Some checks failed
Run Check Script / check (pull_request) Failing after 51s
The env var name was a misnomer — ConfigClient resolves both config and secrets, not just secrets. The struct field was already config_namespace. Legacy SecretManager keeps the old var; this forces migration to ConfigClient for new code.
5.4 KiB
5.4 KiB
Fleet Platform v0.3 — Staging to production-ready
Written 2026-05-31. Picks up after OpenBao + Zitadel + NATS + callout + operator are deployed and functional on staging (2-3 weeks old versions).
Current state
- OpenBao running at
secrets-stg.cb1.nationtech.io - Zitadel running at
sso-stg.cb1.nationtech.io - NATS + auth callout deployed in
fleet-stagingnamespace - Operator deployed (older version, 2-3 weeks old)
- Config-driven OpenBao installer (
examples/openbao) harmony-fleet-deploybinary readsFleetDeployConfig+FleetDeploySecretsfrom OpenBao
Immediate next steps
1. Provision operator credentials in OpenBao
- Fetch existing creds from the running cluster:
oc -n fleet-staging get secret harmony-fleet-operator-secrets -o jsonpath='{.data.credentials\.toml}' | base64 -d - Seed into OpenBao at
secret/data/fleet-staging/FleetDeploySecrets:export VAULT_ADDR=https://secrets-stg.cb1.nationtech.io export VAULT_TOKEN=<root token> oc -n fleet-staging get secret harmony-fleet-operator-secrets -o jsonpath='{.data.credentials\.toml}' | base64 -d \ | jq -Rs '{value: ({operator_credentials_toml: .} | tojson)}' \ | bao kv put secret/fleet-staging/FleetDeploySecrets - - Verify the secret is readable:
bao kv get secret/fleet-staging/FleetDeploySecrets
2. Private repo deploy script
- Create
.envrcwith minimal env:export OPENBAO_URL=https://secrets-stg.cb1.nationtech.io export HARMONY_CONFIG_NAMESPACE=fleet-staging # export OPENBAO_TOKEN=<root token for now; SSO later> - Write deploy invocation (shell script or just
harmony-fleet-deploycall):harmony-fleet-deploy --from-tag harmony-fleet-operator-vX.Y.Z --yes - Commit
.envrc+ script to private repo (shared with teammates)
3. Execute operator upgrade
- Run the deploy script from the private repo
- Verify operator pod starts and connects to NATS
- Verify operator reconciles existing CRs (check logs)
- Confirm no regression in existing fleet functionality
4. Operator UI ingress (trivial)
- Expose operator UI with TLS ingress on
fleet-stg.<base_domain> - Verify the UI loads and serves the SPA
- Confirm no auth gate yet (SSO is next)
5. SSO login flow
- Wire operator UI to Zitadel SSO at
sso-stg.<base_domain> - Test login/logout flow end-to-end
- Verify session persistence across page reloads
- Confirm RBAC: only authorized Zitadel users can access the UI
6. Real data in UI
- Replace mock device list with live
device-infoKV data - Replace mock deployment list with live
DeploymentCR data - Wire per-device drilldown to real
DeviceInfo+ last-heartbeat + agent version - NATS tail panel: SSE stream of
device-infoanddevice-stateupdates (plain text) - Verify data refreshes without manual reload
Configuration model
Environment (minimal, committed in private repo)
OPENBAO_URL=https://secrets-stg.cb1.nationtech.io
HARMONY_CONFIG_NAMESPACE=fleet-staging
# SSO auth or root token (SSO is the goal)
OpenBao (read via ConfigClient)
FleetDeployConfig(k8s namespaces, NATS URL, chart coords) atsecret/data/fleet-staging/FleetDeployConfigFleetDeploySecrets(operator creds) atsecret/data/fleet-staging/FleetDeploySecrets
Missing features (post-UI)
Auth & credentials
- Per-device OpenBao policies (templated policies, one role per device type)
- Device identity claim in JWT (Zitadel
client_idwithdevice-prefix) - OpenBao JWT auth role granularity (extend
OpenbaoJwtAuthto list of roles) - Move k8s namespaces + chart coords into
ConfigClientconfig struct (env = only identifier + auth)
Operator capabilities
- Agent upgrade path (ADR-022 exists; implementation pending)
- Device enrollment flow (operator-facing runbook)
- Revoke device / rotate key operations
- Fleet-wide rollout strategies (canary, %-based) on top of agent-upgrade primitive
Observability
- Operator logs every CR it acquires (verify output reads well)
- NATS debugging one-liners in hand-off menu
- Journald log streaming (currently only
.status.aggregate.lastError) - Metrics dashboard (deferred until >100 devices)
Quality & hardening
- Agent config-driven labels (
[labels]in agent toml → DeviceInfo) matchExpressionsin selectors (currentlymatchLabelsonly)Device.status.conditionspopulated from heartbeat staleness- Operator graceful degradation on bad device_id (log + skip, don't restart-loop)
- Persist
nats_auth_passand issuer NKey viaharmony_secret(regenerate-every-run footgun)
Refactors (deferred, non-blocking)
- Decompose
FleetServerScoreinto independent, ConfigClient-glued Scores - Move
harmony/modules/fleet/→fleet/harmony-fleet/(ADR-021 pending) - Delete
examples/fleet_staging_deploy(superseded byfleet_staging_install) - Drop
K8sAnywhereTopologyfor ad-hoc Score execution; introduceK8sBareTopology
Principles (carried forward)
- No yaml in framework code paths
- Scores describe desired state; topologies expose capabilities
- Cross-boundary wire types in
harmony-reconciler-contracts - Never ship untested code
- Prove claims about upstream before blaming upstream
- Design the brick before moving the brick