access-engine/workplans/FLEX-WP-0036-runtime-rename-access-engine.md
tegwick 16ca97b277
Some checks failed
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 2s
Build and Publish Container Image / build-and-push (push) Failing after 1m2s
Complete renamed image source and verify bounded consumer migration readiness
Assistant: codex
Assistant-Model: gpt-6.1-sol
Assistant-Session: 01a126e1-76f8-7643-8cf3-e93336539ade
2026-10-11 23:25:19 +02:00

291 lines
15 KiB
Markdown

---
id: FLEX-WP-0036
type: workplan
title: "Runtime rename: flex-auth to access-engine, run side by side, consumers switch one at a time"
domain: infotech
repo: access-engine
status: active
owner: claude
topic_slug: netkingdom
created: "2026-10-10"
updated: "2026-10-10"
origin: demand
origin_ref: founder 2026-10-10 ("the relicts are too disorienting")
related: [FLEX-WP-0020]
state_hub_workstream_id: "af0f8d77-af04-569b-8e79-6c47eb8f1087"
---
`FLEX-WP-0020` renamed the repository only. `FLEX-DEC-2026-013` kept every
runtime name as `flex-auth` and deferred a product/runtime rebrand to "a later
workplan after soak". This is that workplan. The founder asked for it on
2026-10-10 because a repo called `access-engine` that deploys, builds, logs and
authenticates as `flex-auth` disorients every agent that works on it.
`FLEX-WP-0020-T11` (the soak) is still open. This plan does not wait for it to
start drafting, but T06 onward must not apply until T11 closes.
## What is renamed
| Surface | Today | Target | Who has to act |
| --- | --- | --- | --- |
| Kubernetes namespace | `flex-auth` | `access-engine` | operator |
| Deployments, Services, SAs, NetworkPolicies, Helm releases | `flex-auth-<consumer>` (9 pins) | `access-engine-<consumer>` | operator |
| TokenReview ClusterRoleBindings | `flex-auth-*-tokenreviewer` (9) | `access-engine-*-tokenreviewer` | operator |
| Service DNS consumers call | `flex-auth-<c>.flex-auth.svc.cluster.local.` | `access-engine-<c>.access-engine.svc.cluster.local.` | each consumer |
| Consumer egress NetworkPolicies | name namespace `flex-auth` | `access-engine` | `informed-decision`, `tenant-engine`, `user-engine` (net-kingdom `sso-mfa/k8s`) |
| Caller token audience | `flex-auth` | `access-engine` | PDP first, then each consumer |
| `resource.system` naming the PDP itself | `flex-auth` (ops-warden lane `flex-auth-policy-check`) | `access-engine` | PDP first, then ops-warden (`FLEX-DEC-2026-015`) |
| Container package | `coulomb/flex-auth` | `coulomb/access-engine` | this repo + Forgejo operator |
| Helm chart | `charts/flex-auth` | `charts/access-engine` | this repo |
| Go module, `cmd/`, binary | `github.com/netkingdom/flex-auth`, `flex-auth` | `.../access-engine`, `access-engine` | this repo; external callers of the binary |
| Env vars | `FLEX_AUTH_*` | `ACCESS_ENGINE_*` | this repo |
| Prose: SCOPE, INTENT, docs, canon, routing catalog | `flex-auth` | `access-engine` | this repo, net-kingdom, ops-warden |
Not renamed: the `FLEX-WP-`, `FLEX-DEC-` and `FLEX-IN-` id prefixes, all State
Hub UUIDs, and historical records (decision logs, archived workplans, decision
records already emitted). Each of those names what existed at the time it was
written.
## Rules for every task
1. **Run side by side, never rename in place.** Renaming a namespace, Service
or audience in place cuts off every consumer still using the old name. That
is a fail-closed outage nobody decided on. Each new `access-engine-*` pin is
started next to its `flex-auth-*` twin at the same image digest, and the
twin is retired only after its consumer has switched.
2. **The PDP accepts the new value before any consumer sends it**
(`FLEX-DEC-2026-015`): ship acceptance of both values first, then let
consumers flip, then retire the old value.
3. **One consumer at a time.** Each consumer switches on its own schedule and
under its own owner's work record. This plan records the handoff id and
waits for evidence. It never closes a task owned by another repo.
4. **Every live step needs the founder's go**, per pin, as T04 of
`FLEX-WP-0035` did.
## Decide the target names and supersede the retain table
```task
id: FLEX-WP-0036-T01
status: wait
needs_human: true
intervention_note: "Exact new runtime principals/naming are included in the single daily red batch with the nine TokenReview grants; no per-item question."
priority: high
state_hub_task_id: "4f9a796d-832f-5936-8a34-eb980f6e1b57"
```
Record `FLEX-DEC-2026-021`: runtime names move to `access-engine`, following
the table above. It supersedes the runtime rows of `FLEX-DEC-2026-013`. Fix
the namespace, release and DNS pattern, the audience value, and the
`resource.system` value. Settle the binary question: rename it outright, or
ship `access-engine` with a `flex-auth` alias for one release.
Gate: decision accepted by the founder.
2026-10-10: binary settled by `FLEX-DEC-2026-020` (founder): renamed outright,
`flex-auth` is a stub that prints the rename and exits 2. The other target
names in the table still need the founder's acceptance.
2026-10-11: target names are captured as proposed FLEX-DEC-2026-021. The new reviewer subjects are the scope-expansion component and are already in the single daily red batch. Binary disposition FLEX-DEC-2026-020 remains accepted; no founder acceptance of this broader table is fabricated.
## Inventory every consumer reference and open handoffs
```task
id: FLEX-WP-0036-T02
status: progress
priority: high
depends_on: [FLEX-WP-0036-T01]
state_hub_task_id: "8f597899-677e-544b-ad58-26aab2614312"
```
For each consumer, list every place that names the runtime: URL, audience,
egress policy, routing catalog entries, scripts, tests, stance maps, and the
`resource.system` value. Known consumers: `ops-warden` (two pins, plus
`flex-auth-policy-check`, registry and manifest scripts), `secrets-engine`,
`user-engine` (two pins: check and readback), `tenant-engine`,
`informed-decision` (lifecycle, sitting and t03 pins), and `net-kingdom`
(`sso-mfa/k8s/**`). Also check the State Hub, `state-hub` CLI and
`railiance-fabric` discovery for runtime names. Request one work-record id
from each owner.
Gate: `docs/runtime-rename-inventory.md` with one row per reference and an
owner handoff id per consumer.
2026-10-10 inventory written. Handoff ids are not yet requested. They wait for
T01, so each owner gets the exact new URL and audience in one message. Found
that informed-decision's component tests and ops-warden's production smoke
script run a binary whose path comes from an env var. A binary built from
current source must now come from `./cmd/access-engine`.
## Make the PDP accept both names
```task
id: FLEX-WP-0036-T03
status: wait
priority: high
depends_on: [FLEX-WP-0036-T01, FLEX-WP-0036-T04, RPF-WP-0044-T07]
state_hub_task_id: "487525e5-c89e-586e-8e9b-54ac6da0804e"
```
Caller auth accepts `access-engine` and `flex-auth` audiences, and packages
and caller bindings accept both `resource.system` values. Decision records
must still state which value the request carried, so nothing is rewritten
silently. Add tests for both values on every pin's package.
Gate: tests; a new image rolled to the existing `flex-auth-*` pins through
reviewed reversible blue execution under
`the-custodian/canon/standards/build-mode-change-authorization_v0.1.md`.
The ordered live prerequisite must be verified before rollout.
2026-10-11 source preparation: bounded compatibility accepts both exact PDP
audiences in Kubernetes TokenReview and records an actually reviewed spelling.
Custom audiences, installed service principals and protected-consumer system
bindings retain their existing meanings. All nine current pins authenticate both
audiences; eight enforcement packages retain their fixture outcomes, canonical
request bindings and original `resource.system` spelling. The ninth pin is the
separate application-readback route, whose installed binding is
`application-readback`, not a protected-system grant. There is no current
PDP-self resource lane in these pin packages; the old inventory's
`flex-auth-policy-check` example is a policy-routing lane outside this secret
fetch package. No new self-lane or caller grant is created to fill that absence.
T03 remains progress until the new image and ordered live checks are proven.
2026-10-11: source compatibility is tested and locally built. The next native step waits for the accepted runtime table, published exact artifact and the ordered Argo ownership proof (RPF-WP-0044-T07). No new founder rollout question; the source/public-package scope lives in the existing single batch.
## Rename the source
```task
id: FLEX-WP-0036-T04
status: wait
needs_human: true
intervention_note: "New public coulomb/access-engine image package publication is included in the single daily red batch. Local source/build checks pass; paired publication has not run."
priority: medium
depends_on: [FLEX-WP-0036-T01]
state_hub_task_id: "ff7fd369-326f-5044-aaa4-96c5f45c80e9"
```
Go module, `cmd/access-engine`, binary, `ACCESS_ENGINE_*` env vars (read the
old names for one release, with a warning), `charts/access-engine`, and CI
publishing to `coulomb/access-engine`. Publish to `coulomb/flex-auth` too
until T07 is done, so the old pins can still roll back. No runtime change.
Gate: build, tests and stage-1 render green; CI publishes both packages, and
both resolve to the same digest.
2026-10-10 binary part done ahead of the rest (`FLEX-DEC-2026-020`). The CLI is
now `cmd/access-engine` and the image entrypoint is
`/usr/local/bin/access-engine`. The image also ships the `flex-auth` stub.
Updated: Makefile, CI, `tools/verify-posture.sh`, `tenancy.yaml` and the live
docs. Workplans, decisions and ADRs are left as written.
2026-10-10 also done: Go module `github.com/netkingdom/access-engine`; no other
repo imports it. `ACCESS_ENGINE_RUN_TOPAZ_INTEGRATION` replaces the only
`FLEX_AUTH_*` variable, and the old name skips the test with a message naming
the new one. README identity paragraph and title updated. Still to do, and
deliberately left for T05: the chart name, because it changes selector labels,
which are immutable on a `helm upgrade`, so it ships with the new releases.
Also the image package and the `/opt/flex-auth` paths, which every live pin's
args reference.
2026-10-11: source now includes canonical `/opt/access-engine/examples` and retains the legacy paths for existing pins. The lifecycle Containerfile starts the renamed engine rather than the retired stub. The old Topaz integration variable remains usable for one release with a warning; an explicit new value takes precedence. CI builds once, publishes to the already admitted legacy package on normal push, and has an explicit paired-publication dispatch for the new package. `tools/publish-runtime-images.sh` independently compares both registry manifest digests; a local image ID or equal config alone is not publication proof. The new public package is a daily red-batch item; no paired dispatch has run. T04 remains progress until accepted publication resolves both repositories to the same immutable digest.
## Stand up the access-engine namespace next to flex-auth
```task
id: FLEX-WP-0036-T05
status: wait
priority: medium
depends_on: [FLEX-WP-0036-T03, FLEX-WP-0036-T04]
state_hub_task_id: "13b93d7a-8341-5c76-9ae8-0f36b45b10fc"
```
Create the `access-engine` namespace, then one `access-engine-<consumer>` pin
for each live `flex-auth-<consumer>` pin. Each new pin runs the same image
digest and package as its twin and gets its own TokenReview binding. Ingress
admits the same consumer as the twin does. Extend
`TestReadbackPinMatchesParticipantPins` to cover the new pins, and add a test
that each twin pair runs the same digest.
Gate: every new pin Ready, `/healthz` 200, anonymous requests 401; founder go.
Blocked on `FLEX-WP-0020-T11` closing.
2026-10-11: native wait qualified once against existing `depends_on` T03/T04:
coordinator must prove predecessor image rollout on existing pins before any
new namespace/pin creation. No wait-task inspection or new grant was needed.
Actual source preparation adds the renamed chart, exact-content generated twins,
explicit audience/reviewer paths and render/fixture correspondence checks.
Twins retain the exact existing image repository/digest and policy argv; these
are migration compatibility details until an identical digest is published under
the renamed repository and the baked new paths are proven. No namespace or
release has been created by this preparation. Local source preparation is reversible blue work under
`the-custodian/canon/standards/build-mode-change-authorization_v0.1.md`.
Native rollout creates nine new ServiceAccount subjects bound to cluster-scoped
`create` on `authentication.k8s.io/tokenreviews`; matching the prior role's verbs
does not make those new principals an unchanged access scope. It is one red
access-scope-expansion item in the single daily batch, including exact binding
revocation recovery. No per-item founder request is needed. The first prepared
blue receipt is preserved as superseded by the classification correction; no
native action occurred. T03/T04 and ordered predecessor gates remain unchanged.
## Consumers switch over, one at a time
```task
id: FLEX-WP-0036-T06
status: wait
priority: medium
depends_on: [FLEX-WP-0036-T02, FLEX-WP-0036-T05]
state_hub_task_id: "f0470e5c-3ea7-506c-b8bb-3ade0b57ff88"
```
Each consumer changes its URL, its egress policy (add the new namespace before
removing the old one), its audience and, for ops-warden, `owner_repo`. One row
per consumer in the inventory, closed with that owner's evidence: a live
allow and a live deny against the new pin, and no traffic left on the twin
(from twin logs).
Gate: every inventory row closed with owner evidence.
2026-10-11 native dependency wait qualified once on existing T02/T05 and
ordered predecessor live proof; no consumer switch executed. Actual source work
adds a deterministic `MigrationReadiness` classifier and public-source
re-derivation/owner handoff report. All five owners still require source work;
operator and tunnel custody remains with their owners. The verifier refuses to
turn comments, spoofed hosts, unsigned readiness flags or source-only changes
into native readiness. Owner-authenticated live allow/deny and timed old-twin
no-traffic/soak evidence remain necessary; this task is not closed by the report.
## Retire the flex-auth runtime
```task
id: FLEX-WP-0036-T07
status: todo
priority: medium
depends_on: [FLEX-WP-0036-T06]
state_hub_task_id: "e46c9c8e-a114-5c49-b035-22fdbd6a435b"
```
After a soak with no traffic on the old pins: uninstall each `flex-auth-*`
release and its TokenReview binding, delete the `flex-auth` namespace, drop
the old audience and `resource.system` from acceptance (`FLEX-DEC-2026-015`
order), and stop publishing `coulomb/flex-auth`. Keep the flattened emergency
manifests in `deploy/` only as rewritten for the new names.
Gate: founder go; `kubectl get ns flex-auth` NotFound; every consumer's checks
still allow and deny as expected.
## Prose, canon and layer declaration
```task
id: FLEX-WP-0036-T08
status: todo
priority: low
depends_on: [FLEX-WP-0036-T06]
state_hub_task_id: "be5440aa-4584-5e4f-9459-242ecf5c7199"
```
SCOPE, INTENT and its layer declaration, the conformance doc, the stance
register reviews, `.claude/rules`, the `deploy/` and `railiance/` READMEs,
`railiance/app.toml`, and `ops-warden`'s routing catalog and credential-routing
canon. Raise the canon-naming item (`FLEX-WP-0030` B5) as closed. Historical
records stay as written.
Gate: `grep -rn flex-auth` over live docs returns only historical records and
deliberate notes about the old name.