Environment Access Control

Executive Summary

The rhea family is fronted by Cloudflare Workers that route the *.pdcoe.dev hostnames to the correct EDS origins. Administrative access is expressed in each env's site.json (access.admin with role.admin / role.config_admin), and Cloudflare cache-purge is driven by a per-env Cloudflare API token stored in site.json .cdn. The routing worker performs host→origin mapping, cache control and ?ref= branch previews, and forwards the env's SITE_TOKEN so it can still reach an origin that has been locked with access.site.

Managed via config-as-code
The Cloudflare fronting, worker deployment, and site.json access/cdn blocks for rhea are all managed in abbvie-migrated/eds-site-config via its config-as-code workflows (deploy-cloud.yml, deploy-worker.yml). This page describes what those configs produce.

Architecture

Environments, One Code Repo

The rhea family's environments consume the same code repo (abbvie-migrated/rhea) via repoless configuration. Only the branch and content source differ per env. The Cloudflare-fronted envs are prod, stage, and dev; a fourth env, sandbox-dev, exists but is not Cloudflare-fronted.

EDS Site Branch Content Source Cloudflare-fronted Purpose
rhea main AEM Prod environment Yes Production
rhea-stage stage AEM Stage environment Yes QA / UAT
rhea-dev develop AEM Dev environment Yes Development
rhea (sandbox-dev) develop AEM sandbox program No Sandbox

CNAME Domains, Workers and DNS

Each Cloudflare-fronted env exposes a publish (live) and a preview URL, both routed by a single per-env routing worker. There are three workers (one per env), each bound to its env's two hostnames — deployed as rhea-<env>-routing-worker.

This is the authoritative hostname-to-origin mapping for the rhea family. Every row is also the DNS record: a proxied CNAME in the Cloudflare dashboard whose name is the hostname's first label and whose target is the EDS origin.

Worker CNAME (DNS name) EDS Origin (CNAME target) Surface
rhea-prod-routing-worker rhea.pdcoe.dev main--rhea--aa-pdcoe.aem.live live
rhea-preview.pdcoe.dev main--rhea--aa-pdcoe.aem.page preview
rhea-stage-routing-worker rhea-stage.pdcoe.dev stage--rhea-stage--aa-pdcoe.aem.live live
rhea-stage-preview.pdcoe.dev stage--rhea-stage--aa-pdcoe.aem.page preview
rhea-dev-routing-worker rhea-dev.pdcoe.dev develop--rhea-dev--aa-pdcoe.aem.live live
rhea-dev-preview.pdcoe.dev develop--rhea-dev--aa-pdcoe.aem.page preview

All six records must be proxied (orange cloud) with SSL/TLS mode Full (strict) — Worker routes and Zero Trust cannot intercept traffic otherwise.

`aa-pdcoe` in these hostnames is correct and is not a leftover
The GitHub org became abbvie-migrated in October 2026, but the aem.live delivery org did not move — it is still aa-pdcoe. The trailing --aa-pdcoe segment of every EDS hostname above, and the admin.hlx.page/config/aa-pdcoe/... paths later on this page, are therefore still right. Rewriting them to match the GitHub owner points them at an org that does not exist on aem.live and takes all six hostnames down.
Change this table in one place
This mapping previously appeared three times on this page and a fourth time in the Architecture Overview, so an environment change needed four edits and any missed one became a silent contradiction. Hostnames, branches and CDN config are generated from eds-site-config (sites/rhea/envs/<env>/env.yaml) — treat that as the source of truth and this table as its summary.

Request Flow

Traffic reaches EDS origins through Cloudflare via the routing worker.

User / Crawler
      │
      ▼
CNAME (*.pdcoe.dev, proxied)
      │
      ▼
Cloudflare Edge → routing worker
      ├── HOST_MAP: hostname → EDS origin (unknown host → 404)
      ├── strips/normalizes query params; blocks /drafts/ on live
      ├── sets x-forwarded-host, x-byo-cdn-type: cloudflare,
      │        x-push-invalidation: enabled
      ├── optional ?ref=<branch> preview (rewrites branch, forces .aem.page)
      ├── cache: cacheEverything on live; cacheTtl:0 on preview/?ref=
      ├── authorization: token <SITE_TOKEN>  (only if the secret is set)
      │
      ▼
EDS Origin (.aem.live / .aem.page)
      │   locked envs 401 anything without the token
      │
      │
      ▼
Cloudflare returns response to user

Key Principle

The worker routes and caches; it does not decide who may administer the site. That is configured server-side in site.json (access.admin), and Cloudflare cache invalidation uses a separate Cloudflare API token.

The one credential the worker does carry is the per-env SITE_TOKEN, forwarded as authorization: token <SITE_TOKEN> on origin fetches so a locked origin still answers the worker — see Layer 2 for the mechanism and the ordering constraint, and Layer 4 for invalidation.

Three distinct credentials are easy to conflate:

Credential Held by Purpose
access.admin roles site.json, server-side Who may administer the site
SITE_TOKEN The env's worker, as a secret Lets the worker reach a locked origin
Cloudflare API token GitHub secret, injected at deploy Cache purge on publish, and worker deploys

Security Layers

Layer 1: EDS Access Control (access.admin)

Access control is expressed inside each env's site.json under access.admin, not in a separate access/site.json file. The live rhea prod config uses:

"access": {
  "admin": {
    "requireAuth": "auto",
    "role": {
      "admin": ["philip.park@abbvie.com"],
      "config_admin": ["35461E63622A71650A495E62@techacct.adobe.com"]
    }
  }
}
  • role.admin — human administrators (by email) allowed to manage the site via the Admin API / Sidekick.
  • role.config_admin — the AEM technical account (IMS tech-account id) that CI uses to drive deploy-cloud writes for that env.
  • requireAuth: "auto" — Adobe's default admin-auth behavior.
Origin lock-down is being rolled out per environment

access.admin governs who may administer the site. Locking the raw .aem.live / .aem.page origins so they cannot be reached directly is a separate control, access.site, applied per environment by the Lock EDS Site workflow in eds-site-config.

Verified 2026-09-17 — only rhea-stage is locked so far:

Environment Raw .aem.live / .aem.page Custom *.pdcoe.dev
rhea-stage 401 — locked 200
rhea (prod) 200 — not yet locked 200
rhea-dev 200 — not yet locked 200

So a raw origin URL is currently still reachable on prod and dev. Treat the lock as the target state, not the current one, and check the env before assuming a preview URL is private. access.site needs allow only — a secretId is required solely so a fronting CDN can still reach the locked origin.

The table reports observed status. A lock is applied by the workflow against the Admin API, and the resulting access.site block is committed back afterwards — so an env can be enforcing 401 before its config lands in eds-site-config. Probe the URL rather than reading the repo if you need the current answer.

Layer 2: Cloudflare Routing Worker

Each Cloudflare-fronted env runs one routing worker (rhea-<env>-routing-worker), deployed from a single source file via wrangler deploy --env <name> with per-env route bindings in wrangler.toml. The worker:

  • Maps the request hostname to an EDS origin via a shared HOST_MAP (unknown host → 404).
  • Normalizes/strips query params (keeps format/height/optimize/width for media, limit/offset/sheet for JSON), and redirects non-443 ports to default.
  • Blocks /drafts/ on live surfaces (allowed on preview).
  • Sets x-forwarded-host, x-byo-cdn-type: cloudflare, and (unless disabled) x-push-invalidation: enabled.
  • Forwards authorization: token <SITE_TOKEN> when that secret is set for the env (see below).
  • Supports ?ref=<branch> branch previews: rewrites the branch segment, forces the .aem.page preview surface, and disables edge caching for that response.
  • Caches everything on live surfaces (cacheEverything) and bypasses cache on preview / ?ref= (cacheTtl: 0, no-store, strips ETag/Last-Modified).
The worker forwards a site token when one is set

The worker sets authorization: token <SITE_TOKEN> on origin fetches when the per-env SITE_TOKEN secret exists, and omits the header entirely when it does not — so it is inert until a token is stored.

That is what lets an origin be locked without taking the site down: the locked origin 401s every direct request, while the worker still gets through, making the custom hostname the only way in. Ordering matters — the token must reach the worker before access.site is enabled, or the custom hostnames start returning 401 too. The Lock EDS Site workflow does both steps in that order.

Layer 3: Cloudflare Zero Trust

Non-public preview/stage/dev hostnames can be gated behind a Cloudflare Zero Trust Access Application (policy: allow *@abbvie.com via SSO or one-time PIN), while rhea.pdcoe.dev (production publish) stays public. Zero Trust is configured in the Cloudflare dashboard, outside eds-site-config.

Layer 4: CDN Push Invalidation

CDN push invalidation is configured in each env's site.json under .cdn.<env>:

"cdn": {
  "live":    { "host": "rhea.pdcoe.dev" },
  "preview": { "host": "rhea-preview.pdcoe.dev" },
  "prod": {
    "type": "cloudflare",
    "plan": "enterprise",
    "host": "rhea.pdcoe.dev",
    "zoneId": "…",
    "apiToken": "${RHEA_PROD_CLOUDFLARE_API_TOKEN}"
  }
}

apiToken is a Cloudflare API token placeholder resolved at deploy time from the <FAMILY>_<ENV>_CLOUDFLARE_API_TOKEN GitHub secret (via envsubst in deploy-cloud.yml). AEM uses this token to purge the Cloudflare cache on publish. The same GitHub secret is reused by deploy-worker.yml for wrangler deploy. See the eds-site-config README — Secrets for the required token scopes.

Token and Secret Management

Two credentials matter for this setup. Both are managed centrally in eds-site-config as GitHub secrets — not hand-installed into Cloudflare workers.

Credential Format Used by Purpose
Admin API key opaque token deploy-cloud, pull-cloud-config, regenerate-sitemap, resync-code, key-health Authenticates CI to the AEM Admin API to write site.json / query.yaml / sitemap.yaml
Cloudflare API token opaque token deploy-cloud (embedded into site.json .cdn.<env>.apiToken), deploy-worker (wrangler deploy) AEM cache-purge on publish + worker deploys

Admin API key (ADMIN_API_KEY)

Stored as the ADMIN_API_KEY GitHub secret. Create/rotate it in tools.aem.live. The key-health.yml workflow warns weekly as the key nears expiry (<60 days warning, <30 days error). To rotate: create a new key, update the ADMIN_API_KEY secret, verify a deploy, then delete the old key. See the eds-site-config README for details.

Cloudflare API token (<FAMILY>_<ENV>_CLOUDFLARE_API_TOKEN)

One token per (family, env), e.g. RHEA_PROD_CLOUDFLARE_API_TOKEN. It is referenced as a placeholder in site.json (.cdn.<env>.apiToken = ${RHEA_PROD_CLOUDFLARE_API_TOKEN}) and substituted at deploy time — the value never lands in the repo. The same secret is reused by wrangler deploy. Required scopes: Zone: Cache Purge: Purge (the env's live + preview zones), Account: Workers Scripts: Edit, and Zone: Workers Routes: Edit.

Currently provisioned:

Secret Zones
RHEA_DEV_CLOUDFLARE_API_TOKEN rhea-dev.pdcoe.dev, rhea-dev-preview.pdcoe.dev
RHEA_STAGE_CLOUDFLARE_API_TOKEN rhea-stage.pdcoe.dev, rhea-stage-preview.pdcoe.dev
RHEA_PROD_CLOUDFLARE_API_TOKEN rhea.pdcoe.dev, rhea-preview.pdcoe.dev

To add a new Cloudflare-fronted env: mint the token in the Cloudflare dashboard, add the GitHub secret, and wire it into the env: blocks of deploy-cloud.yml, deploy-worker.yml, and pull-cloud-config.yml.

Deploying / redeploying workers

Workers are deployed by deploy-worker.yml (on push to workers/** or manual dispatch), which runs wrangler deploy --dry-run then the real deploy per [env.*] block. To redeploy manually:

gh workflow run deploy-worker.yml -f family=rhea -f worker=routing            # all envs
gh workflow run deploy-worker.yml -f family=rhea -f worker=routing -f env=dev # one env

Verifying access configuration

Inspect the live access/cdn blocks for an env:

curl -s https://admin.hlx.page/config/aa-pdcoe/sites/rhea.json \
  -H 'x-auth-token: <admin-auth-token>' | jq '{access, cdn}'

Sidekick Compatibility

The AEM Sidekick extension uses .aem.page URLs for preview and publish actions.

  • Authenticated AEM authors (signed in via Adobe IMS) can use Sidekick normally — their auth token is passed to the Admin API.
  • Sidekick preview/publish operations work regardless of whether the origin is locked, because they authenticate through the Admin API rather than through direct origin access.
  • A shared .aem.page link behaves differently per environment. Where access.site is enabled it returns 401 to the recipient; where it is not, the link is publicly reachable. Verified 2026-09-17: 401 on rhea-stage, 200 on rhea and rhea-dev.

Share the *.pdcoe.dev hostname rather than a raw .aem.page URL either way — it is the supported surface, and it does not change meaning as environments are locked.

Configuration Checklist

# Action Owner
1 Set access.admin (role.admin + role.config_admin) in each env's site.json Platform team
2 Add <FAMILY>_<ENV>_CLOUDFLARE_API_TOKEN GitHub secrets and wire into workflows Platform team
3 Create 6 DNS CNAME records (all proxied, SSL Full strict) Infra team
4 Deploy 3 routing workers (one per env, via deploy-worker.yml) Platform team
5 (Optional) Create Zero Trust Access Application for non-public CNAMEs Infra team
6 Configure .cdn.<env> (Cloudflare API token, zoneId) in site.json for push invalidation Platform team
7 Store SITE_TOKEN on the env's worker before enabling access.site Platform team
8 Lock the env's edge URLs (Lock EDS Site workflow) and commit the resulting access.site Platform team
9 Verify: CNAMEs return 200 via the routing worker QA team
10 Verify: prod CNAME is public; non-public CNAMEs gated as intended QA team
11 Verify: raw .aem.live / .aem.page return 401 for locked envs QA team

Crawler-facing items — robots, canonical tags, sitemap scope and auditing for hardcoded origin URLs — are on the SEO and Crawler Hygiene checklist.

On this page
h3
true