fix(workspaces): stop a destroy that succeeded from answering 500

packages/control-plane/core/workspace-tunnels.ts

@@ -85,8 +85,9 @@ export class WorkspaceTunnels {
   }
 
   /** Deletes tunnel resources, clearing each column only after its
-   * confirmed deletion. Callers log the returned errors; anything left
-   * behind stays on the row for the janitor to retry. */
+   * confirmed deletion. Never throws, like the client it wraps: callers log
+   * the returned errors, and anything left behind stays on the row for the
+   * janitor to retry. */
   async cleanup(db: Db, row: WorkspaceTunnelRow): Promise<SurfaceCleanupResult> {
     if (row.tunnel_id === null && row.dns_record_id === null) {
       return { dnsDeleted: true, tunnelDeleted: true, errors: [] };
@@ -101,11 +102,23 @@ export class WorkspaceTunnels {
       assignments.push("tunnel_id = NULL", "tunnel_hostname = NULL");
     }
     if (assignments.length > 0) {
-      await rows(db, {
-        q: `UPDATE workspaces SET ${assignments.join(", ")}, updated_at = ?1
-            WHERE id = ?2`,
-        v: [Date.now(), row.id],
-      });
+      try {
+        await rows(db, {
+          q: `UPDATE workspaces SET ${assignments.join(", ")}, updated_at = ?1
+              WHERE id = ?2`,
+          v: [Date.now(), row.id],
+        });
+      } catch (error) {
+        // The client never throws and neither may this, because destroy calls
+        // cleanup AFTER the VM is gone: a throw here answered 500 for work
+        // that had already half-succeeded irreversibly, and left the caller
+        // with no way to tell that from a destroy which did nothing.
+        //
+        // Not clearing the columns is the safe half of this failure. Both
+        // deletes tolerate an already-deleted resource, so the janitor's retry
+        // is a no-op against Cloudflare and clears the columns then.
+        result.errors.push(error instanceof Error ? error.message : String(error));
+      }
     }
     return result;
   }

packages/control-plane/core/workspaces.ts

@@ -1179,6 +1179,15 @@ export function addWorkspaceRoutes(
       if (cleanup.errors.length > 0) {
         // Honest destroy: the row stays in destroying with its remaining
         // identifiers; the janitor retries until Cloudflare cleanup lands.
+        //
+        // Report them. The caller gets a 200 and the janitor's own transition
+        // sets `error` back to NULL, so these errors are the only account of
+        // why a destroy needed two attempts, and dropping them is what made
+        // the first such destroy unexplainable.
+        runtime.reportError(
+          "workspace_destroy_cleanup_incomplete",
+          new Error(`workspace ${id}: ${cleanup.errors.join("; ")}`),
+        );
         const pending = await workspaceById(runtime.db, id);
         if (pending === null) throw new Error("workspace disappeared during destroy");
         return context.json<CreateWorkspaceResponse>({

packages/control-plane/test/workspace-tunnels.test.ts

@@ -1,7 +1,9 @@
 import { describe, expect, it } from "vitest";
 import { rows } from "../core/db.js";
+import type { Db } from "../core/db.js";
 import { CloudflareTunnels } from "../core/compute/cloudflare-tunnels.js";
 import { WorkspaceTunnels } from "../core/workspace-tunnels.js";
+import type { WorkspaceTunnelRow } from "../core/workspace-tunnels.js";
 import { WorkspaceWebAppAuth } from "../core/webapp-tickets.js";
 import type { WorkspaceRow } from "../core/workspaces.js";
 import type { WorkspaceView } from "../core/wire.js";
@@ -90,6 +92,41 @@ describe("workspace tunnels", () => {
     expect(row?.dns_record_id).toBeNull();
   });
 
+  it("reports a failed column clear rather than throwing it at the caller", async () => {
+    // Destroy calls cleanup AFTER the VM is already deleted. A throw here
+    // answered an opaque 500 for a destroy that had irreversibly succeeded:
+    // the server was gone, the janitor finished the row and cleared `error`
+    // to NULL, and nothing was left to explain the 500 to whoever saw it.
+    const workspaceTunnels = new WorkspaceTunnels(
+      new CloudflareTunnels({
+        accountId: "test-account",
+        zoneId: "test-zone-id",
+        apiToken: "test-api-token",
+        fetcher: async () => Response.json({ success: true, result: {} }),
+      }),
+      "webapp.test",
+      "test-webapp-root-secret",
+      async () => Response.json({ ok: true }),
+    );
+    const unreachable = new Error("D1_ERROR: Network connection lost");
+    const failingDb: Db = {
+      rawSQL: () => ({ run: async () => { throw unreachable; } }),
+      rawSQLTransaction: () => ({ run: async () => { throw unreachable; } }),
+    };
+
+    const result = await workspaceTunnels.cleanup(failingDb, {
+      id: "ws-1",
+      tunnel_id: "tun-1",
+      tunnel_hostname: "ws-1.webapp.test",
+      dns_record_id: "dns-1",
+    } satisfies WorkspaceTunnelRow);
+
+    // Cloudflare did its half; only the column clear failed.
+    expect(result.tunnelDeleted).toBe(true);
+    expect(result.dnsDeleted).toBe(true);
+    expect(result.errors).toEqual(["D1_ERROR: Network connection lost"]);
+  });
+
   it("stays inert without configuration", async () => {
     const providers = new FakeProviders();
     const app = appWithVmProviders([providers], providers);