feat(heartbeat): add process memory and backup health checks

src/automation/heartbeat.test.ts

@@ -8,9 +8,11 @@ function makeConfig(overrides?: Partial<HeartbeatConfig>): HeartbeatConfig {
   return {
     enabled: true,
     interval: '5m',
-    checks: ['gateway', 'model', 'channels', 'memory', 'disk'],
+    checks: ['gateway', 'model', 'channels', 'memory', 'disk', 'process_memory', 'backup'],
     failure_threshold: 2,
     disk_threshold_mb: 100,
+    process_memory_threshold_mb: 1500,
+    backup_failure_threshold: 1,
     ...overrides,
   };
 }
@@ -29,6 +31,12 @@ function makeDeps(overrides?: Partial<HeartbeatDeps>): HeartbeatDeps {
     memoryDir: '/tmp/flynn-test-memory',
     dataDir: '/tmp',
     channelLookup: { get: vi.fn() },
+    processMemoryUsageMb: () => 256,
+    backupHealthProvider: () => ({
+      enabled: false,
+      hasRun: false,
+      consecutiveFailures: 0,
+    }),
     ...overrides,
   };
 }
@@ -436,4 +444,73 @@ describe('HeartbeatMonitor', () => {
       expect(check.healthy).toBe(false);
     });
   });
+
+  describe('process_memory check', () => {
+    it('passes when RSS is below threshold', async () => {
+      const deps = makeDeps({
+        config: makeConfig({ checks: ['process_memory'], process_memory_threshold_mb: 512 }),
+        processMemoryUsageMb: () => 200,
+      });
+      monitor = new HeartbeatMonitor(deps);
+
+      const result = await monitor.runChecks();
+      const check = result.checks.find((c) => c.name === 'process_memory');
+      if (!check) {throw new Error('Expected process_memory check result');}
+      expect(check.healthy).toBe(true);
+    });
+
+    it('fails when RSS is above threshold', async () => {
+      const deps = makeDeps({
+        config: makeConfig({ checks: ['process_memory'], process_memory_threshold_mb: 128 }),
+        processMemoryUsageMb: () => 512,
+      });
+      monitor = new HeartbeatMonitor(deps);
+
+      const result = await monitor.runChecks();
+      const check = result.checks.find((c) => c.name === 'process_memory');
+      if (!check) {throw new Error('Expected process_memory check result');}
+      expect(check.healthy).toBe(false);
+      expect(check.message).toContain('High memory usage');
+    });
+  });
+
+  describe('backup check', () => {
+    it('passes when backup is disabled', async () => {
+      const deps = makeDeps({
+        config: makeConfig({ checks: ['backup'] }),
+        backupHealthProvider: () => ({
+          enabled: false,
+          hasRun: false,
+          consecutiveFailures: 0,
+        }),
+      });
+      monitor = new HeartbeatMonitor(deps);
+
+      const result = await monitor.runChecks();
+      const check = result.checks.find((c) => c.name === 'backup');
+      if (!check) {throw new Error('Expected backup check result');}
+      expect(check.healthy).toBe(true);
+      expect(check.message).toContain('disabled');
+    });
+
+    it('fails when backup consecutive failures exceed threshold', async () => {
+      const deps = makeDeps({
+        config: makeConfig({ checks: ['backup'], backup_failure_threshold: 2 }),
+        backupHealthProvider: () => ({
+          enabled: true,
+          hasRun: true,
+          consecutiveFailures: 3,
+          lastError: 'minio unavailable',
+        }),
+      });
+      monitor = new HeartbeatMonitor(deps);
+
+      const result = await monitor.runChecks();
+      const check = result.checks.find((c) => c.name === 'backup');
+      if (!check) {throw new Error('Expected backup check result');}
+      expect(check.healthy).toBe(false);
+      expect(check.message).toContain('Backup failing');
+      expect(check.message).toContain('minio unavailable');
+    });
+  });
 });