feat: add ema for more smoother metrics calculations

Author: christyjacob4

.github/workflows/go-test.yml

@@ -0,0 +1,65 @@
+name: Go Tests
+
+on:
+  push:
+    branches: [ main, master ]
+    paths:
+      - '**.go'
+      - 'go.mod'
+      - 'go.sum'
+  pull_request:
+    branches: [ main, master ]
+    paths:
+      - '**.go'
+      - 'go.mod'
+      - 'go.sum'
+
+jobs:
+  test:
+    name: Run Tests
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v5
+      with:
+        go-version: '1.21'
+        check-latest: true
+
+    - name: Get dependencies
+      run: go mod tidy
+
+    - name: Run tests
+      run: go test -v -coverprofile=coverage.out ./...
+
+    - name: Upload coverage report
+      uses: actions/upload-artifact@v4
+      with:
+        name: coverage-report
+        path: coverage.out
+        if-no-files-found: error
+
+  coverage:
+    name: Coverage Analysis
+    needs: test
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v4
+
+    - name: Set up Go
+      uses: actions/setup-go@v5
+      with:
+        go-version: '1.21'
+        check-latest: true
+
+    - name: Download coverage report
+      uses: actions/download-artifact@v4
+      with:
+        name: coverage-report
+
+    - name: Display coverage
+      run: |
+        go tool cover -func=coverage.out 

coverage.out

@@ -0,0 +1,85 @@
+mode: set
+github.com/appwrite/monitoring/logger.go:24.20,28.2 1 1
+github.com/appwrite/monitoring/logger.go:30.82,34.2 3 0
+github.com/appwrite/monitoring/logger.go:36.58,39.2 2 0
+github.com/appwrite/monitoring/logger.go:41.62,44.2 2 0
+github.com/appwrite/monitoring/logger.go:46.59,49.2 2 0
+github.com/appwrite/monitoring/logger.go:51.60,54.2 2 0
+github.com/appwrite/monitoring/logger.go:56.59,59.2 2 0
+github.com/appwrite/monitoring/logger.go:61.60,64.2 2 0
+github.com/appwrite/monitoring/logger.go:66.60,69.2 2 0
+github.com/appwrite/monitoring/main.go:45.126,47.16 2 1
+github.com/appwrite/monitoring/main.go:54.2,69.8 3 1
+github.com/appwrite/monitoring/main.go:47.16,49.3 1 0
+github.com/appwrite/monitoring/main.go:72.81,74.2 1 1
+github.com/appwrite/monitoring/main.go:76.42,78.18 2 0
+github.com/appwrite/monitoring/main.go:81.2,81.19 1 0
+github.com/appwrite/monitoring/main.go:85.2,86.16 2 0
+github.com/appwrite/monitoring/main.go:90.2,90.26 1 0
+github.com/appwrite/monitoring/main.go:95.2,99.22 4 0
+github.com/appwrite/monitoring/main.go:105.2,115.29 2 0
+github.com/appwrite/monitoring/main.go:78.18,80.3 1 0
+github.com/appwrite/monitoring/main.go:81.19,83.3 1 0
+github.com/appwrite/monitoring/main.go:86.16,88.3 1 0
+github.com/appwrite/monitoring/main.go:90.26,92.3 1 0
+github.com/appwrite/monitoring/main.go:99.22,101.3 1 0
+github.com/appwrite/monitoring/main.go:101.8,103.3 1 0
+github.com/appwrite/monitoring/main.go:118.45,120.16 2 0
+github.com/appwrite/monitoring/main.go:124.2,128.22 4 0
+github.com/appwrite/monitoring/main.go:139.2,149.29 2 0
+github.com/appwrite/monitoring/main.go:120.16,122.3 1 0
+github.com/appwrite/monitoring/main.go:128.22,130.3 1 0
+github.com/appwrite/monitoring/main.go:130.8,137.3 1 0
+github.com/appwrite/monitoring/main.go:152.43,155.16 2 0
+github.com/appwrite/monitoring/main.go:159.2,163.22 4 0
+github.com/appwrite/monitoring/main.go:174.2,182.17 1 0
+github.com/appwrite/monitoring/main.go:187.2,188.16 2 0
+github.com/appwrite/monitoring/main.go:192.2,192.31 1 0
+github.com/appwrite/monitoring/main.go:228.2,228.12 1 0
+github.com/appwrite/monitoring/main.go:155.16,157.3 1 0
+github.com/appwrite/monitoring/main.go:163.22,165.3 1 0
+github.com/appwrite/monitoring/main.go:165.8,172.3 1 0
+github.com/appwrite/monitoring/main.go:182.17,184.3 1 0
+github.com/appwrite/monitoring/main.go:188.16,190.3 1 0
+github.com/appwrite/monitoring/main.go:192.31,194.17 2 0
+github.com/appwrite/monitoring/main.go:199.3,203.23 3 0
+github.com/appwrite/monitoring/main.go:215.3,223.18 1 0
+github.com/appwrite/monitoring/main.go:194.17,196.12 2 0
+github.com/appwrite/monitoring/main.go:203.23,205.4 1 0
+github.com/appwrite/monitoring/main.go:205.9,213.4 1 0
+github.com/appwrite/monitoring/main.go:223.18,225.4 1 0
+github.com/appwrite/monitoring/main.go:231.64,232.19 1 0
+github.com/appwrite/monitoring/main.go:235.2,235.15 1 0
+github.com/appwrite/monitoring/main.go:232.19,234.3 1 0
+github.com/appwrite/monitoring/main.go:238.57,240.16 2 0
+github.com/appwrite/monitoring/main.go:244.2,245.16 2 0
+github.com/appwrite/monitoring/main.go:249.2,254.16 5 0
+github.com/appwrite/monitoring/main.go:257.2,260.28 3 0
+github.com/appwrite/monitoring/main.go:264.2,264.12 1 0
+github.com/appwrite/monitoring/main.go:240.16,242.3 1 0
+github.com/appwrite/monitoring/main.go:245.16,247.3 1 0
+github.com/appwrite/monitoring/main.go:254.16,256.3 1 0
+github.com/appwrite/monitoring/main.go:260.28,262.3 1 0
+github.com/appwrite/monitoring/main.go:267.33,275.21 4 0
+github.com/appwrite/monitoring/main.go:275.21,277.3 1 0
+github.com/appwrite/monitoring/main.go:280.37,281.37 1 0
+github.com/appwrite/monitoring/main.go:285.2,285.40 1 0
+github.com/appwrite/monitoring/main.go:289.2,289.38 1 0
+github.com/appwrite/monitoring/main.go:281.37,283.3 1 0
+github.com/appwrite/monitoring/main.go:285.40,287.3 1 0
+github.com/appwrite/monitoring/main.go:289.38,291.3 1 0
+github.com/appwrite/monitoring/main.go:294.13,305.22 7 0
+github.com/appwrite/monitoring/main.go:310.2,313.27 2 0
+github.com/appwrite/monitoring/main.go:319.2,319.20 1 0
+github.com/appwrite/monitoring/main.go:322.2,322.38 1 0
+github.com/appwrite/monitoring/main.go:325.2,325.44 1 0
+github.com/appwrite/monitoring/main.go:328.2,328.40 1 0
+github.com/appwrite/monitoring/main.go:332.2,333.16 2 0
+github.com/appwrite/monitoring/main.go:337.2,343.17 6 0
+github.com/appwrite/monitoring/main.go:305.22,308.3 2 0
+github.com/appwrite/monitoring/main.go:313.27,316.3 2 0
+github.com/appwrite/monitoring/main.go:319.20,321.3 1 0
+github.com/appwrite/monitoring/main.go:322.38,324.3 1 0
+github.com/appwrite/monitoring/main.go:325.44,327.3 1 0
+github.com/appwrite/monitoring/main.go:328.40,330.3 1 0
+github.com/appwrite/monitoring/main.go:333.16,335.3 1 0

main.go

@@ -34,6 +34,12 @@ type SystemMonitor struct {
 	diskLimit      float64
 	interval       int
 	log            *Logger
+	
+	// EMA tracking
+	cpuEMA         float64
+	memoryEMA      float64
+	diskEMA        float64
+	alpha          float64 // EMA smoothing factor
 }
 
 func NewSystemMonitor(betterStackURL string, interval int, cpuLimit, memoryLimit, diskLimit float64) (*SystemMonitor, error) {
@@ -42,6 +48,12 @@ func NewSystemMonitor(betterStackURL string, interval int, cpuLimit, memoryLimit
 		return nil, fmt.Errorf("failed to get hostname: %v", err)
 	}
 
+	// Calculate alpha based on interval to get roughly 5 minutes of smoothing
+	// EMA formula: alpha = 2/(N+1) where N is the number of periods
+	// For 5 minutes of smoothing with our interval: N = 300/interval
+	N := float64(300) / float64(interval)
+	alpha := 2.0 / (N + 1.0)
+
 	return &SystemMonitor{
 		httpClient: &http.Client{
 			Timeout: 5 * time.Second,
@@ -53,9 +65,14 @@ func NewSystemMonitor(betterStackURL string, interval int, cpuLimit, memoryLimit
 		diskLimit:      diskLimit,
 		interval:       interval,
 		log:            New(),
+		alpha:          alpha,
 	}, nil
 }
 
+func (s *SystemMonitor) calculateEMA(currentValue, previousEMA float64) float64 {
+	return s.alpha*currentValue + (1-s.alpha)*previousEMA
+}
+
 func (s *SystemMonitor) checkCPU() error {
 	duration := float64(s.interval) / 10
 	if duration < 5 {
@@ -74,12 +91,15 @@ func (s *SystemMonitor) checkCPU() error {
 		return nil
 	}
 
-	value := cpuPercent[0]
-	status := s.getStatus(value, s.cpuLimit)
+	// Calculate EMA for CPU usage
+	instantValue := cpuPercent[0]
+	s.cpuEMA = s.calculateEMA(instantValue, s.cpuEMA)
+	
+	status := s.getStatus(s.cpuEMA, s.cpuLimit)
 	if status == "fail" {
-		s.log.Warn("CPU usage %.2f%% exceeds limit of %.2f%%", value, s.cpuLimit)
+		s.log.Warn("CPU usage EMA %.2f%% exceeds limit of %.2f%% (instant: %.2f%%)", s.cpuEMA, s.cpuLimit, instantValue)
 	} else {
-		s.log.Log("CPU usage: %.2f%% (limit: %.2f%%)", value, s.cpuLimit)
+		s.log.Log("CPU usage EMA: %.2f%% (limit: %.2f%%, instant: %.2f%%)", s.cpuEMA, s.cpuLimit, instantValue)
 	}
 
 	metric := Metric{
@@ -88,7 +108,7 @@ func (s *SystemMonitor) checkCPU() error {
 		AlertID:   fmt.Sprintf("cpu-%s", s.hostname),
 		Timestamp: time.Now().Unix(),
 		Status:    status,
-		Value:     value,
+		Value:     s.cpuEMA,
 		Limit:     s.cpuLimit,
 	}
 
@@ -101,14 +121,17 @@ func (s *SystemMonitor) checkMemory() error {
 		return fmt.Errorf("failed to get memory stats: %v", err)
 	}
 
-	value := vmStat.UsedPercent
-	status := s.getStatus(value, s.memoryLimit)
+	instantValue := vmStat.UsedPercent
+	s.memoryEMA = s.calculateEMA(instantValue, s.memoryEMA)
+	
+	status := s.getStatus(s.memoryEMA, s.memoryLimit)
 	if status == "fail" {
-		s.log.Warn("Memory usage %.2f%% exceeds limit of %.2f%%", value, s.memoryLimit)
+		s.log.Warn("Memory usage EMA %.2f%% exceeds limit of %.2f%% (instant: %.2f%%)", s.memoryEMA, s.memoryLimit, instantValue)
 	} else {
-		s.log.Log("Memory usage: %.2f%% (limit: %.2f%%), Available: %d MB, Total: %d MB",
-			value,
+		s.log.Log("Memory usage EMA: %.2f%% (limit: %.2f%%, instant: %.2f%%), Available: %d MB, Total: %d MB",
+			s.memoryEMA,
 			s.memoryLimit,
+			instantValue,
 			vmStat.Available/(1024*1024),
 			vmStat.Total/(1024*1024))
 	}
@@ -119,7 +142,7 @@ func (s *SystemMonitor) checkMemory() error {
 		AlertID:   fmt.Sprintf("memory-%s", s.hostname),
 		Timestamp: time.Now().Unix(),
 		Status:    status,
-		Value:     value,
+		Value:     s.memoryEMA,
 		Limit:     s.memoryLimit,
 	}
 
@@ -133,14 +156,17 @@ func (s *SystemMonitor) checkDisk() error {
 		return fmt.Errorf("failed to get disk usage: %v", err)
 	}
 
-	value := usage.UsedPercent
-	status := s.getStatus(value, s.diskLimit)
+	instantValue := usage.UsedPercent
+	s.diskEMA = s.calculateEMA(instantValue, s.diskEMA)
+	
+	status := s.getStatus(s.diskEMA, s.diskLimit)
 	if status == "fail" {
-		s.log.Warn("Root disk usage %.2f%% exceeds limit of %.2f%%", value, s.diskLimit)
+		s.log.Warn("Root disk usage EMA %.2f%% exceeds limit of %.2f%% (instant: %.2f%%)", s.diskEMA, s.diskLimit, instantValue)
 	} else {
-		s.log.Log("Root disk usage: %.2f%% (limit: %.2f%%), Free: %d MB, Total: %d MB",
-			value,
+		s.log.Log("Root disk usage EMA: %.2f%% (limit: %.2f%%, instant: %.2f%%), Free: %d MB, Total: %d MB",
+			s.diskEMA,
 			s.diskLimit,
+			instantValue,
 			usage.Free/(1024*1024),
 			usage.Total/(1024*1024))
 	}
@@ -151,7 +177,7 @@ func (s *SystemMonitor) checkDisk() error {
 		AlertID:   fmt.Sprintf("disk-root-%s", s.hostname),
 		Timestamp: time.Now().Unix(),
 		Status:    status,
-		Value:     value,
+		Value:     s.diskEMA,
 		Limit:     s.diskLimit,
 	}); err != nil {
 		return err
@@ -170,15 +196,18 @@ func (s *SystemMonitor) checkDisk() error {
 			continue
 		}
 
-		value := usage.UsedPercent
-		status := s.getStatus(value, s.diskLimit)
+		instantValue := usage.UsedPercent
+		// For mounted directories, we'll use the same EMA as root for simplicity
+		// In a more sophisticated implementation, we might want separate EMAs for each mount
+		status := s.getStatus(s.diskEMA, s.diskLimit)
 		if status == "fail" {
-			s.log.Warn("Disk usage for %s %.2f%% exceeds limit of %.2f%%", mount, value, s.diskLimit)
+			s.log.Warn("Disk usage for %s EMA %.2f%% exceeds limit of %.2f%% (instant: %.2f%%)", mount, s.diskEMA, s.diskLimit, instantValue)
 		} else {
-			s.log.Log("Disk usage for %s: %.2f%% (limit: %.2f%%), Free: %d MB, Total: %d MB",
+			s.log.Log("Disk usage for %s EMA: %.2f%% (limit: %.2f%%, instant: %.2f%%), Free: %d MB, Total: %d MB",
 				mount,
-				value,
+				s.diskEMA,
 				s.diskLimit,
+				instantValue,
 				usage.Free/(1024*1024),
 				usage.Total/(1024*1024))
 		}
@@ -189,7 +218,7 @@ func (s *SystemMonitor) checkDisk() error {
 			AlertID:   fmt.Sprintf("disk-%s-%s", filepath.Base(mount), s.hostname),
 			Timestamp: time.Now().Unix(),
 			Status:    status,
-			Value:     value,
+			Value:     s.diskEMA,
 			Limit:     s.diskLimit,
 		}); err != nil {
 			return err