build(deps): update plotly requirement from >=6.9.0 to >=7.0.0 in the dash-network group #11

Workflow file for this run

	# CI for llms.2plot.dev — the documentation site for dash-improve-my-llms.
	#
	# Inherited from dash-documentation-boilerplate, which is where the shape of
	# this file is the deliverable. Here it is just the network's CI baseline
	# (2plot.ai and 2plot.dev run the same one):
	#
	# * least-privilege `permissions` and a cancel-in-progress `concurrency`
	# group, so a workflow cannot write more than it reads and a rapid second
	# push does not race the first;
	# * an explicit `timeout-minutes` on every job — the default is six hours,
	# which is how a hung `curl` burns a day of runner minutes unnoticed;
	# * the real Docker image, built with a buildx GHA cache, then BOOTED, then
	# probed by the same battery that runs against production;
	# * version fingerprints asserted INSIDE the image, because pip metadata is
	# invisible from the outside and a stale artifact serves quietly;
	# * a secretless in-process pytest suite — no CLERK_*, no
	# CROSS_APP_WEBHOOK_SECRET — because the fail-closed behaviour is only
	# provable when nothing is configured;
	# * an advisory pip-audit.
	#
	# A fork changes the matrix and the image name — both done here. Everything
	# else is the standard; see the 2plot network standard on
	# boilerplate.2plot.dev (the template's own live site, not this one).

	name: CI

	# Deliberately NOT `push: branches: [main]`. cd.yml runs on that push and its
	# first job `uses:` this workflow, so a push to main used to start two runs of
	# it — which then contended for the `ci-${{ github.ref }}` concurrency group
	# below and cancelled each other. The work still got done (whichever run
	# survived did all of it), but every push left a `cancelled` CI run sitting
	# next to the green CD one, which reads as a failure at a glance.
	#
	# So: pull requests get their own CI, and `main` is owned by CD. There is no
	# coverage gap — CD cannot deploy without this workflow passing first.
	on:
	pull_request:
	workflow_dispatch:
	# Called by cd.yml so a deploy can never ship something the matrix rejected.
	workflow_call:

	# Read-only. Nothing in this workflow publishes, comments or tags; the deploy
	# lives in cd.yml behind a `production` environment.
	permissions:
	contents: read

	concurrency:
	group: ci-${{ github.ref }}
	cancel-in-progress: true

	env:
	PIP_DISABLE_PIP_VERSION_CHECK: "1"
	FORCE_COLOR: "1"
	# Never let a CI run inherit production behaviour: the base-URL guard keys
	# off RENDER / APP_ENV, and the satellite reporter keys off the webhook
	# secret. Both must stay inert here.
	APP_ENV: ci

	jobs:
	lint:
	name: lint
	runs-on: ubuntu-latest
	timeout-minutes: 10
	steps:
	- uses: actions/checkout@v7
	- uses: actions/setup-python@v7
	with:
	# The fleet Python — tests/test_python_version.py pins every
	# encoding of it (Dockerfile FROM tag, matrix main, render.yaml,
	# this job and pip-audit) to one minor.
	python-version: "3.14"
	cache: pip
	- run: pip install flake8
	- name: flake8
	run: flake8 lib components pages tests scripts run.py

	# The workflows lint themselves. This is not belt-and-braces: an invalid
	# workflow file is the one defect CI structurally cannot report, because
	# the run dies before a job exists to fail. A double-quoted string in a
	# ${{ }} expression on line 67 of THIS file silently killed every CI and
	# CD run on this repo for four days — nothing to click, nothing in the
	# API but `conclusion: failure` with zero jobs. actionlint catches it in
	# a second, and (this is the point) a pull request that reintroduces it
	# gets a red check with the offending column underlined.
	- name: actionlint
	run: \|
	bash <(curl -fsSL https://raw.githubusercontent.com/rhysd/actionlint/v1.7.7/scripts/download-actionlint.bash) 1.7.7
	./actionlint -color

	test:
	# SINGLE quotes inside ${{ }}. GitHub Actions expressions accept no other
	# string delimiter, and a double quote is a LEX error — which invalidates
	# the whole workflow file, not just this line. An invalid file is not
	# reported as a failing check: the run is created, dies in zero seconds
	# with zero jobs scheduled, and shows only "Failure" with nothing to open.
	# Every CI and CD run on this repo did that, unnoticed, for four days —
	# CD included, because its first job `uses:` this file. The `actionlint`
	# step in the lint job exists so this can never be silent again.
	name: py${{ matrix.python }} · ${{ matrix.backend }} · dash${{ matrix.dash && format(' {0}', matrix.dash) \|\| ' (pinned)' }}
	runs-on: ubuntu-latest
	timeout-minutes: 20
	strategy:
	fail-fast: false
	matrix:
	# Every backend gets a run on the FLEET Python — decided by
	# evidence, not preference: 3.14 ran the template's full suite and
	# its docker boot/battery green before it became the main
	# (2026-08-25), with dash 4.4.1, dash-improve-my-llms >=2.8.0 and
	# cryptography >=50 all importing. It is also what the production
	# image serves (Dockerfile FROM tag, same minor —
	# tests/test_python_version.py pins the two together), which the old
	# matrix never tested: main said 3.12 while the image was a
	# patch-pinned 3.11.8 that nothing here exercised.
	python: ["3.14"]
	backend: [flask, fastapi, quart]
	dash: [""]
	include:
	# ...the two adjacent minors keep the compat window three wide on
	# the default backend...
	- python: "3.13"
	backend: flask
	dash: ""
	- python: "3.12"
	backend: flask
	dash: ""
	# ...and the bottom of the `~=4.4.1` range is pinned explicitly on
	# both backends that matter, so a 4.4.0-only regression cannot hide
	# behind pip resolving to 4.4.1. 4.3.0 is deliberately absent: its
	# FastAPI backend 500s on every non-root URL (upstream, fixed in
	# 4.4.0), which is why requirements.txt floors at 4.4. These two
	# rows stay on a window leg (3.12), not the main: each include row
	# varies ONE axis — these vary dash, the rows above vary python —
	# so a red here always names its cause.
	- python: "3.12"
	backend: fastapi
	dash: "4.4.0"
	- python: "3.12"
	backend: flask
	dash: "4.4.0"
	steps:
	- uses: actions/checkout@v7

	- uses: actions/setup-python@v7
	with:
	python-version: ${{ matrix.python }}
	cache: pip

	- name: Install the app
	run: \|
	pip install -r requirements.txt
	# markdown2dash 0.1.2 declares gunicorn<22 against the CVE-driven
	# gunicorn>=23 floor. Same two-command install as the Dockerfile,
	# render.yaml and scripts/dev.sh.
	pip install --no-deps markdown2dash==0.1.2
	# Dash's own extras are required for the ASGI backends: a bare
	# `fastapi` install is not enough for dash.backends._fastapi to
	# import. httpx backs starlette's TestClient.
	if [ "${{ matrix.backend }}" != "flask" ]; then
	pip install "dash[${{ matrix.backend }}]" httpx
	# The backend extra from PyPI, same floor as requirements.txt.
	pip install "dash-improve-my-llms[${{ matrix.backend }}]>=2.8.0"
	fi
	pip install pytest pytest-cov
	# An explicit matrix pin overrides requirements.txt, to prove the
	# bottom of the supported range still works.
	if [ -n "${{ matrix.dash }}" ]; then
	pip install "dash[${{ matrix.backend }}]==${{ matrix.dash }}" \
	\|\| pip install "dash==${{ matrix.dash }}"
	fi

	- name: Confirm the pinned dependency versions
	run: \|
	python - <<'PY'
	import dash, dash_improve_my_llms as pkg, gunicorn

	def parts(v):
	return tuple(int(x) for x in v.split(".")[:3] if x.isdigit())

	# 4.4 floor: 4.3.0's FastAPI backend 500s on every non-root URL.
	assert parts(dash.__version__)[:2] >= (4, 4), dash.__version__
	# 2.8.0 is this app's floor: lib/analytics_tracker imports
	# `classify` / `_ledger.EVENT_FIELDS` at module scope and run.py
	# calls `on_document_read` unconditionally, so anything lower is
	# an ImportError at boot, not a degraded feature.
	assert parts(pkg.__version__) >= (2, 8, 0), pkg.__version__
	# 21.x carried two request-smuggling CVEs (CVE-2024-6827,
	# CVE-2024-1135). markdown2dash's spurious <22 pin must not win.
	assert parts(gunicorn.__version__)[:2] >= (23, 0), gunicorn.__version__
	print(f"dash {dash.__version__}, dash-improve-my-llms "
	f"{pkg.__version__}, gunicorn {gunicorn.__version__}")
	PY

	# No CLERK_*, no CROSS_APP_WEBHOOK_SECRET, no SESSION_SECRET here ON
	# PURPOSE. tests/conftest.py pins them empty and the fail-closed checks
	# in tests/test_access.py depend on that posture; a secret injected here
	# would make the suite pass for the wrong reason.
	- name: Test suite (${{ matrix.backend }}, zero secrets)
	env:
	DASH_BACKEND: ${{ matrix.backend }}
	run: pytest tests -q

	- name: Boot under a production server
	if: matrix.backend == 'flask'
	env:
	# This gunicorn serves the MATRIX LEG's interpreter (3.13/3.12 on
	# the window rows) while the Dockerfile declares the fleet Python —
	# off-contract for the battery's python_matches_declared check by
	# design, so it stands down here. The docker job below and CD's
	# verify leave it armed: there the interpreter IS the artifact.
	SMOKE_PYTHON_DECLARED: ignore
	run: \|
	gunicorn run:server -b 127.0.0.1:8550 --daemon --access-logfile - --error-logfile -
	for _ in $(seq 1 30); do
	curl -sf http://127.0.0.1:8550/healthz && break
	sleep 1
	done
	# A page that renders under the test client can still fail under a
	# real WSGI worker — different import path, different working
	# directory, no test-client conveniences.
	curl -sf http://127.0.0.1:8550/ > /dev/null
	curl -sf http://127.0.0.1:8550/networks > /dev/null
	# The battery, against the same server a satellite deploys.
	python3 scripts/network_smoke.py --base-url http://127.0.0.1:8550

	docker:
	name: docker image · boot · battery
	runs-on: ubuntu-latest
	timeout-minutes: 25
	needs: [test]
	steps:
	- uses: actions/checkout@v7

	# The same build Render runs. This is where a dependency-resolution
	# failure surfaces — at CI time, not deploy time, where the only signal
	# is a dashboard log while the old image keeps serving.
	- uses: docker/setup-buildx-action@v4
	- name: Build the production image
	uses: docker/build-push-action@v7
	with:
	context: .
	tags: llms-2plot-dev:ci
	load: true
	cache-from: type=gha
	cache-to: type=gha,mode=max

	# pip metadata is invisible from outside a running host, so the versions
	# are asserted here, inside the artifact that actually ships.
	- name: Version fingerprints inside the image
	run: \|
	docker run --rm llms-2plot-dev:ci python -c "
	from importlib.metadata import version

	def parts(v):
	return tuple(int(x) for x in v.split('.')[:3] if x.isdigit())

	v = version('dash')
	print('dash', v)
	assert parts(v)[:2] >= (4, 4), f'expected dash >=4.4, image has {v}'

	v = version('dash-improve-my-llms')
	print('dash-improve-my-llms', v)
	assert parts(v) >= (2, 8, 0), f'expected >=2.8.0 (one classifier, the read event, geo guardrail, panel, vendor policy), image has {v}'

	# markdown2dash installs with --no-deps to dodge its gunicorn<22
	# pin; this assert is what proves the dodge kept working. 21.x
	# carried two request-smuggling CVEs (CVE-2024-6827, CVE-2024-1135).
	v = version('gunicorn')
	print('gunicorn', v)
	assert parts(v)[:2] >= (23, 0), f'expected gunicorn>=23, image has {v}'

	# ...and that skipping its dependency graph did not skip the package.
	import markdown2dash # noqa: F401
	print('markdown2dash importable')
	"

	# Boot with no secrets: Clerk falls open (dev mode), the reporter stays
	# dormant, the hub client reports itself disabled. What this catches is
	# any import-time or preload crash — the class of failure where the
	# platform loops the worker and the deploy never goes live.
	- name: Boot the container and wait for /healthz
	run: \|
	docker run -d --name docs -p 8550:8550 llms-2plot-dev:ci
	for i in $(seq 1 60); do
	if curl -sf http://127.0.0.1:8550/healthz > /dev/null; then
	echo "healthy after ~$((i*2))s"
	exit 0
	fi
	if [ "$(docker inspect -f '{{.State.Running}}' docs)" != "true" ]; then
	echo "container exited during boot:"
	docker logs docs
	exit 1
	fi
	sleep 2
	done
	echo "never became healthy; last logs:"
	docker logs --tail 100 docs
	exit 1

	# The SAME script CD runs against the live host (once the SITE_URL repo
	# variable is set — see cd.yml's header), so a failure in CI and a
	# failure in production read identically.
	- name: Smoke battery against the booted container
	run: python3 scripts/network_smoke.py --base-url http://127.0.0.1:8550

	# The guard on the guard (emojimart's F2 finding, 2026-08-24): the
	# boot step curls /healthz from OUTSIDE, which proves the app binds
	# and answers — but says nothing about the Dockerfile's HEALTHCHECK
	# instruction itself. A broken probe (wrong port, missing binary,
	# bad quoting) ships silently while every external check stays
	# green, and the orchestrator that relies on container health sees
	# an opaque box. Poll Docker's own verdict: with --interval=30s and
	# --start-period=20s the first probe lands within ~50s.
	- name: Assert Docker's own health verdict
	run: \|
	for i in $(seq 1 30); do
	status="$(docker inspect -f '{{.State.Health.Status}}' docs 2>/dev/null \|\| echo none)"
	if [ "$status" = "healthy" ]; then
	echo "docker reports healthy after ~$((i*3))s"
	exit 0
	fi
	if [ "$status" = "none" ]; then
	echo "image has NO HEALTHCHECK — the container is opaque to its orchestrator"
	exit 1
	fi
	sleep 3
	done
	echo "HEALTHCHECK never reported healthy (last: $status) — the probe itself is broken:"
	docker inspect -f '{{json .State.Health}}' docs \|\| true
	exit 1

	- name: Container logs (for the record)
	if: always()
	run: docker logs --tail 40 docs 2>/dev/null \|\| true

	pip-audit:
	name: pip-audit (advisory)
	runs-on: ubuntu-latest
	timeout-minutes: 10
	# Advisory on purpose. A CVE in a transitive dependency of a docs site is
	# worth knowing about the day it lands, and worth nobody's broken build at
	# 2am. The report is the value; flip this off once the baseline is quiet.
	continue-on-error: true
	steps:
	- uses: actions/checkout@v7
	- uses: actions/setup-python@v7
	with:
	python-version: "3.14"
	- run: pip install pip-audit
	# Skip local vendor/ paths — pip-audit can only assess PyPI dists.
	- run: \|
	grep -v '^\./vendor/' requirements.txt > /tmp/req-pypi.txt
	pip-audit -r /tmp/req-pypi.txt --skip-editable

Navigation Menu

build(deps): update plotly requirement from >=6.9.0 to >=7.0.0 in the dash-network group #11

Workflow file

build(deps): update plotly requirement from >=6.9.0 to >=7.0.0 in the dash-network group #11

Uh oh!

Workflow file for this run