@platformatic/watt-extra 0.1.8-alpha.0 → 0.1.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@platformatic/watt-extra",
-  "version": "0.1.8-alpha.0",
+  "version": "0.1.8",
   "description": "The Platformatic runtime manager",
   "type": "module",
   "scripts": {

package/plugins/alerts.js

@@ -4,6 +4,10 @@ async function alerts (app, _opts) {
   const healthCache = [] // It's OK to have this in memory, this is per-pod.
   const podHealthWindow =
     app.instanceConfig?.config?.scaler?.podHealthWindow || 60 * 1000
+  const alertRetentionWindow =
+    app.instanceConfig?.config?.scaler?.alertRetentionWindow || 10 * 1000
+
+  const lastServicesAlertTime = {}
 
   async function setupAlerts () {
     // Skip alerts setup if ICC is not configured
@@ -62,6 +66,17 @@ async function alerts (app, _opts) {
       // }
 
       if (healthInfo.unhealthy) {
+        const currentTime = Date.now()
+
+        const serviceId = healthInfo.id
+        const lastAlertTime = lastServicesAlertTime[serviceId]
+
+        if (lastAlertTime && currentTime - lastAlertTime < alertRetentionWindow) {
+          app.log.debug('Skipping alert, within retention window')
+          return
+        }
+
+        lastServicesAlertTime[serviceId] = currentTime
         delete healthInfo.healthConfig
 
         const authHeaders = await app.getAuthorizationHeader()
@@ -86,7 +101,6 @@ async function alerts (app, _opts) {
         }
 
         const alert = await body.json()
-        const serviceId = healthInfo.id
 
         try {
           await app.sendFlamegraphs({

package/test/alerts.test.js

@@ -327,6 +327,94 @@ test('should not fail when health info is missing', async (t) => {
   assert.strictEqual(alertReceived, null, 'No alert should have been received')
 })
 
+test('should respect alert retention window', async (t) => {
+  const applicationName = 'test-app'
+  const applicationId = randomUUID()
+  const applicationPath = join(__dirname, 'fixtures', 'service-1')
+
+  const alertsReceived = []
+
+  const getAuthorizationHeader = async (headers) => {
+    return { ...headers, authorization: 'Bearer test-token' }
+  }
+
+  const icc = await startICC(t, {
+    applicationId,
+    applicationName,
+    iccConfig: {
+      scaler: {
+        alertRetentionWindow: 500
+      }
+    },
+    processAlerts: (req) => {
+      const alert = req.body
+      assert.equal(req.headers.authorization, 'Bearer test-token')
+      alertsReceived.push(alert)
+      return { id: 'test-alert-id', ...alert }
+    }
+  })
+
+  setUpEnvironment({
+    PLT_APP_NAME: applicationName,
+    PLT_APP_DIR: applicationPath,
+    PLT_ICC_URL: 'http://127.0.0.1:3000'
+  })
+
+  const app = await start()
+
+  app.getAuthorizationHeader = getAuthorizationHeader
+
+  await app.setupAlerts()
+
+  t.after(async () => {
+    await app.close()
+    await icc.close()
+  })
+
+  // Create a health info template
+  const createHealthInfo = (serviceId, unhealthy = true) => ({
+    id: serviceId,
+    service: serviceId,
+    currentHealth: {
+      elu: unhealthy ? 0.95 : 0.5,
+      heapUsed: 76798040,
+      heapTotal: 99721216
+    },
+    unhealthy,
+    healthConfig: {
+      enabled: true,
+      interval: 1000,
+      gracePeriod: 1000,
+      maxUnhealthyChecks: 10,
+      maxELU: 0.99,
+      maxHeapUsed: 0.99,
+      maxHeapTotal: 4294967296
+    }
+  })
+
+  // Send first unhealthy event - should trigger alert
+  app.watt.runtime.emit('application:worker:health', createHealthInfo('service-1', true))
+  await sleep(50)
+
+  // Send second unhealthy event immediately - should trigger alert
+  app.watt.runtime.emit('application:worker:health', createHealthInfo('service-2', true))
+  await sleep(50)
+
+  // Send second unhealthy event immediately - should be ignored due to retention window
+  app.watt.runtime.emit('application:worker:health', createHealthInfo('service-1', true))
+  await sleep(100)
+
+  assert.strictEqual(alertsReceived.length, 2, 'Only one alert should be sent within retention window')
+
+  await sleep(500)
+
+  // Send third unhealthy event - should trigger second alert
+  app.watt.runtime.emit('application:worker:health', createHealthInfo('service-1', true))
+  await sleep(100)
+
+  assert.strictEqual(alertsReceived.length, 3, 'Second alert should be sent after retention window expires')
+})
+
 test('should not set up alerts when scaler URL is missing', async (t) => {
   const applicationName = 'test-app'
   const applicationId = randomUUID()

package/pnpm-workspace.yaml

@@ -1,2 +0,0 @@
-ignoredBuiltDependencies:
-  - better-sqlite3