a53 reached main at 17:43:00Z and release at 17:45:24Z; forty minutes later /healthz still served ae1dce6, where the previous deploy had been live within six minutes of its push. Render reacted to neither branch. Item 13 assumes autoDeploy is the mechanism — true of the template, not true here — so render.yaml's `branch: release` is documentation on this service and the dashboard is the switch. The owner step is Branch = release AND Auto-Deploy ON; it is not a code change and not doable from a session, so it is listed, not claimed. The same run also proved item 13's verify fix on its first opportunity: verify was SKIPPED after the failed deploy. The old `always() && != 'cancelled' && != 'skipped'` gate would have run it and reported GREEN against ae1dce6 — the previous build — which is precisely the defect the change exists to prevent. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EjfiNPKsgLwAxtN3BexPSJ
1 parent 5c73a53 commit 0081f65

2 files changed

Lines changed: 45 additions & 3 deletions

File tree

‎CHANGELOG.md‎

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -41,6 +41,25 @@ fork's shape, not byte-copied. Per-item dispositions are in the sync report.
4141
- **A `posture` fence in `DIVERGENCES.md`** — measured, not intended:
4242
`ai_bots`, `healthz: full`, `runtime: python`, `deploy: release-branch`.
4343

44+
### Known issue
45+
46+
- **This host's Render service does not auto-deploy from a branch, so
47+
production is frozen until the owner switches the dashboard.** The
48+
deploy hook this release removed was this service's ONLY deploy
49+
trigger: `5c73a53` reached `main` at 17:43:00Z and `release` at
50+
17:45:24Z, and forty minutes later `/healthz` still served `ae1dce6`,
51+
where the previous deploy had been live within six minutes. CD run
52+
33266359801 is red for exactly this — its promote step succeeded and
53+
its build-match wait timed out. Remedy (owner, Render dashboard, not a
54+
code change): Branch = `release` AND Auto-Deploy ON. Recorded as
55+
`DIVERGENCES.md` 7.
56+
- On the same run, item 13's `verify` fix proved itself: the job was
57+
**skipped** rather than run. Under the previous
58+
`always() && != 'cancelled' && != 'skipped'` gate it would have run
59+
after the failed deploy and reported GREEN against `ae1dce6` — the
60+
previous build. That is the defect the gate change exists to prevent,
61+
observed here on the first run that could exhibit it.
62+
4463
### Changed
4564

4665
- **`human_hits` DROPS and `bot_hits` RISES from this release.** UA-less

‎DIVERGENCES.md‎

Lines changed: 26 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -56,9 +56,10 @@ that inherits it verbatim polls SOMEBODY ELSE'S `/healthz` for
5656
fifteen minutes and reports red on day one. An empty `SITE_URL` is
5757
this fork's signal that no live target is configured yet; the
5858
build-match wait and the whole verify job skip with a notice. The
59-
guard stays. (Live status, 2026-08-26: the variable is still unset
60-
while the site is live — a repo-settings change for the owner, not a
61-
code divergence.) Since the 1.6.35 promote round the same guard is a
59+
guard stays. (Live status CORRECTED 2026-08-29: the variable IS
60+
now set — run 33266359801's "Say plainly that the live half is dormant"
61+
step SKIPPED, which is `env.SITE_URL != ''` measured from inside the
62+
workflow. The 2026-08-26 note that it was still unset is retired.) Since the 1.6.35 promote round the same guard is a
6263
SECOND conjunct on `verify`'s `if`, after the item's required
6364
`needs.deploy.result == 'success'`; `tests/test_cd_promotes_release.py`
6465
is therefore PORTED, not byte-copied — it asserts that deploy-success is
@@ -79,6 +80,28 @@ soak that gated the 2.7.0 tag), `PHASE1-REPORT.md` (this fork's build
7980
report) and `DEVELOPMENT-LOG.md` (its running log). They are records,
8081
not machinery — a sync should neither restore nor remove them.
8182

83+
**7. This host's Render service does not auto-deploy from a branch.**
84+
FOUND 2026-08-29 by sync item 13's first promoted run (33266359801),
85+
and the reason that run went red: with the deploy-hook POST removed,
86+
`5c73a53` was pushed to `main` at 17:43:00Z and fast-forwarded onto a
87+
newly created `release` at 17:45:24Z, and forty minutes later
88+
`/healthz` still served `ae1dce6` — while the PREVIOUS deploy had been
89+
live within six minutes of its push. Render reacted to neither branch,
90+
so the deploy hook was this host's ONLY deploy trigger and autoDeploy
91+
is off on the service. Item 13 assumes the opposite ("autoDeploy stays
92+
unset/on — it is the mechanism"), which is true of the template and is
93+
NOT true here. `render.yaml` correctly says `branch: release`, but this
94+
service is evidently not Blueprint-managed, so that line is
95+
documentation and the dashboard is the switch.
96+
97+
OWNER STEP, not a code change and not doable from a session: in the
98+
Render dashboard set Branch = `release` AND turn Auto-Deploy ON. Until
99+
both are set, CD's build-match wait will time out on every run and
100+
production stays frozen at the last hook-deployed build. This entry is
101+
a record of the host's state, not a divergence from the template's
102+
code — delete it once the dashboard is switched and a promoted run
103+
goes green.
104+
82105
## Byte-owned paths
83106

84107
Paths this fork owns byte-for-byte. The F3b fan-out never overwrites

0 commit comments

Comments
 (0)