@platformatic/watt-extra 1.7.0 → 1.7.1-alpha.2

package/.claude/settings.local.json

@@ -1,14 +1,11 @@
 {
   "permissions": {
     "allow": [
-      "Read(//work/workspaces/workspace-platformatic/platformatic/**)",
-      "Bash(npx borp:*)",
-      "Bash(timeout 30 npx borp -c 1 --timeout=20000 ./test/trigger-flamegraphs.test.js)",
-      "Bash(xargs cat:*)",
-      "Bash(pnpm install)",
-      "Bash(find:*)",
-      "Bash(cat:*)",
-      "WebFetch(domain:github.com)"
+      "Bash(node --test-only:*)",
+      "Bash(node --test:*)",
+      "Bash(for i in {1..3})",
+      "Bash(do echo \"=== Run $i ===\")",
+      "Bash(done)"
     ],
     "deny": [],
     "ask": []

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@platformatic/watt-extra",
-  "version": "1.7.0",
+  "version": "1.7.1-alpha.2",
   "description": "The Platformatic runtime manager",
   "type": "module",
   "scripts": {

package/plugins/alerts.js

@@ -1,6 +1,9 @@
 import { request } from 'undici'
 
 async function alerts (app, _opts) {
+  const pauseEluThreshold = app.env.PLT_FLAMEGRAPHS_PAUSE_ELU_TRESHOLD
+  const pauseTimeout = app.env.PLT_FLAMEGRAPHS_PAUSE_TIMEOUT
+
   const healthCache = [] // It's OK to have this in memory, this is per-pod.
   const podHealthWindow =
     app.instanceConfig?.scaler?.podHealthWindow || 60 * 1000
@@ -61,6 +64,12 @@ async function alerts (app, _opts) {
       const healthWithTimestamp = { ...healthInfo, timestamp, service: serviceId }
       delete healthWithTimestamp.healthConfig // we don't need to store this
 
+      const elu = healthInfo.currentHealth.elu
+      process._rawDebug('--------ALERT-------', elu)
+      if (elu >= pauseEluThreshold) {
+        app.pauseProfiling({ serviceId, timeout: pauseTimeout })
+      }
+
       healthCache.push(healthWithTimestamp)
 
       const cutoffTime = timestamp - podHealthWindow
@@ -136,12 +145,8 @@ async function alerts (app, _opts) {
 
         const alert = await body.json()
 
-        app.sendFlamegraphs({
-          workerIds: [workerId],
-          alertId: alert.id
-        }).catch(err => {
-          app.log.error({ err }, 'Failed to send a flamegraph')
-        })
+        app.requestFlamegraphs({ serviceIds: [serviceId], alertId: alert.id })
+          .catch(err => app.log.error({ err }, 'Failed to send a flamegraph'))
       }
     }
 

package/plugins/env.js

@@ -19,10 +19,9 @@ const schema = {
     PLT_CACHE_CONFIG: { type: 'string' },
     PLT_DISABLE_FLAMEGRAPHS: { type: 'boolean', default: false },
     PLT_FLAMEGRAPHS_INTERVAL_SEC: { type: 'number', default: 60 },
-    PLT_FLAMEGRAPHS_ELU_THRESHOLD: { type: 'number', default: 0.4 },
     PLT_FLAMEGRAPHS_GRACE_PERIOD: { type: 'number', default: 3000 },
-    PLT_FLAMEGRAPHS_ATTEMPT_TIMEOUT: { type: 'number', default: 10000 },
-    PLT_FLAMEGRAPHS_CACHE_CLEANUP_INTERVAL: { type: 'number', default: 120000 },
+    PLT_FLAMEGRAPHS_PAUSE_ELU_TRESHOLD: { type: 'number', default: 0.95 },
+    PLT_FLAMEGRAPHS_PAUSE_TIMEOUT: { type: 'number', default: 5 * 60 * 1000 },
     PLT_JWT_EXPIRATION_OFFSET_SEC: { type: 'number', default: 60 },
     PLT_UPDATES_RECONNECT_INTERVAL_SEC: { type: 'number', default: 1 },
     PLT_ELU_HEALTH_SIGNAL_THRESHOLD: { type: 'number', default: 0.8 },