feat: Implement load balancing and health check features for service groups, including DNS-based load balancing modes and health check configurations, enhancing service reliability and performance monitoring

apps/api/test/health-check.test.ts

@@ -0,0 +1,124 @@
+import { describe, expect, it, beforeAll, afterAll } from "vitest";
+import { createServer, type Server } from "node:net";
+import * as healthCheckService from "../src/services/health-check-service.js";
+import type { HealthCheckTarget } from "@cfdm/shared";
+
+function startTcpServer(): Promise<{ server: Server; port: number }> {
+  return new Promise((resolve) => {
+    const server = createServer();
+    server.listen(0, "127.0.0.1", () => {
+      const address = server.address();
+      const port =
+        typeof address === "object" && address ? address.port : 0;
+      resolve({ server, port });
+    });
+  });
+}
+
+describe("health-check probeTarget", () => {
+  let server: Server;
+  let port: number;
+
+  beforeAll(async () => {
+    const started = await startTcpServer();
+    server = started.server;
+    port = started.port;
+  });
+
+  afterAll(async () => {
+    await new Promise<void>((resolve) => server.close(() => resolve()));
+  });
+
+  it("tcp probe succeeds for open port", async () => {
+    const target: HealthCheckTarget = {
+      scope: "binding",
+      ref_id: 1,
+      ip: "127.0.0.1",
+      hostname: "test.local",
+      type: "tcp",
+      port,
+      path: null,
+      expected_status: null,
+      timeout_ms: 1000,
+    };
+    const result = await healthCheckService.probeTarget(target);
+    expect(result.ok).toBe(true);
+    expect(result.error).toBeNull();
+    expect(result.latencyMs).toBeGreaterThanOrEqual(0);
+  });
+
+  it("tcp probe fails for closed port", async () => {
+    const target: HealthCheckTarget = {
+      scope: "binding",
+      ref_id: 1,
+      ip: "127.0.0.1",
+      hostname: "test.local",
+      type: "tcp",
+      port: 1,
+      path: null,
+      expected_status: null,
+      timeout_ms: 500,
+    };
+    const result = await healthCheckService.probeTarget(target);
+    expect(result.ok).toBe(false);
+    expect(result.error).not.toBeNull();
+  });
+});
+
+describe("health-check state derivation via runAllChecks", () => {
+  it("marks ip down after threshold failures and up after recovery", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "example.com", "zone-id");
+    const service = repos.createService(db, "Svc", "svc");
+    const binding = repos.insertBinding(
+      db,
+      domain.id,
+      service.id,
+      "@",
+      null,
+    );
+    repos.updateBindingLbConfig(db, binding.id, {
+      health_check_enabled: true,
+      health_check_type: "tcp",
+      health_check_port: 1,
+      health_check_timeout_ms: 200,
+    });
+    repos.replaceBindingIpsWithMeta(db, binding.id, [
+      { ip: "127.0.0.1", weight: 1, priority: 1 },
+    ]);
+
+    // Запуск с closed port (port 1) — должен зафиксировать down после 2 проверок
+    await healthCheckService.runAllChecks(db, {
+      thresholds: {
+        degradedFailures: 1,
+        downFailures: 2,
+        latencyWarnMs: 1000,
+      },
+    });
+    let status = repos.getIpHealthStatusRow(
+      db,
+      "binding",
+      binding.id,
+      "127.0.0.1",
+    );
+    expect(status?.status).toBe("degraded");
+
+    await healthCheckService.runAllChecks(db, {
+      thresholds: {
+        degradedFailures: 1,
+        downFailures: 2,
+        latencyWarnMs: 1000,
+      },
+    });
+    status = repos.getIpHealthStatusRow(
+      db,
+      "binding",
+      binding.id,
+      "127.0.0.1",
+    );
+    expect(status?.status).toBe("down");
+  });
+});