feat(telemetry): add canonical VPS host monitoring

2026-08-22 17:29:30 +03:00
parent a3e7a015de
commit 23ea96fbb2
33 changed files with 2541 additions and 51 deletions
@@ -0,0 +1,102 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { readFile } from "node:fs/promises";
+
+import {
+  recordInfrastructureHostTelemetry,
+} from "../src/host-telemetry-repository.mjs";
+
+const edgeId = "73da0c42-a641-4559-b8f7-23509b60bfe9";
+const hostId = "adf2a5b6-3c0b-4a39-998c-07dfb7818ad1";
+const serviceId = "01f14736-f5c2-4867-9cbc-2d268996a871";
+const projectId = "ad7b357c-c7ac-4bf8-a638-c7f956e9aa71";
+const ownerId = "78da71d5-f48f-4de0-8e47-729f6d644151";
+
+test("records a normalized host observation only through the Edge-to-host graph", async () => {
+  const queries = [];
+  const now = new Date();
+  const client = {
+    async query(sql, parameters) {
+      queries.push({ sql, parameters });
+      if (queries.length === 1) return { rows: [{
+        service_instance_id: serviceId,
+        host_id: hostId,
+        project_id: projectId,
+        owner_scope_id: ownerId,
+        host_key: "robot2b-b2-edge-vps",
+      }] };
+      if (queries.length === 2) return { rows: [{
+        id: "7ea94f66-6eed-4a27-8f04-e67060fa7e94",
+        received_at: now,
+        expires_at: new Date(now.valueOf() + 15_000),
+        replayed: false,
+      }] };
+      return { rows: [] };
+    },
+  };
+
+  const result = await recordInfrastructureHostTelemetry(client, {
+    authenticatedEdgeRef: `edge:${edgeId}`,
+    snapshot: snapshot(now.toISOString()),
+  });
+
+  assert.equal(result.status, "recorded");
+  assert.equal(result.replayed, false);
+  assert.equal(result.hostRef, `host:${hostId}`);
+  assert.match(queries[0].sql, /device_infrastructure_service_instances/);
+  assert.equal(queries[0].parameters[0], edgeId);
+  assert.match(queries[1].sql, /device_infrastructure_host_telemetry_samples/);
+  assert.equal(queries[1].parameters[5], edgeId);
+  assert.equal(JSON.stringify(queries).includes("password"), false);
+  assert.match(queries[2].sql, /delete from device_infrastructure_host_telemetry_samples/);
+});
+
+test("rejects telemetry whose host key disagrees with the canonical graph", async () => {
+  const client = {
+    async query() {
+      return { rows: [{
+        service_instance_id: serviceId,
+        host_id: hostId,
+        project_id: projectId,
+        owner_scope_id: ownerId,
+        host_key: "canonical-host",
+      }] };
+    },
+  };
+  await assert.rejects(
+    () => recordInfrastructureHostTelemetry(client, {
+      authenticatedEdgeRef: `edge:${edgeId}`,
+      snapshot: snapshot(new Date().toISOString()),
+    }),
+    /device_host_telemetry_host_key_mismatch/,
+  );
+});
+
+test("host telemetry migration is additive, observation-backed and secret-free", async () => {
+  const sql = await readFile(new URL("../migrations/017_infrastructure_host_telemetry.sql", import.meta.url), "utf8");
+  assert.match(sql, /create table if not exists device_infrastructure_host_telemetry_samples/);
+  assert.match(sql, /observation\.observation/);
+  assert.match(sql, /references device_infrastructure_hosts/);
+  assert.match(sql, /references device_infrastructure_service_instances/);
+  assert.doesNotMatch(sql, /insert into device_infrastructure_hosts/i);
+  assert.doesNotMatch(sql, /password|private.key|credential/i);
+});
+
+function snapshot(observedAt) {
+  return {
+    schemaVersion: "nodedc.infrastructure.host-telemetry.v1",
+    profile: "linux-host-telegraf-v1",
+    hostKey: "robot2b-b2-edge-vps",
+    observedAt,
+    source: { agent: "telegraf", agentVersion: "1.38.4", collectorRef: "service:nodedc-host-telemetry-agent" },
+    hardware: { hostname: "koffyvngij", architecture: "x64", platform: "linux", kernelRelease: "6.8.0", cpuModel: "KVM CPU", logicalProcessors: 1 },
+    cpu: { usagePercent: 20, load1: 0.2, load5: 0.1, load15: 0.05 },
+    memory: { totalBytes: 1024, availableBytes: 700, freeBytes: null, usedBytes: 324, usedPercent: 31.6 },
+    swap: { totalBytes: 0, availableBytes: null, freeBytes: 0, usedBytes: 0, usedPercent: 0 },
+    system: { uptimeSeconds: 100, users: 1, processes: { total: 10, running: 1, sleeping: 9, blocked: 0, zombies: 0 } },
+    disks: [],
+    network: [],
+    services: [],
+  };
+}