fix

apps/edr-freight-api/src/modules/health/health.controller.spec.ts

@@ -0,0 +1,100 @@
+import 'reflect-metadata';
+
+import type { Response } from 'express';
+import type { DataSource } from 'typeorm';
+
+import type { MatrixClient } from '../chat/matrix.client';
+import type { EmailClientService } from '../notifications/email-client.service';
+import type { SmsClientService } from '../notifications/sms-client.service';
+import { HealthController } from './health.controller';
+
+type ReadinessBody = {
+  status: string;
+  checks: {
+    chat: { status: string; enabled: boolean; actingAs?: string; error?: string };
+  };
+};
+
+/** Captures what the controller wrote, in place of an express Response. */
+function recorder() {
+  const sent: { code?: number; body?: ReadinessBody } = {};
+  const res = {
+    status(code: number) {
+      sent.code = code;
+      return this;
+    },
+    json(body: ReadinessBody) {
+      sent.body = body;
+      return this;
+    },
+  };
+  return { sent, res: res as unknown as Response };
+}
+
+function controllerWith(matrix: Partial<MatrixClient>) {
+  const dataSource = { query: jest.fn(async () => [{ '?column?': 1 }]) };
+  return new HealthController(
+    dataSource as unknown as DataSource,
+    { brokerConnected: true } as unknown as SmsClientService,
+    { brokerConnected: true } as unknown as EmailClientService,
+    matrix as MatrixClient,
+  );
+}
+
+describe('HealthController readiness — chat check', () => {
+  it('reports degraded, not 503, when MATRIX_ADMIN_TOKEN is not a server admin', async () => {
+    // The dev outage. Chat is broken, but chat is not worth pulling the pod
+    // out of the load balancer for — bookings and billing still work.
+    const controller = controllerWith({
+      enabled: true,
+      adminCheck: jest.fn(async () => ({
+        ok: false,
+        actingAs: '@super-admin.f15347:matrixdev.edrsc.com',
+        error: 'Matrix GET /_synapse/admin/v2/users?limit=1 -> 403: not a server admin',
+      })),
+    });
+
+    const { sent, res } = recorder();
+    await controller.readiness(res);
+
+    expect(sent.code).toBe(200);
+    expect(sent.body?.status).toBe('degraded');
+    expect(sent.body?.checks.chat.status).toBe('error');
+    // The account name is the actionable half — it says *which* token is wired up.
+    expect(sent.body?.checks.chat.actingAs).toBe(
+      '@super-admin.f15347:matrixdev.edrsc.com',
+    );
+  });
+
+  it('reports ok when the token really is a server admin', async () => {
+    const controller = controllerWith({
+      enabled: true,
+      adminCheck: jest.fn(async () => ({
+        ok: true,
+        actingAs: '@edrbot:matrixdev.edrsc.com',
+      })),
+    });
+
+    const { sent, res } = recorder();
+    await controller.readiness(res);
+
+    expect(sent.body?.status).toBe('ok');
+    expect(sent.body?.checks.chat).toMatchObject({
+      status: 'ok',
+      enabled: true,
+      actingAs: '@edrbot:matrixdev.edrsc.com',
+    });
+  });
+
+  it('does not call Synapse, or degrade, when chat is switched off', async () => {
+    const adminCheck = jest.fn();
+    const controller = controllerWith({ enabled: false, adminCheck });
+
+    const { sent, res } = recorder();
+    await controller.readiness(res);
+
+    expect(adminCheck).not.toHaveBeenCalled();
+    expect(sent.body?.status).toBe('ok');
+    expect(sent.body?.checks.chat).toEqual({ status: 'unknown', enabled: false });
+  });
+});

apps/edr-freight-api/src/modules/health/health.controller.ts

@@ -7,6 +7,7 @@ import { Public } from "@edr/api-common";
 import { Response } from "express";
 import { DataSource } from "typeorm";
 
+import { MatrixClient } from "../chat/matrix.client";
 import { EmailClientService } from "../notifications/email-client.service";
 import { SmsClientService } from "../notifications/sms-client.service";
 
@@ -32,6 +33,7 @@ export class HealthController {
     private readonly dataSource: DataSource,
     private readonly smsClient: SmsClientService,
     private readonly emailClient: EmailClientService,
+    private readonly matrix: MatrixClient,
   ) {}
 
   @Get()
@@ -45,7 +47,7 @@ export class HealthController {
   @Public()
   @ApiOperation({
     summary:
-      "Readiness probe — database plus SMS/email broker connectivity. Broker failures report as degraded unless READINESS_REQUIRES_BROKER=true.",
+      "Readiness probe — database, SMS/email broker connectivity, and the Matrix admin token. Broker failures report as degraded unless READINESS_REQUIRES_BROKER=true; chat failures always report as degraded.",
   })
   async readiness(@Res() res: Response) {
     const startedAt = Date.now();
@@ -76,23 +78,55 @@ export class HealthController {
       enabled: process.env.RABBITMQ_ENABLED !== "false",
     };
 
+    const chat = await this.chatCheck();
+
     const brokerDown =
       broker.sms.status === "error" || broker.email.status === "error";
     const failed =
       database.status === "error" ||
       (READINESS_REQUIRES_BROKER && brokerDown);
 
-    const status = failed ? "error" : brokerDown ? "degraded" : "ok";
+    const status = failed
+      ? "error"
+      : brokerDown || chat.status === "error"
+        ? "degraded"
+        : "ok";
 
     return res
       .status(failed ? HttpStatus.SERVICE_UNAVAILABLE : HttpStatus.OK)
       .json({
         status,
         timestamp: new Date().toISOString(),
-        checks: { database, broker },
+        checks: { database, broker, chat },
       });
   }
 
+  /**
+   * Chat provisioning runs entirely on MATRIX_ADMIN_TOKEN, and a token that is
+   * valid but not *server admin* fails only the `/_synapse/admin` half: rooms
+   * are never created, joins never happen, and the sole symptom is an empty
+   * Element for every employee. Nothing else in the probe would catch that.
+   *
+   * Degraded, never a 503 — chat is not worth pulling the pod out of the load
+   * balancer for, by the same reasoning as the broker check above. `unknown`
+   * when MATRIX_ENABLED is off: a feature that is switched off is not a fault.
+   */
+  private async chatCheck(): Promise<{
+    status: CheckStatus;
+    enabled: boolean;
+    actingAs?: string;
+    error?: string;
+  }> {
+    if (!this.matrix.enabled) return { status: "unknown", enabled: false };
+    const check = await this.matrix.adminCheck();
+    return {
+      status: check.ok ? "ok" : "error",
+      enabled: true,
+      actingAs: check.actingAs,
+      error: check.error,
+    };
+  }
+
   @Get("info")
   @Public()
   @ApiOperation({ summary: "App info — version, environment, uptime" })

apps/edr-freight-api/src/modules/health/health.module.ts

@@ -2,13 +2,15 @@
 
 import { Module } from "@nestjs/common";
 
+import { ChatModule } from "../chat/chat.module";
 import { HealthController } from "./health.controller";
 import { NotificationsModule } from "../notifications/notifications.module";
 
 @Module({
   // NotificationsModule exports the SMS/email clients; the readiness probe reads
   // their broker connection state rather than opening a second connection.
-  imports: [NotificationsModule],
+  // ChatModule exports MatrixClient for the MATRIX_ADMIN_TOKEN check.
+  imports: [NotificationsModule, ChatModule],
   controllers: [HealthController],
 })
 export class HealthModule {}