terdut-operator/DESIGN.md

terdut-operator design

This is the design reference for implementing terdut-operator. It exists so implementation can start from settled decisions instead of re-litigating them mid-PR. It is deliberately more detailed than the README; the README stays as the short pitch and now points here instead of carrying open questions.

Written against terdut-server as of the Postgres-only, per-team-resources version (teams, escalation policies, dead man's switches and integrations are all rows scoped to a team, managed over internal/api/* — not env/config-file driven; see that repo's charts/terdut-server for the current deploy story this operator supersedes).

1. Goals & non-goals

Goal: let a terdut-server install — the server itself, its teams, their escalation policies, dead man's switches and alert-source integrations — be fully described as Kubernetes objects and managed through gitops, following controller-runtime / Kubebuilder conventions.

The operator creates and owns every TerdutServer it manages. It never adopts a pre-existing, independently-deployed terdut-server — whether deployed by hand or by charts/terdut-server. There is no migration path from an existing chart-based install, and none is planned (§10): starting with the operator means applying a fresh TerdutServer CR, not converting one. This is the root a few things downstream hang off of — notably §6's bootstrap flow, which only has to handle the operator bootstrapping a server it just created, never a server something else already bootstrapped first.

Non-goals (v1):

• Not a general-purpose Postgres operator. It consumes a database that either the Zalando postgres-operator or something else already provides.
• Not managing Alertmanager itself, or the routing rules that decide which alerts reach which integration webhook — only the terdut-server side (creating the integration and handing back its URL/key).
• Not OLM packaging. Plain Kubebuilder manifests + Helm chart for install, matching how terdut-server itself ships.
• Cross-namespace references are limited to exactly one edge: TerdutTeam.spec.serverRef may name a TerdutServer in a different namespace, gated by that TerdutServer's own spec.allowedTeams consent field (§4.1, §4.2) — this is the multi-tenant shape the operator exists for (one platform team owns a TerdutServer; other teams self-service a TerdutTeam against it without needing write access to the server's namespace). Every other reference (teamRef on the escalation rule/dead-man-switch/alert-source CRDs) stays same-namespace-as-its-TerdutTeam only, in v1 — those manage a specific team's own resources and are expected to live alongside it.
• No validating/mutating admission webhooks in v1. CEL validation rules on the CRDs (OpenAPI x-kubernetes-validations) cover what they can; anything that needs a live look at another object (e.g. "does this teamRef exist") is a status condition, not an admission rejection — keeps v1 to a controller-only deployment with no cert-manager/webhook dependency.

2. The README's open questions, resolved

How do team-crd connect with server-crd?

Explicit spec.serverRef: {name, namespace} on TerdutTeam — same reasoning as before (explicit, greppable, trivially validated), but namespace is deliberately part of the reference: one team can run and own a TerdutServer, and other teams — in their own namespaces, without any write access to the server-owning team's namespace — self-service a TerdutTeam against it. namespace defaults to the TerdutTeam's own namespace when omitted, so the common single-tenant case (serverRef: {name: terdut}) is unchanged.

This cross-namespace edge needs the target namespace's explicit consent — otherwise any namespace in the cluster could point a TerdutTeam at someone else's TerdutServer and have the operator provision a team on its behalf, which is a namespace-boundary violation, not a gitops convenience. (This consent gate is about which TerdutTeams the operator will act on, not about credential exposure — no terdut-server credential is ever placed in a TerdutTeam's own namespace regardless of this setting; see §6.) Kubernetes has two established patterns for this kind of consent, and Gateway API itself uses both, for two different relationships:

• ReferenceGrant (used for a Route reaching into an arbitrary Service/Secret): a separate object, living in the target namespace, enumerating exact {fromNamespace, fromKind} → {toKind, toName} pairs. No wildcard, no selector — every permitted namespace is spelled out.
• An inline field on the parent (used for a ListenerSet attaching to a shared Gateway, GA since Gateway API v1.5): the parent carries spec.allowedListeners.namespaces: {from: None|Same|All|Selector, selector} directly, no separate CRD.

TerdutTeam attaching to a shared TerdutServer is structurally the second case, not the first — a bounded set of expected children attaching to a parent they were deliberately made shareable, not an arbitrary backend reference — so this design follows the ListenerSet precedent: TerdutServer.spec.allowedTeams (§4.1), no extra CRD. §4.2 covers how TerdutTeam resolves against it. No other reference in this design (teamRef on the child CRDs) crosses a namespace boundary, so this is the only place cross-namespace consent is needed at all (§1, §5, §6, §9).

How does escalationrules, switches and alertsources connect to a team?

Explicit spec.teamRef: {name} on each of TerdutEscalationRule, TerdutDeadmanSwitch, TerdutAlertSource — same reasoning, and it mirrors terdut-server's own data model, where every one of these rows carries a team_id foreign key already. A matcher/selector on TerdutTeam would be inventing a second source of truth for an association the server already models as a plain reference.

Support for both postgres-operator (Zalando) and bring-your-own, how do we design that to be user friendly?

TerdutServer.spec.database is a oneOf, mirroring the chart's existing database.dsn / database.passwordSecret contract (see §8):

• dsn + passwordSecretRef — bring-your-own, exactly today's chart inputs.
• postgresClusterRef — points at a Zalando postgresql.acid.zalan.do CR; the operator derives the DSN and resolves the generated credentials Secret itself (see §8). CloudNativePG support is a natural follow-up using the same shape and is called out as deferred (§13), not designed in detail now.

3. API group, versions, CRD catalog

• Group: terdut.ryuvia.com, version: v1alpha1 (matches the ryuvia.com domain terdut-server already uses; bump to v1beta1/v1 per the normal Kubernetes API graduation criteria once the shapes below have proven stable against a real install).
• Module: git.ryuvia.com/niklas/terdut-operator, scaffolded with Kubebuilder (controller-runtime), matching terdut-server's Go toolchain and house style.

Kind	Scope	Purpose
TerdutServer	Namespaced	One terdut-server install: Deployment, Service, database wiring, bootstrap, operator credentials, cross-namespace team consent.
TerdutTeam	Namespaced	One team on a TerdutServer, possibly in another namespace: name, OIDC group mapping.
TerdutEscalationRule	Namespaced	A team's escalation policy (levels, targets, repeat).
TerdutDeadmanSwitch	Namespaced	One dead man's switch on a team.
TerdutAlertSource	Namespaced	One alert-ingest integration on a team (currently: Alertmanager webhook).

4. Per-CRD spec

4.1 TerdutServer

apiVersion: terdut.ryuvia.com/v1alpha1
kind: TerdutServer
metadata:
  name: terdut
  namespace: oncall
spec:
  image:
    repository: git.ryuvia.com/niklas/terdut-server
    tag: v0.9.3
  replicas: 1                      # terdut-server is not horizontally-scale-tested; keep the field, default 1
  networking:
    hostname: terdut.example.com
    servicePort: 8080
    gatewayListener: ""            # same semantics as chart's networking.listener
  database:
    dsn: "postgres://terdut@terdut-postgres:5432/terdut?sslmode=require"   # mutually exclusive with postgresClusterRef
    passwordSecretRef: {name: terdut.terdut-postgres.credentials.postgresql.acid.zalan.do, key: password}
    # --- OR ---
    postgresClusterRef: {name: terdut-postgres}   # Zalando `postgresql` CR in the same namespace
  sweeper:
    staleAfter: 6h
    archiveAfter: 168h
  deadman:
    matchers: "alertname=Watchdog"
    timeout: 15m
    severity: critical
  notify:
    ntfyURL: "http://ntfy.ntfy.svc.cluster.local"
    fallbackTopic: ""
    repeatEvery: 15m
    tokenSecretRef: {name: "", key: token}
  oidc:
    enabled: false
    issuer: ""
    clientID: ""
    clientSecretRef: {name: "", key: client-secret}
    name: SSO
    scopes: "openid profile email"
    allowedGroups: []
    adminGroup: ""
    sessionMaxAge: 12h
  passwordLogin: true
  # Consent for TerdutTeams in OTHER namespaces to set serverRef at this
  # TerdutServer. Same-namespace TerdutTeams never need this. Modeled on
  # Gateway API's Gateway.spec.allowedListeners.namespaces (the ListenerSet
  # attachment pattern, not ReferenceGrant — see §2 for why).
  allowedTeams:
    namespaces:
      from: None        # None (default) | Same | All | Selector
      # selector:       # required, and only meaningful, when from: Selector
      #   matchLabels:
      #     terdut.ryuvia.com/allowed: "true"
status:
  conditions: [...]                 # Ready, DatabaseReady, Bootstrapped
  observedGeneration: 3
  serviceName: terdut
  credentialsSecretRef: {name: terdut.platform-oncall-instance-credentials, key: token}  # see §6; pure output -- generated by the controller's own self-registration flow, always in the OPERATOR's namespace (always, implicitly -- not stored here, since it's never anything else), under a fixed key ("token").

Field-for-field this is the chart's values.yaml reshaped as a spec — not for migrating an existing chart-based install (§1: there is no such path), just because the shape is already familiar from the chart, and the operator absorbs what the chart's Deployment/Service/bootstrap-job templates used to do.

spec.database fields are +kubebuilder:validation:XValidation guarded to be mutually exclusive (dsn xor postgresClusterRef); mirrors the chart's "chart provisions no database" stance — this operator provisions no database either, only wires up one that exists.

spec.allowedTeams.namespaces.from defaults to None, matching allowedListeners's own default — a fresh TerdutServer accepts no cross-namespace TerdutTeam until its owner opts in, same-namespace TerdutTeams are unaffected either way. Selector deliberately has no per-name allowlist (no "and only these teams") — namespace-level consent is the right granularity here, same reasoning as ListenerSet: the namespace is the tenancy boundary, not the object.

4.2 TerdutTeam

spec:
  serverRef:
    name: terdut
    namespace: platform-oncall      # optional; defaults to this TerdutTeam's own namespace.
                                     # Cross-namespace requires that TerdutServer's spec.allowedTeams
                                     # (§4.1) to admit this namespace — otherwise Ready: False, reason: RefNotPermitted.
  displayName: "Platform"          # -> POST /api/teams {"name": ...}; server assigns the ID
  oidc:
    memberGroup: "terdut-platform-members"
    ownerGroup: "terdut-platform-owners"
status:
  conditions: [...]
  teamID: 42                       # the server-side ID; needed by every child object's controller
  credentialsSecretRef: {name: platform-oncall.platform-team-credentials, key: token}  # see §6; this team's own scoped key, always in the OPERATOR's own namespace (not stored here — same reasoning as TerdutServer's, §4.1)
  serverEndpoint: "http://terdut.platform-oncall.svc:8080"  # resolved once, here, from spec.serverRef -- see §5: this is what actually makes "a child never needs to chain up to TerdutServer" true, not just a stated intent. Set alongside teamID/credentialsSecretRef, same reconcile.
  observedGeneration: 1

Team membership (which users belong, team_members) is explicitly not modeled as a CRD field in v1: terdut-server already manages membership via OIDC group sync at login for SSO installs, and manual membership for password-login installs is a people-management action, not infrastructure — forcing it through gitops would mean a human's team change goes through a PR review. Flagged in §13 as revisitable if a real gitops-membership need shows up.

4.3 TerdutEscalationRule

spec:
  teamRef: {name: platform-team}
  repeatCount: 2
  fallbackTopic: "platform-oncall"
  levels:
    - timeout: 5m
      targets:
        - kind: oncall            # "oncall" or "user"
        - kind: user
          username: alice         # resolved to a user ID by the controller at apply time
    - timeout: 15m
      targets:
        - kind: oncall
status:
  conditions: [...]
  observedGeneration: 1

One TerdutEscalationRule per team — the server itself models a policy as one row (escalation_policies) with an owned list of levels, so a one-CRD-to-one-policy mapping (not one-CRD-per-level) matches the server's own aggregate and lets the whole thing be reconciled with the single PUT /api/teams/{teamID}/escalation the API actually exposes (see §5). Not enforced at admission (no webhooks in v1, §1) — two TerdutEscalationRules naming the same team would both PUT it and clobber each other every reconcile; a footgun worth this one sentence, not a technical guard.

A username target is resolved to the user_id PUT /api/teams/{teamID}/escalation actually requires (confirmed against source: escalationTargetJSON.UserID *int64, no username field at all) via GET /api/users — confirmed open to any authenticated caller, not gated by team membership or admin (internal/api/router.go's own comment: "readable by anyone signed in"), so the team-scoped credential this controller already holds is enough; no extra RBAC-equivalent server-side needed. Re-resolved every reconcile rather than cached, in case a username is renamed. An unresolvable username is Ready: False, reason: UnknownUser, naming which one.

4.4 TerdutDeadmanSwitch

spec:
  teamRef: {name: platform-team}
  name: "prod-watchdog"
  matcher: "alertname=Watchdog,cluster=prod"
  timeout: 15m
  severity: critical
status:
  conditions: [...]
  switchID: 7

No unique-name constraint server-side (§5's table row, confirmed against source) — the controller's own idempotent-create step is a GET-list and name match, not a conflict to recover from. name is optional, same as the API: left empty, terdut-server derives it from matcher's own canonical form, and that's what the lookup matches against too.

4.5 TerdutAlertSource

spec:
  teamRef: {name: platform-team}
  kind: alertmanager
  name: "prod-alertmanager"
status:
  conditions: [...]
  integrationID: 3
  webhookURLSecretRef: {name: prod-alertmanager-terdut-webhook}   # url + key, never in status/spec

The integration key is shown by the API exactly once, at creation (Integration.Key/URL in terdut-server's own model) — never re-readable, same shape as the bootstrap admin key. The controller writes it straight into a generated, owner-referenced Secret on the create it caused and never logs or stores it anywhere else; the CR's status carries only the Secret reference, matching how e.g. cert-manager's Certificate exposes spec.secretName rather than the key material itself.

4.6 Cross-namespace consent: TerdutServer.spec.allowedTeams

No separate CRD — the consent lives on TerdutServer itself (§4.1), following Gateway API's Gateway.spec.allowedListeners (ListenerSet attachment) rather than its ReferenceGrant, since a TerdutTeam attaching to a shared TerdutServer is the same shape of relationship: a bounded set of expected children attaching to a parent explicitly designed to be shared, not an arbitrary cross-namespace backend reference (see §2 for the full comparison of both patterns).

• from: None (default) — no cross-namespace TerdutTeam may resolve a serverRef into this TerdutServer. Same-namespace TerdutTeams are always allowed regardless of this field.
• from: Same — equivalent to None in effect (same-namespace is already unrestricted) but kept for parity with the upstream enum and to make the policy self-documenting in a diff.
• from: All — any namespace in the cluster may reference in. Appropriate for a genuinely shared, cluster-wide TerdutServer; the audit trail is "check allowedTeams plus who has RBAC to create a TerdutTeam anywhere," which is materially weaker than Selector.
• from: Selector — only namespaces matching spec.allowedTeams.namespaces.selector (a standard metav1.LabelSelector over Namespace objects, exactly like allowedListeners's own selector) may reference in. This is the recommended mode for the platform-team-owns-a-shared-server scenario this design targets: label the consuming namespaces once (e.g. terdut.ryuvia.com/allowed-server: platform-oncall/terdut) and new namespaces opt in by carrying the label, without editing the TerdutServer again.
• A TerdutTeam's controller re-evaluates allowedTeams on every reconcile before it will resolve a cross-namespace serverRef — for Selector, this means a Get on its own Namespace object plus reading the target TerdutServer's spec, not a List across the cluster. Same-namespace serverRef never consults this field at all.
• Narrowing or clearing allowedTeams (or unlabeling a namespace, under Selector) is a live revocation: the next reconcile of any TerdutTeam it used to authorize finds itself no longer permitted, flips Ready: False, reason: RefNotPermitted, and — deliberately — does not delete the team server-side on revocation alone; it stops reconciling further changes until access is restored or the TerdutTeam CR itself is deleted (whose finalizer still needs the credentials Secret described in §6 to clean up, so blocking new changes rather than forcing an immediate, possibly credential-less deletion is the safer failure mode).

5. Reconciliation semantics

terdut-server's REST surface (internal/api/router.go) does not give every resource a full update verb, so reconciliation strategy is per-resource:

Resource	Verbs available	Strategy
Team	POST create, PUT rename, DELETE, PUT oidc-groups, GET by name	Real update-in-place: diff spec vs. last-applied, PUT the changed pieces. GET /api/teams?name= (terdut-server's TEAM-LOOKUP.md, landed 2026-10-01) is what makes the idempotent-create general rule below actually true for Team — confirmed by checking: until that endpoint existed, an instance-scoped service account had no way to recover a team's id after a 409, unlike every other resource in this table, where the adopt-on-conflict rule had a real lookup to call.
Escalation policy	GET/PUT whole-policy	Update-in-place: PUT the full desired policy every reconcile that finds drift; cheap because whole-policy is small and already loaded whole server-side.
Dead man's switch	POST create, PUT update-in-place (added in terdut-server v0.33.0), DELETE, GET-list	Real update-in-place, same shape as Team/Escalation: PUT the whole switch every reconcile once its id is known. No unique-name constraint server-side (confirmed against source — handleCreateTeamDeadman has no conflict handling at all, unlike Team/service-account creation), so idempotent-create here can't rely on a 409 to adopt from: before POSTing, GET /api/teams/{teamID}/deadman/switches and match by name first: found → adopt its id; not found → POST. handleUpdateTeamDeadman's own doc comment (terdut-server) confirms the motivation directly: "Added alongside create/delete so an automated caller (terdut-operator) can reconcile a spec change without deleting and recreating the switch, which would otherwise ... needlessly rotate its id for no reason a reconciler's diff should ever manufacture." An earlier draft of this row, written before that endpoint existed, described delete-and-recreate; corrected here, confirmed against source rather than left stale.
Integration (alert source)	POST create, PATCH rename, DELETE	Rename via PATCH; any other spec change (kind) is delete-and-recreate, which rotates the webhook key — called out loudly in the CRD's field docs and in a Warning event, since it breaks whatever sends to the old URL/key until the new Secret is picked up. (See the general rule below for what happens if the Secret is lost with no spec change.)

General rules for every controller:

• Idempotent create: before POSTing, check status.<serverSideID> is unset; if the server already has a same-named object from a previous partial reconcile (e.g. after a crash between POST and status-write), treat a 409/name-conflict as "adopt" — GET-by-name and populate status, rather than erroring forever. terdut-server's list endpoints in each of these areas return objects by name, so this is a straightforward correlation.
• Periodic resync in addition to watch-triggered reconciles (Kubebuilder default RequeueAfter on success, e.g. every 5–10 minutes) to catch drift from someone changing state directly against the server's API/UI, since gitops correctness means the CR wins, not "first write wins".
• Finalizers on every CRD that has a server-side counterpart, so deletion calls the corresponding DELETE before the Kubernetes object disappears. Failure to delete server-side (e.g. server unreachable) blocks finalizer removal and surfaces as a Degraded condition + event, rather than silently orphaning a row.
• Generated Secrets holding unrecoverable server-issued material are watched, and their loss is fail-closed, not self-healed. Currently this is just TerdutAlertSource's webhook Secret (§4.5): the controller adds it to its Owns() watches, not just the CR. If it disappears while status.integrationID is still set, the controller does not attempt to recreate it — the key is genuinely gone (§4.5: never stored anywhere but that one Secret), so silently minting a replacement would rotate a live production webhook URL with no corresponding spec change to explain why. Instead it flips Ready: False, reason: WebhookSecretLost and fires a Warning event telling the operator to delete and recreate the TerdutAlertSource. No new mechanism is needed for recovery: deleting the CR runs the existing finalizer (DELETE the still-live integration server-side, above), and recreating it runs the existing idempotent-create path (this same section) — a fresh POST, a new key, a new Secret. This is deliberately the same recovery motion as the kind-change rotation above, just human-triggered instead of spec-triggered.
• Owner chain for status resolution, not API calls: TerdutTeam's controller does not call any other controller; every child CRD's controller independently resolves its own teamRef → TerdutTeam.status for the teamID, credentialsSecretRef, and serverEndpoint it needs to call the API (§6) — it never needs to chain further up to TerdutServer at all, not just as a stated intent but literally: serverEndpoint is resolved once, by TerdutTeam's own controller, and stored in its status specifically so no child ever needs its own TerdutServer RBAC (get/ list/watch on terdutservers) to find out where to send a request — the team's own scoped credential plus that one status field is everything a child resource's controller requires. If the referenced TerdutTeam isn't Ready yet (which includes not having a credentialsSecretRef set), the child requeues with backoff and reports Ready: False, reason: WaitingForTeam — no cross-controller RPC.
• Cross-namespace serverRef is re-checked every reconcile, not just at creation: TerdutTeam's controller reads the target TerdutServer's spec.allowedTeams (and, under Selector, a Get on its own Namespace object for labels) on every pass before touching a cross-namespace TerdutServer — revocation (§4.6) takes effect on the team's very next reconcile, not just when the CR is first applied.

6. Bootstrap & authentication to terdut-server's API

terdut-server's scoped service-account credential type (terdut-server's SERVICE-ACCOUNTS.md) has shipped — confirmed against source: internal/api/service_accounts.go, migration 014_service_accounts.sql, and internal/api/router.go wiring it in under AuthMiddleware. This section is no longer blocked on it; the "v1-blocking" framing here was accurate when this section was first written and is stale now.

GET /api/service-accounts?name= is not an unauthenticated lookup — confirmed against internal/api/middleware.go's AuthMiddleware, which hard-rejects any request carrying neither a Bearer token nor a session cookie with 401 before any handler ever runs. This would matter a great deal if the operator's /api/bootstrap call could ever lose a race to something else bootstrapping the same server first — a credential-less loser would have no authenticated way to recover. It doesn't matter here, by construction (§1): the operator only ever calls /api/bootstrap against a TerdutServer it just created, so there is nothing else in a position to race it. An earlier draft of this section added a spec.credentialsSecretRef bring-your-own input specifically to work around that race, for a world where the operator might adopt a server something else had already bootstrapped. That world doesn't exist (§1), so the field was removed rather than kept as unused flexibility — self-registration (point 1, below) is simply the only path, not one of two.

Every credential the operator holds — the one instance-scoped key per TerdutServer, and one team-scoped key per TerdutTeam — lives in a Secret in the operator's own namespace, never in the namespace of the CR it authenticates for. Reconciliation happens entirely inside the operator's controller loop, which is a single Deployment/ServiceAccount already watching every namespace it's granted (§9); nothing about calling terdut-server's API on a CR's behalf requires the credential to be physically located near that CR, and no CR owner (human or otherwise) ever needs to see, hold, or have RBAC to read a terdut-server credential. This is a straight simplification of an earlier draft of this section, which mirrored a shared credential into each consenting namespace instead — that version conflated "the CR's owner never needs to see this" (true, and preserved here) with "so the credential must live in the CR's namespace" (a non-sequitur once you don't need to grant anyone else namespace-local read access). Dropping that assumption also removes an entire class of complexity: no on-demand mirroring, no garbage-collecting an orphaned copy when allowedTeams narrows, no "OwnerReferences can't cross namespaces so track it in status instead" workaround — none of that machinery is needed when nothing ever crosses into a tenant namespace in the first place.

1. First reconcile, confirmed against source (internal/api/users.go's handleBootstrap): /api/bootstrap is single-shot per install, gated on SELECT COUNT(*) FROM users — once non-zero, every call 403s regardless of identity. There are two crash windows between "get an admin key" and "have a lasting, usable credential" — getting from /api/bootstrap's key to a minted service-account key, and getting from that key to a persisted Secret — and both get a checkpoint rather than being left as a theoretical gap, the same rigor §5's general idempotent-create rule already applies elsewhere:
   • status.credentialsSecretRef already set: done, nothing to do.
   • Otherwise, check for an intermediate <namespace>.<name>-bootstrap-admin Secret in the operator's own namespace first. If it exists, its key is a still- valid admin credential from an earlier, interrupted attempt — skip /api/bootstrap entirely and reuse it. If not, call /api/bootstrap once the Deployment this TerdutServer created has a ready replica; on 201, immediately checkpoint its response's raw admin key ({"user": ..., "api_key": {"key": "<raw>", ...}}) into that Secret before doing anything else with it. A 403 with neither status.credentialsSecretRef nor this checkpoint Secret present is the one genuinely pathological case left (the checkpoint deleted out from under a reconcile already past this point) — handled the same way the design already handles unrecoverable server-issued material elsewhere (§5's webhook-Secret-loss rule): fail closed, Ready: False, reason: BootstrapStateLost, with the same recovery as that case, delete and recreate the TerdutServer (its finalizer tears down the Deployment/database-backing and server-side rows; a fresh create starts clean) — not a workaround peculiar to this one path.
   • With an admin key in hand (fresh or checkpointed): POST /api/service-accounts {name: "terdut-operator", scope: "instance"}. A 409 here means a prior attempt got this far before being interrupted — adopt rather than error, per §5's general rule: GET /api/service-accounts?name=terdut-operator (authenticated with the checkpointed admin key, not an unauthenticated lookup) to find its id, then POST /api/service-accounts/{id}/keys to mint a fresh key — an orphaned first key some interrupted attempt minted and never used is inert, not a cleanup obligation.
2. The resulting instance-scoped key — not the checkpointed admin key, which is deleted once this step succeeds — is written to a generated Secret in the operator's own namespace (e.g. <serverRef.namespace>.<serverRef.name>-instance-credentials, under a fixed data key, token), referenced back from TerdutServer.status.credentialsSecretRef: {name, key} (§4.1). No OwnerReference (those can't cross namespaces, and this Secret doesn't share a namespace with the TerdutServer that caused it); the TerdutServer's finalizer deletes this Secret directly as part of its own teardown, the same way it already has to clean up the server-side resources it created (§5's general finalizer rule extends naturally to this Secret).
3. When a TerdutTeam first becomes Ready (its serverRef resolved, allowedTeams satisfied if cross-namespace), its controller uses the TerdutServer's instance-scoped credential (read from the operator's own namespace, resolved via the owner chain in §5) to mint a team-scoped service account for itself: POST /api/service-accounts with scope: team, teamID: <status.teamID>. The resulting key is written to its own generated Secret, again in the operator's own namespace (e.g. <teamNamespace>.<teamName>-team-credentials), referenced from TerdutTeam.status.credentialsSecretRef (§4.2). Same finalizer pattern as point 2: the TerdutTeam's finalizer deletes this Secret as part of its own teardown.
4. Every child controller (TerdutEscalationRule, TerdutDeadmanSwitch, TerdutAlertSource) reads its team's credentialsSecretRef — resolved through its teamRef → TerdutTeam.status (§5) — and never touches the instance-scoped credential at all. Since child CRDs stay same-namespace- as-their-TerdutTeam in v1 (§1), and the credential itself lives in the operator's namespace regardless of where the TerdutTeam or its children are, this works identically whether the TerdutTeam is same-namespace or cross-namespace relative to its TerdutServer — there is no separate cross-namespace case to handle here at all, unlike the mirroring design this replaced.
5. Blast radius: a team-scoped key can only touch its own team_id's escalation policy, dead-man switches, integrations, schedule and OIDC group bindings server-side (enforced by terdut-server itself, per SERVICE-ACCOUNTS.md) — compromising one such Secret (e.g. a bug that leaks operator-namespace Secrets, or an overly broad RBAC grant on that one namespace) exposes exactly one team, never the whole server. This is the real fix for what an earlier draft of this section called out as its weak point (every mirrored copy being server-admin-equivalent); it falls out of team-scoped credentials existing at all, independent of where they're stored — the operator-private storage described above closes the RBAC-footprint half of the problem, team scoping closes the credential- privilege half.
6. Rotation: POST /api/service-accounts/{id}/keys mints a new key on the existing account without recreating it; the old key is revoked via DELETE /api/service-accounts/{id}/keys/{keyID}; the operator's local Secret is updated in place. No DB-level workaround, no re-triggering a single-shot endpoint that can't fire twice (which is what made rotation unworkable under the old /api/bootstrap-only design).

7. Ownership, status, garbage collection

• Every generated object that lives in the same namespace as the CR that caused it (Deployment, Service, webhook Secret) carries a standard metav1.OwnerReference — GC handles these, no finalizer needed. The two credential Secrets from §6 are the one exception: they live in the operator's own namespace regardless of where their owning CR lives, so OwnerReference doesn't apply (cross-namespace) and cleanup instead runs through that CR's finalizer directly, alongside the server-side DELETE it already has to issue (§5).
• Status conditions follow the standard metav1.Condition shape with at least Ready on every kind, plus kind-specific ones (TerdutServer: DatabaseReady, Bootstrapped; children: Synced).
• status.observedGeneration on every kind, bumped only after a successful reconcile against that generation's spec — the standard way a client (or kubectl wait) tells "applied" from "seen".
• No cluster-scoped aggregation object (e.g. no cluster-wide "all servers" status) in v1 — kubectl get terdutservers -A is the aggregate view.

8. Postgres integration

Mirrors the chart's existing two-path contract (values.yaml database.dsn/passwordSecret), because that contract is already documented and tested operationally:

• Bring-your-own: spec.database.dsn (no password) + spec.database.passwordSecretRef — the operator sets PGPASSWORD on the Deployment's container env from that Secret, exactly like the chart does today, and does nothing else. No connectivity check beyond what the Deployment's own readiness probe already gives.
• Zalando postgres-operator: spec.database.postgresClusterRef names a postgresql.acid.zalan.do CR in the same namespace. The controller:
  • Reads that CR's status for the primary Service name/port to build the DSN host — <cluster>.<namespace>.svc:5432 — and database name convention.
  • Resolves the generated credentials Secret (<user>.<cluster>.credentials.postgresql.acid.zalan.do) the same way the chart's comment already documents, and wires it in as PGPASSWORD the same way.
  • Watches that Secret (not just the postgresql CR) so a credential rotation triggers a requeue — the chart today requires a manual pod restart for this; the operator can at least detect and report it via a condition even if restarting on rotation is left as a §13 follow-up rather than done automatically (a rolling restart on credential change is a behavior change worth its own design pass, not folded in here).
  • Requires read RBAC on postgresql.acid.zalan.do (optional CRD — the operator's ClusterRole/Role should not hard-fail if the CRD isn't installed and a given TerdutServer uses BYO DSN instead).

9. RBAC

• The operator's own ServiceAccount needs, per namespace it's granted: get/list/watch/create/update/patch/delete on Deployments, Services it owns, and get/list/watch on postgresql.acid.zalan.do (optional, degrade gracefully if absent per §8), plus cluster-wide get/list on Namespace (labels only, for allowedTeams: {from: Selector} evaluation — §4.1, §4.6).
• Two different Secret scopes, not one — corrected from an earlier draft of this section. That earlier draft said Secret access was "scoped to the operator's own namespace only... nowhere else," reasoning that with every credential held privately in the operator's own namespace (§6) there was no legitimate reason to touch a Secret anywhere else. That was wrong once §4.5 existed:
  • The §6 credential Secrets (one instance-scoped key per TerdutServer, one team-scoped key per TerdutTeam) do live in, and are only ever touched from, the operator's own namespace — get/list/watch/create/update/patch/delete there, nowhere else. The rest of the original reasoning stands for these Secrets specifically: no human or team's own RBAC is ever granted access to a terdut-server credential by this design, and the operator itself never needs cross-namespace access to reach them.
  • The §4.5 webhook Secret is different: it's owned by and lives beside its TerdutAlertSource, in that CR's own tenant namespace, not the operator's. The per-namespace Role already granted for Deployments/Services in each watched namespace (below) must carry the same Secret verbs there too, or the controller cannot create, watch, or even detect the loss of (§5) that Secret at all.
  • This necessarily widens the operator's footprint in each watched tenant namespace to "any Secret in that namespace," not just the ones it created — Kubernetes RBAC has no owner-scoped grant finer than the namespace itself, and the design already accepts this same granularity for Deployments/Services there. Flagged as an accepted trade-off, not a silent gap (§13).
• No cluster-scoped resources are created by this operator (namespaced CRDs only, per §1) — a Role + RoleBinding per watched (tenant) namespace is sufficient for Deployments/Services/the webhook Secret/the optional Zalando CRD, plus a separate Role + RoleBinding in the operator's own namespace for the §6 credential Secrets; a ClusterRole is only needed for watching CRDs across all namespaces (the normal Kubebuilder multi-tenant-operator default) — none of the Secret access above needs to be cluster-scoped.
• terdut-server's own RBAC is unaffected — the operator talks to it purely over HTTP with service-account API keys (§6), never via the Kubernetes API for app-level state.

10. Relationship to charts/terdut-server

The chart's Deployment/Service/bootstrap-job templates are redundant once TerdutServer exists — running both would mean two controllers (Helm and this operator) reconciling the same Deployment, which is exactly the conflict Kubernetes operators exist to avoid. The chart is repurposed into an installer chart: it installs the operator + CRDs (and optionally one TerdutServer CR from values.yaml, for users who want "helm install and get a server" without hand-writing a CR) rather than templating the Deployment directly.

No migration path from an existing chart-based install, and none is planned (§1). An earlier draft of this section spent most of its length on one — helm template the current release's values.yaml into an equivalent TerdutServer CR, uninstall or shrink the old release, and a whole sub-question about who gets to call /api/bootstrap first, the chart's Job or the operator — all of which presupposed the operator might end up managing a server the chart had already deployed and bootstrapped. §1 rules that out: the operator only ever manages servers it created itself, so there's nothing to migrate and no bootstrap race to settle (§6 covers why that race doesn't exist either). Adopting the operator means applying a fresh TerdutServer CR; whatever the chart deployed before stays exactly what it was, a separate install, until someone deletes it.

11. Testing strategy

• envtest (controller-runtime's fake API server) for every controller's reconcile logic against the Kubernetes side.
• terdut-server's REST API is faked with a small httptest.Server per controller test driven by fixtures matching the real handlers' request/response shapes (already well-documented in internal/api/*_test.go on the server side) — no real Postgres or real terdut-server binary needed for controller unit tests.
• A smaller number of true end-to-end tests (kind cluster + real terdut-server image + real Postgres) covering the golden path per CRD: create TerdutServer → TerdutTeam → one of each child kind → verify via terdut-server's own API that the objects exist with the right shape → delete the CR → verify the server-side object is gone.

12. Observability

• Standard controller-runtime metrics (reconcile duration/error counts) are enough for v1 — no custom metrics.
• Every externally-visible action (bootstrap, key rotation-needed, delete- and-recreate on the no-PUT resources, adopt-on-conflict) emits a Kubernetes Event on the CR, since that's what shows up in kubectl describe and gitops tooling (Argo CD/Flux) surfaces without extra wiring.

13. Deferred / explicitly out of scope for this design

• A real scoped service-account/token type in terdut-server — not merely deferred, this is v1-blocking for §6 as written (verified: without it, §6's bootstrap flow has no working credential-rotation path and no clean answer to the chart-vs-operator bootstrap race; see §6 points 1, 5, 6 and §10). Sequence this server-side change before implementing the TerdutServer controller's bootstrap logic, not after.
• A version-discovery endpoint on terdut-server (e.g. GET /api/version). Neither this operator nor terdut-tui has one today — both independently detect capability by probing specific routes (terdut-tui via GET /api/teams 404-checking; this operator would otherwise need to invent its own equivalent probe). An unattended reconciler is more exposed to a silent breaking API change than an interactive TUI a human is watching; raising this alongside the service-account request rather than inventing another route-probe here.
• CloudNativePG support — same spec.database shape as Zalando should extend to it, but the concrete field/Secret-naming conventions need their own look.
• Cross-namespace teamRef on the child CRDs (TerdutEscalationRule, TerdutDeadmanSwitch, TerdutAlertSource) — only TerdutTeam.serverRef crosses namespaces in v1 (§2, §4.2, §4.6); these stay same-namespace as their TerdutTeam until a real need for splitting them out shows up.
• Narrower-than-namespace RBAC for the §4.5 webhook Secret (Kubernetes RBAC has no owner-scoped grant below the namespace itself, per §9) — revisit if the widened per-tenant-namespace Secret access proves too broad in practice.
• Gitops-managed team membership (see §4.2).
• Automatic Deployment restart on upstream Postgres credential rotation.
• Admission webhooks / CEL-only validation limits (e.g. verifying a teamRef exists at admission time rather than surfacing it as a status condition after the fact).
• OLM packaging.