Suspendedremoval

src/app/api/tenants/[name]/resume-request/route.ts

@@ -0,0 +1,154 @@
+import { NextRequest, NextResponse } from "next/server";
+import { getSessionUser, canMutate } from "@/lib/session";
+import { getTenant, setTenantAnnotation } from "@/lib/k8s";
+import { canUserSeeTenant } from "@/lib/visibility";
+import {
+  createResumeRequest,
+  getPendingResumeRequestForTenant,
+  getTenantRequestByTenantName,
+} from "@/lib/db";
+import { safeError } from "@/lib/errors";
+
+/**
+ * POST /api/tenants/[name]/resume-request
+ *
+ * Owner-initiated request to reactivate a suspended tenant (Bug 37a).
+ * Creates a pending tenant_request of type 'resume' for admin review,
+ * and stamps the PiecedTenant CR with an annotation that pauses the
+ * operator's 60-day deletion timer.
+ *
+ * Why a request flow at all
+ * -------------------------
+ * Customers can self-serve cancel; resume requires admin oversight.
+ * Reactivation may involve re-validating billing, confirming the
+ * customer still wants to be active, or other manual steps. The
+ * request flow gives admins a queue to review, with the same approve/
+ * reject UX as initial provision requests.
+ *
+ * Authorization
+ * -------------
+ * Owners and platform admins. Platform admins shouldn't normally use
+ * this endpoint — they have direct PATCH suspend access — but it's
+ * permissive in case admin tooling pivots.
+ *
+ * Validation
+ * ----------
+ *  - Tenant must exist and be visible to the caller.
+ *  - Tenant must be currently suspended. Resuming an active tenant
+ *    is meaningless.
+ *  - At most one pending resume request per tenant. Enforced by the
+ *    DB's partial unique index, but we also check explicitly here to
+ *    return a friendly 409 instead of a 500.
+ *
+ * Side effects on success
+ * -----------------------
+ *  - INSERT into tenant_requests (request_type='resume', status='pending')
+ *  - PATCH annotation `pieced.ch/resume-request-pending=<request-id>` on
+ *    the CR. This is the operator's signal to pause its 60-day deletion
+ *    timer until the request transitions to terminal.
+ *
+ * The annotation set is best-effort: if the K8s PATCH fails after the
+ * DB insert, the row exists without the annotation. The customer
+ * sees the request as pending; admin can still approve. The only
+ * functional consequence is the 60-day timer doesn't pause until the
+ * next request transition, which is fine in practice (admin response
+ * times are dramatically shorter than 60 days).
+ */
+export async function POST(
+  req: NextRequest,
+  { params }: { params: Promise<{ name: string }> }
+) {
+  const user = await getSessionUser();
+  if (!user) {
+    return NextResponse.json({ error: "Unauthorized" }, { status: 401 });
+  }
+  if (!canMutate(user)) {
+    return NextResponse.json({ error: "Forbidden" }, { status: 403 });
+  }
+
+  const { name } = await params;
+  const tenant = await getTenant(name);
+  if (!tenant) {
+    return NextResponse.json({ error: "Not found" }, { status: 404 });
+  }
+  if (!(await canUserSeeTenant(user, tenant))) {
+    return NextResponse.json({ error: "Not found" }, { status: 404 });
+  }
+
+  if (!tenant.spec.suspend) {
+    return NextResponse.json(
+      { error: "Tenant is not suspended; nothing to resume." },
+      { status: 409 }
+    );
+  }
+
+  // Already a pending request? Don't duplicate.
+  const existing = await getPendingResumeRequestForTenant(name);
+  if (existing) {
+    return NextResponse.json(
+      {
+        error: "A resume request for this tenant is already pending.",
+        request: { id: existing.id, createdAt: existing.createdAt },
+      },
+      { status: 409 }
+    );
+  }
+
+  // Pull traceability fields (companyName, agentName) from the original
+  // provision request. The schema marks these NOT NULL, so we have to
+  // populate them; copying from the provision row keeps the resume
+  // row navigable in the admin UI without making up values.
+  const provision = await getTenantRequestByTenantName(name);
+
+  try {
+    const resumeRequest = await createResumeRequest({
+      tenantName: name,
+      zitadelOrgId: tenant.metadata.labels?.[
+        "pieced.ch/zitadel-org-id"
+      ] ?? user.zitadelOrgId,
+      zitadelUserId: user.id,
+      contactName: user.name ?? user.email ?? "Unknown",
+      contactEmail: user.email ?? "unknown@example.invalid",
+      companyName: provision?.companyName ?? tenant.spec.displayName ?? name,
+      agentName: provision?.agentName ?? "Assistant",
+    });
+
+    // Stamp the annotation so the operator pauses its TTL. If this
+    // fails the request still exists; surface the error so admin
+    // tooling can re-stamp if needed, but don't roll back.
+    try {
+      await setTenantAnnotation(
+        name,
+        "pieced.ch/resume-request-pending",
+        resumeRequest.id
+      );
+    } catch (e) {
+      console.warn(
+        "resume request created but annotation could not be set; operator's 60-day timer will not pause until next reconcile triggered by request transition",
+        e
+      );
+    }
+
+    return NextResponse.json(
+      {
+        message: "Resume request submitted. An admin will review shortly.",
+        request: { id: resumeRequest.id, status: resumeRequest.status },
+      },
+      { status: 201 }
+    );
+  } catch (e: any) {
+    // Unique violation (a pending row already exists for this tenant)
+    // is friendly-handled above; this catches everything else.
+    if (e.code === "23505") {
+      return NextResponse.json(
+        { error: "A resume request for this tenant is already pending." },
+        { status: 409 }
+      );
+    }
+    console.error("Resume request creation failed:", e);
+    return NextResponse.json(
+      { error: safeError(e, "Failed to submit resume request") },
+      { status: 500 }
+    );
+  }
+}

src/app/api/tenants/[name]/suspend/route.ts

@@ -1,7 +1,7 @@
 import { NextRequest, NextResponse } from "next/server";
 import { z } from "zod";
 import { getSessionUser, canMutate } from "@/lib/session";
-import { getTenant, patchTenantSpec } from "@/lib/k8s";
+import { getTenant, patchTenantSpec, setTenantAnnotation } from "@/lib/k8s";
 import { canUserSeeTenant } from "@/lib/visibility";
 import { safeError } from "@/lib/errors";
 
@@ -12,37 +12,38 @@ const patchSchema = z.object({
 /**
  * PATCH /api/tenants/[name]/suspend
  *
- * Customer-side "Cancel subscription" / "Resume" toggle (Bug 31).
+ * Direct suspend control on the PiecedTenant CR. Sets `spec.suspend`
+ * to true (cancel) or false (resume).
  *
- * Sets `spec.suspend` on the PiecedTenant CR. The operator interprets
- * this flag as "stop reconciling this tenant" — workloads, packages,
- * and channel-user changes are no longer applied. Existing data is
- * preserved (namespace, ConfigMaps, OpenBao secrets, CNPG database,
- * billing records). Resuming sets the flag back to false and the
- * operator picks up reconciliation on the next loop.
+ * Authorization (Bug 37a)
+ * -----------------------
+ * - suspend=true  → owners and platform admins may call.
+ * - suspend=false → platform admins ONLY. Owners must go through the
+ *   resume-request flow (POST /api/tenants/[name]/resume-request),
+ *   which creates a pending request for admin approval. This
+ *   asymmetry is by design: cancellation is self-service (low risk;
+ *   reversible by request); reactivation requires admin oversight
+ *   (e.g. to re-validate billing, confirm intent).
  *
- * Authorization
- * -------------
- * - Customer-side: only an `owner` of the tenant's org may call this.
- *   `canMutate` is the right gate (mirrors the rest of the customer
- *   API surface). User-role members cannot cancel a subscription.
- * - Platform staff: allowed via `canMutate`'s isPlatform branch, but
- *   in practice they should use admin tooling for this — the action
- *   is exposed here for the customer's benefit.
+ * Customer flow:
+ *   - Cancel:  PATCH suspend=true  here
+ *   - Resume:  POST /resume-request — creates a 'resume' tenant_request,
+ *              admin approves via /api/admin/requests/[id]/approve which
+ *              then PATCHes suspend=false here as a platform user.
  *
- * Visibility check is via `canUserSeeTenant` — same notFound() trick
- * as the detail page, so we don't leak existence of tenants the
- * caller can't see.
+ * Workload behaviour
+ * ------------------
+ * On suspend=true the operator deletes the OpenClawInstance, stopping
+ * the pod within seconds. Tenant data — namespace, ConfigMaps,
+ * OpenBao secrets, CNPG database, LiteLLM team — is retained.
  *
- * Note on workload teardown
- * -------------------------
- * As of this writing, the operator's `suspend` handling is "skip
- * reconciliation and set status.phase to Suspended". The underlying
- * StatefulSet keeps running until next reconciliation, which won't
- * happen while suspended. Group D will add scale-to-zero so cancelled
- * subscriptions actually stop incurring compute. Until then, an
- * operator following up with a `kubectl scale` is the workaround.
- * Customer data is preserved either way.
+ * Suspended tenants enter a 60-day retention window (operator
+ * constant `retentionAfterSuspend`); after that, the tenant is fully
+ * deleted unless a pending resume request exists. The operator
+ * checks the `pieced.ch/resume-request-pending` annotation to know
+ * about pending requests; we set it here when admin approves the
+ * resume (transitively, via the admin-approve endpoint), and clear
+ * it when the request reaches a terminal state.
  */
 export async function PATCH(
   req: NextRequest,
@@ -76,6 +77,18 @@ export async function PATCH(
   }
   const { suspend } = parsed.data;
 
+  // Bug 37a: resume (suspend=false) is platform-admin only via this
+  // endpoint. Owners must go through the resume-request flow.
+  if (!suspend && !user.isPlatform) {
+    return NextResponse.json(
+      {
+        error:
+          "Resume requires platform-admin approval. Submit a resume request via /api/tenants/[name]/resume-request.",
+      },
+      { status: 403 }
+    );
+  }
+
   // No-op early exit. Avoids a needless K8s patch + status churn when
   // the user double-clicks the button or the UI is briefly out of sync.
   if (Boolean(tenant.spec.suspend) === suspend) {
@@ -87,10 +100,32 @@ export async function PATCH(
 
   try {
     await patchTenantSpec(name, { suspend });
+
+    // On admin-side resume, also clear the pending-resume-request
+    // annotation if it exists. Belt-and-suspenders: the admin-approve
+    // endpoint already clears it on its happy path, but a platform
+    // user resuming directly via this endpoint shouldn't leave the
+    // annotation behind. Best-effort: failure to clear the annotation
+    // is logged but doesn't fail the resume.
+    if (!suspend) {
+      try {
+        await setTenantAnnotation(
+          name,
+          "pieced.ch/resume-request-pending",
+          null
+        );
+      } catch (e) {
+        console.warn(
+          "failed to clear resume-request-pending annotation; operator will see it stale until next request transition",
+          e
+        );
+      }
+    }
+
     return NextResponse.json(
       {
         message: suspend
-          ? "Subscription cancelled. Your data is preserved."
+          ? "Subscription cancelled. Your data is preserved for 60 days."
           : "Subscription resumed.",
         suspend,
       },