RPF-WP-0045: plan Kubernetes-auth migration for openbao-activity-core and openbao-email-connect
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 6s

Declares target store specs, two exact-path policies and two ESO
ServiceAccounts. Nothing applied; live steps wait on the founder.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 63291@bnt-lap001
Assistant-Session: 8bd77868-ca68-4f49-bb1e-d539ecc0d703
This commit is contained in:
codex 2026-09-21 19:18:59 +02:00
parent 2d79c2e4e5
commit 6e399bbe64
6 changed files with 347 additions and 25 deletions

View file

@ -0,0 +1,264 @@
---
id: RPF-WP-0045
type: workplan
title: "Move openbao-activity-core and openbao-email-connect off expired static tokens to Kubernetes auth"
domain: financials
repo: railiance-platform
status: ready
flavor: implementation
owner: railiance-platform
topic_slug: railiance
created: "2026-09-21"
updated: "2026-09-21"
related: [RPF-WP-0037]
---
Two ClusterSecretStores on railiance01 fail validation: OpenBao answers 403 on
`auth/token/lookup-self` for the static tokens they reference
(`openbao-email-connect` since 2026-09-13, `openbao-activity-core` since
2026-09-19). Five ExternalSecrets are `SecretSyncedError`. This plan moves both
stores to the method RPF-WP-0037 used for `openbao-forgejo`, `openbao-reuse`
and `openbao-target-revenue`. It is an authentication migration, not a
credential rotation. No KV value changes.
**This plan changes nothing live by itself.** Every live task is `wait` on the
founder's go-ahead. T01 is `ADMINISTER @ realm:openbao/railiance01` with the
platform-admin credential, `activation=APPROVED`. T03, T05 and T06 are
`ADMINISTER @ realm:kubernetes/railiance01`, `activation=APPROVED`. No step
needs `BREAK_GLASS`. Evidence is `EvidenceBoundary target-audited`
(kubectl status and OpenBao audit log on railiance01).
## Read-only evidence, 2026-09-21
Taken over `ssh railiance01 'kubectl get …'` and `kubectl auth can-i`. No
secret value was requested.
- Recovered method: `spec.provider.vault.auth.kubernetes`, mountPath
`kubernetes`, a **dedicated ServiceAccount in the consumer namespace**,
`audiences: [openbao]`, a per-lane OpenBao role. Source:
`openbao/eso-auth-recovery/lanes.json`, `scripts/repair_eso_kubernetes_auth.py`.
| Store | SA | OpenBao role | Policy |
| --- | --- | --- | --- |
| openbao-forgejo | forgejo/forgejo-mailer-eso | forgejo-mailer-eso | workload-kv-read-forgejo-mailer-eso |
| openbao-reuse | reuse/reuse-surface-eso | reuse-surface-runtime-eso | workload-kv-read-reuse-surface-runtime-eso |
| openbao-target-revenue | target-revenue/target-revenue-eso | target-revenue-runtime-eso | workload-kv-read-target-revenue-runtime-eso |
Role shape: bound to exactly one SA name and one namespace, audience
`openbao`, ttl/max_ttl/explicit_max_ttl 15m, `token_no_default_policy=true`,
one policy. Policy: exact `platform/data/...` read plus `auth/token/lookup-self`
(read) and `auth/token/revoke-self` (update). All three stores are Ready.
- Failing ExternalSecrets and the KV paths they read (live `spec.data[].remoteRef.key`):
| Namespace | ExternalSecret | KV path | Target Secret |
| --- | --- | --- | --- |
| activity-core | actcore-backup-offsite | platform/workloads/railiance/backup/offsite-lane | actcore-runtime-secret (Merge) |
| activity-core | actcore-forgejo-admin | platform/workloads/forgejo/forgejo-admin | actcore-runtime-secret (Merge) |
| activity-core | actcore-issue-core-runtime | platform/workloads/issue-core/issue-core/issue-core-runtime | actcore-runtime-secret (Merge) |
| activity-core | llm-connect-provider-secrets | platform/workloads/activity-core/llm-connect/llm-connect-provider-secrets | llm-connect-provider-secrets (Owner) |
| email-connect | email-connect-runtime | platform/workloads/email-connect/transactional | email-connect-runtime |
The activity-core four last refreshed on 2026-09-19 (so all four KV paths
existed then). Target Secrets are retained with their last synced values;
workloads are running on those.
- ESO's controller SA `external-secrets/external-secrets` may
`create serviceaccount --subresource=token` in `activity-core` and
`email-connect` (`kubectl auth can-i` = yes). No new RBAC is needed.
- ServiceAccounts `activity-core/activity-core-eso` and
`email-connect/email-connect-eso` do not exist yet.
- OpenBao's `auth/kubernetes/config` is written by
`scripts/openbao-apply-initial-config.sh` with `kubernetes_host` only, so it
reviews JWTs with the OpenBao pod's own projected token. No static reviewer
JWT that could expire.
## Declared in the repository (this commit)
- `openbao/policies/workload-kv-read-activity-core-eso.hcl`: the four data
paths above, lookup-self, revoke-self. Nothing else. (The older
`external-secrets-activity-core.hcl` also grants metadata reads; it stays for
its existing callers and is not attached to the new role.)
- `openbao/policies/workload-kv-read-email-connect-transactional-eso.hcl`: one
data path, lookup-self, revoke-self.
- `openbao/eso-auth-recovery/rpf-wp-0045-serviceaccounts.yaml`: the two SAs.
- `argocd/platform-addons/openbao-secretstore/openbao-{activity-core,email-connect}.clustersecretstore.yaml`:
target specs with Kubernetes auth. **These differ from live** until T03.
Nothing syncs them (excluded from `kustomization.yaml`; the railiance01
ArgoCD draft for `openbao-secretstore` is pinned to `d2dbc19`).
## T01 OpenBao roles and policies (founder, admin credential)
```task
id: RPF-WP-0045-T01
status: wait
priority: high
```
Waits on the founder's go-ahead. **The only step that needs the OpenBao
platform-admin credential**, one attended session, four writes plus
read-backs. Run from a checkout of `railiance-platform` at the commit that
holds this plan. No token appears on the command line; `warden access`
supplies it to the child process.
```sh
cd /home/worsch/railiance-platform
warden access openbao-platform-admin-login --exec -- sh -euc '
bao policy write workload-kv-read-activity-core-eso \
openbao/policies/workload-kv-read-activity-core-eso.hcl
bao policy write workload-kv-read-email-connect-transactional-eso \
openbao/policies/workload-kv-read-email-connect-transactional-eso.hcl
bao write auth/kubernetes/role/activity-core-eso \
bound_service_account_names=activity-core-eso \
bound_service_account_namespaces=activity-core \
audience=openbao token_policies=workload-kv-read-activity-core-eso \
token_ttl=15m token_max_ttl=15m token_explicit_max_ttl=15m \
token_no_default_policy=true
bao write auth/kubernetes/role/email-connect-transactional-eso \
bound_service_account_names=email-connect-eso \
bound_service_account_namespaces=email-connect \
audience=openbao token_policies=workload-kv-read-email-connect-transactional-eso \
token_ttl=15m token_max_ttl=15m token_explicit_max_ttl=15m \
token_no_default_policy=true
bao read auth/kubernetes/role/activity-core-eso
bao read auth/kubernetes/role/email-connect-transactional-eso
'
```
Before writing, confirm neither role nor policy exists
(`bao read auth/kubernetes/role/<name>` returns 404); if one does, stop and
compare rather than overwrite. Done when both read-backs show the fields above.
## T02 Confirm the declared manifests are the ones to apply
```task
id: RPF-WP-0045-T02
status: todo
priority: medium
```
Repository and read-only. `ssh railiance01 'kubectl diff -f -' < <file>` for the
SA file and both store manifests (client-side; nothing applied). Expected diff:
two new SAs, and in each store only `spec.provider.vault.auth` changes. Any
other difference stops T03.
## T03 Apply the ServiceAccounts and switch the two stores (founder go-ahead)
```task
id: RPF-WP-0045-T03
status: wait
priority: high
```
Waits on T01 and the founder's go-ahead. No OpenBao credential. Kubernetes
admin on railiance01.
```sh
cd /home/worsch/railiance-platform
ssh railiance01 'kubectl apply -f -' < openbao/eso-auth-recovery/rpf-wp-0045-serviceaccounts.yaml
ssh railiance01 'kubectl apply -f -' < argocd/platform-addons/openbao-secretstore/openbao-activity-core.clustersecretstore.yaml
ssh railiance01 'kubectl apply -f -' < argocd/platform-addons/openbao-secretstore/openbao-email-connect.clustersecretstore.yaml
```
Optional pre-check before the store switch, no admin credential (login with a
short-lived SA JWT; the output holds a 15-minute client token, so discard it):
```sh
ssh railiance01 'J=$(kubectl -n email-connect create token email-connect-eso --audience=openbao --duration=10m); kubectl -n openbao exec -i openbao-0 -- bao write -field=token_policies auth/kubernetes/login role=email-connect-transactional-eso jwt="$J"'
```
Expected output: `[workload-kv-read-email-connect-transactional-eso]`. Same for
`activity-core-eso` in `activity-core`. A wrong namespace or audience must fail.
## T04 Verify
```task
id: RPF-WP-0045-T04
status: wait
priority: high
```
Read-only after T03. Force a refresh, then check.
```sh
ssh railiance01 'kubectl annotate clustersecretstore openbao-activity-core openbao-email-connect force-sync=$(date +%s) --overwrite'
ssh railiance01 'kubectl -n activity-core annotate externalsecret --all force-sync=$(date +%s) --overwrite; kubectl -n email-connect annotate externalsecret email-connect-runtime force-sync=$(date +%s) --overwrite'
ssh railiance01 'kubectl wait --for=condition=Ready clustersecretstore/openbao-activity-core clustersecretstore/openbao-email-connect --timeout=60s'
ssh railiance01 'kubectl -n activity-core wait --for=condition=Ready externalsecret --all --timeout=90s; kubectl -n email-connect wait --for=condition=Ready externalsecret/email-connect-runtime --timeout=90s'
ssh railiance01 'kubectl get externalsecret -A -o custom-columns=NS:.metadata.namespace,NAME:.metadata.name,REASON:.status.conditions[0].reason,AT:.status.refreshTime | grep -v SecretSynced$'
```
(The annotate calls are writes to ESO objects; they belong to the T03 go-ahead.)
Done when both stores are Ready, the five ExternalSecrets show `SecretSynced`
with a refreshTime after T03, and the last command lists no ExternalSecret
other than the header. The earlier 37-count had 5 failing; expect 0.
## T05 Rollback (only if T03/T04 fails)
```task
id: RPF-WP-0045-T05
status: wait
priority: medium
```
The static tokens are already invalid, so rollback cannot restore sync; it
restores the prior, known-failing spec. Workload Secrets are unaffected either
way (ESO keeps the last synced values). Founder go-ahead:
```sh
ssh railiance01 'kubectl patch clustersecretstore openbao-activity-core --type=json -p "[{\"op\":\"replace\",\"path\":\"/spec/provider/vault/auth\",\"value\":{\"tokenSecretRef\":{\"name\":\"openbao-activity-core-eso-token\",\"namespace\":\"external-secrets\",\"key\":\"token\"}}}]"'
ssh railiance01 'kubectl patch clustersecretstore openbao-email-connect --type=json -p "[{\"op\":\"replace\",\"path\":\"/spec/provider/vault/auth\",\"value\":{\"tokenSecretRef\":{\"name\":\"openbao-email-connect-eso-token\",\"namespace\":\"external-secrets\",\"key\":\"token\"}}}]"'
ssh railiance01 'kubectl delete -f -' < openbao/eso-auth-recovery/rpf-wp-0045-serviceaccounts.yaml
```
OpenBao side (admin credential, only if the roles themselves are wrong):
`warden access openbao-platform-admin-login --exec -- sh -euc 'bao delete auth/kubernetes/role/activity-core-eso; bao delete auth/kubernetes/role/email-connect-transactional-eso; bao policy delete workload-kv-read-activity-core-eso; bao policy delete workload-kv-read-email-connect-transactional-eso'`.
Then `git revert` this plan's manifest commit. Correct forward in preference, as
RPF-WP-0037 does.
## T06 Retire the two static-token Secrets
```task
id: RPF-WP-0045-T06
status: wait
priority: medium
```
Waits on T04 passing for 24 hours (a few 1h refreshes) and the founder's
go-ahead. First prove nothing references them:
```sh
ssh railiance01 'kubectl get clustersecretstores,secretstores -A -o jsonpath="{range .items[*]}{.metadata.name} {.spec.provider.vault.auth.tokenSecretRef.name}{\"\n\"}{end}" | grep -E "openbao-(activity-core|email-connect)-eso-token"'
```
Expected: no output. Both tokens already fail `lookup-self` with 403, so no
revocation is needed; the founder may confirm invalidity with the RPF-WP-0037
`cleanup` pattern (the helper reads the value in memory only). Then delete:
```sh
ssh railiance01 'kubectl -n external-secrets delete secret openbao-activity-core-eso-token openbao-email-connect-eso-token'
```
Also retire the bootstrap targets in the owning repos:
`activity-core` `make openbao-eso-token-apply` / `scripts/openbao-eso-token-apply.sh`
and `email-connect/deploy/k8s/railiance/openbao-eso-token-apply.sh`, and update
`docs/argocd-gitops.md`. Their owners do that in their repos.
## Risks
- **Token values in Secret metadata.** Both token Secrets were created with
`kubectl apply`, so `kubectl.kubernetes.io/last-applied-configuration`
carries the token value. A metadata-only read shows it. These two tokens are
already invalid; T06 deletes them. Every other `*-eso-token` Secret in
`external-secrets` that was `kubectl apply`-ed (audit-core, core-hub,
canned-prompts, sbom-nexus, tenant-engine) is likely the same with **live**
tokens. Not inspected. Needs its own finding.
- Dependencies of the new method: OpenBao unsealed and reachable (same as
today), the Kubernetes auth mount's JWT review with OpenBao's own pod token
(auto-rotated, working for the three recovered stores and six more), and
ESO's TokenRequest right (present). Nothing found that expires.
- The forgejo-admin PAT and the offsite-lane values read here have their own
lifetimes (CCR-2026-0006, CCR-2026-0004). A later sync failure on those is a
value problem, not this auth problem.
- Why the static tokens died is not established (TTL expiry or revocation).
Other stores still on static tokens (audit-core, core-hub-database,
canned-prompts, sbom-nexus, tenant-engine) may fail the same way. Migrating
them is out of scope here.