feat: pm2 metrics (#52)

CHANGELOG.md

@@ -7,62 +7,93 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+### Added
+
+- pm2 metrics
+
 ## 2.5.0 - 2023-05-17
+
 ### Added
+
 - CIP68 RFT 444 support
 
 ## 2.4.0 - 2023-03-15
+
 ### Removed
+
 - `addr_vk` support in `paymentCredToBech32Address`
 
 ## 2.3.2 - 2023-03-07
+
 ### Added
+
 - Support for ScriptHash payment credential using `script` addresses.
 
 ### Changed
+
 - do not leak framework in errors
 
 ## 2.3.1 - 2023-02-06
+
 ### Removed
+
 - `addr_vk` is no longer valid address type as returned by `detectAndValidateAddressType`
 
 ### Fixed
+
 - CIP68 `getMetadataFromOutputDatum` parsing of `files` and custom fields
 
 ## 2.3.0 - 2023-02-02
+
 ### Added
+
 - support for `addr_vk` addresses in `paymentCredFromBech32Address`
 - `getPaymentPartBech32` for generating partial bech32 address consisting of bech32 prefix, addr header and payment part
 
 ### Changed
+
 - fastify dependencies
 
 ## 2.2.0 - 2022-12-30
+
 ### Added
+
 - Missing custom 400s
 
 ## 2.1.1 - 2022-12-30
+
 ### Changed
+
 - js implementation of utf-8-validate, no longer requiring node-gyp
 
 ## 2.1.0 - 2022-12-28
+
 ### Added
+
 - CIP68 utils
 
 ## 2.0.0 - 2022-11-27
+
 ### Removed
+
 - BREAKING CHANGE: `getSchemaForEndpoint` is now part of the `@blockfrost/openapi` package
 - `@blockfrost/openapi` dependency
 
 ## 1.1.0 - 2022-11-02
+
 ### Added
+
 - `validatePolicy` and `validateAsset` to validation functions
 - invalid asset and invalid policy 400 handlers
 
 ## 1.0.1 - 2022-10-24
+
 ### Changed
+
 - `getAddressTypeAndPaymentCred` to return undefined instead of `''` when there is not valid `paymentCred`
 
 ## 1.0.0 - 2022-09-23
+
 ### Added
+
 - Initial release

package.json

@@ -23,6 +23,8 @@
     "bech32": "^2.0.0",
     "cbor": "^8.1.0",
     "crc": "^4.2.0",
+    "pm2": "^5.3.0",
+    "prom-client": "^14.2.0",
     "yaml": "^2.1.1"
   },
   "devDependencies": {

src/metrics-collector/pm2.ts

@@ -0,0 +1,104 @@
+import { register, Gauge } from 'prom-client';
+import pm2, { ProcessDescription } from 'pm2';
+
+export const pm2Metrics = async () => {
+  const memoryGauge = new Gauge({
+    name: 'worker_memory_bytes',
+    help: 'Worker memory usage in bytes',
+    labelNames: ['name', 'instance'],
+  });
+
+  const cpuGauge = new Gauge({
+    name: 'worker_cpu_percentage',
+    help: 'Worker CPU usage in percentage',
+    labelNames: ['name', 'instance'],
+  });
+
+  const uptimeGauge = new Gauge({
+    name: 'worker_uptime_seconds',
+    help: 'Worker uptime in seconds',
+    labelNames: ['name', 'instance'],
+  });
+
+  const restartsGauge = new Gauge({
+    name: 'worker_restarts_total',
+    help: 'Total number of process restarts',
+    labelNames: ['name', 'instance'],
+  });
+
+  const statusGauge = new Gauge({
+    name: 'worker_status',
+    help: 'Is the process running',
+    labelNames: ['name', 'instance'],
+  });
+
+  const instancesGauge = new Gauge({
+    name: 'worker_instances',
+    help: 'Process instances',
+    labelNames: ['name', 'instance'],
+  });
+
+  try {
+    await new Promise((resolve, reject) => {
+      pm2.connect(async error => {
+        if (error) {
+          console.error(error);
+          reject(new Error('Error connecting to PM2'));
+        }
+
+        try {
+          const processDescriptionList: ProcessDescription[] =
+            await new Promise((resolve, reject) => {
+              pm2.list((error, list) => {
+                if (error) {
+                  reject(error);
+                } else {
+                  resolve(list);
+                }
+              });
+            });
+
+          for (const proc of processDescriptionList) {
+            const { name, pm_id: instance, monit, pm2_env } = proc;
+            const labels = { name, instance };
+
+            uptimeGauge.set(
+              labels,
+              pm2_env
+                ? Math.round((Date.now() - (pm2_env.pm_uptime || 0)) / 1000)
+                : 0,
+            );
+            statusGauge.set(
+              labels,
+              pm2_env ? (pm2_env.status === 'online' ? 1 : 0) : 0,
+            );
+            instancesGauge.set(
+              labels,
+              pm2_env ? Number(pm2_env.instances) || 1 : 0,
+            );
+            memoryGauge.set(labels, monit ? monit.memory || 0 : 0);
+            cpuGauge.set(labels, monit ? monit.cpu || 0 : 0);
+            restartsGauge.set(
+              labels,
+              pm2_env
+                ? pm2_env.unstable_restarts || 0 + (pm2_env.restart_time || 0)
+                : 0,
+            );
+          }
+        } catch (error) {
+          console.error(error);
+          reject(new Error('Error fetching PM2 metrics'));
+        } finally {
+          pm2.disconnect();
+          resolve(true);
+        }
+      });
+    });
+
+    const metrics = await register.metrics();
+
+    return metrics;
+  } catch {
+    throw new Error('Error fetching PM2 metrics');
+  }
+};