add azr metrics in heartbeats (#2735)

* add azr metrics in heartbeats

* fix bug

* address comment

* address comment

* move heartbeat.go under metric folder

* emit error infos for getHomeAZ failures

* ignore exhuastive check

---------

Co-authored-by: Saksham Mittal <111590532+smittal22@users.noreply.github.com>

cns/logger/constants.go

@@ -8,6 +8,11 @@ const (
 	// Dimensions
 	OrchestratorTypeStr = "OrchestratorType"
 	NodeIDStr           = "NodeID"
+	HomeAZStr           = "HomeAZ"
+	IsAZRSupportedStr   = "IsAZRSupported"
+	HomeAZErrorCodeStr  = "HomeAZErrorCode"
+	HomeAZErrorMsgStr   = "HomeAZErrorMsg"
+
 	// CNS Snspshot properties
 	CnsNCSnapshotEventStr         = "CNSNCSnapshot"
 	IpConfigurationStr            = "IPConfiguration"

cns/metric/heartbeat.go

@@ -0,0 +1,50 @@
+// Copyright 2018 Microsoft. All rights reserved.
+// MIT License
+
+package metric
+
+import (
+	"context"
+	"strconv"
+	"time"
+
+	"github.com/Azure/azure-container-networking/aitelemetry"
+	"github.com/Azure/azure-container-networking/cns"
+	"github.com/Azure/azure-container-networking/cns/logger"
+	"github.com/Azure/azure-container-networking/cns/restserver"
+	"github.com/Azure/azure-container-networking/cns/types"
+)
+
+// SendHeartBeat emits node metrics periodically
+func SendHeartBeat(ctx context.Context, heartbeatInterval time.Duration, homeAzMonitor *restserver.HomeAzMonitor, channelMode string) {
+	ticker := time.NewTicker(heartbeatInterval)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+			metric := aitelemetry.Metric{
+				Name: logger.HeartBeatMetricStr,
+				// This signifies 1 heartbeat is sent. Sum of this metric will give us number of heartbeats received
+				Value:            1.0,
+				CustomDimensions: make(map[string]string),
+			}
+
+			// add azr metrics when channel mode is direct
+			if channelMode == cns.Direct {
+				getHomeAzResp := homeAzMonitor.GetHomeAz(ctx)
+				switch getHomeAzResp.Response.ReturnCode { //nolint:exhaustive // ignore exhaustive types check
+				case types.Success:
+					metric.CustomDimensions[logger.IsAZRSupportedStr] = strconv.FormatBool(getHomeAzResp.HomeAzResponse.IsSupported)
+					metric.CustomDimensions[logger.HomeAZStr] = strconv.FormatUint(uint64(getHomeAzResp.HomeAzResponse.HomeAz), 10)
+				default:
+					metric.CustomDimensions[logger.HomeAZErrorCodeStr] = getHomeAzResp.Response.ReturnCode.String()
+					metric.CustomDimensions[logger.HomeAZErrorMsgStr] = getHomeAzResp.Response.Message
+
+				}
+			}
+			logger.SendMetric(metric)
+		}
+	}
+}

cns/service/main.go

@@ -39,6 +39,7 @@ import (
 	nncctrl "github.com/Azure/azure-container-networking/cns/kubecontroller/nodenetworkconfig"
 	podctrl "github.com/Azure/azure-container-networking/cns/kubecontroller/pod"
 	"github.com/Azure/azure-container-networking/cns/logger"
+	"github.com/Azure/azure-container-networking/cns/metric"
 	"github.com/Azure/azure-container-networking/cns/middlewares"
 	"github.com/Azure/azure-container-networking/cns/multitenantcontroller"
 	"github.com/Azure/azure-container-networking/cns/multitenantcontroller/multitenantoperator"
@@ -932,7 +933,7 @@ func main() {
 	}
 
 	if !disableTelemetry {
-		go logger.SendHeartBeat(rootCtx, cnsconfig.TelemetrySettings.HeartBeatIntervalInMins)
+		go metric.SendHeartBeat(rootCtx, time.Minute*time.Duration(cnsconfig.TelemetrySettings.HeartBeatIntervalInMins), homeAzMonitor, cnsconfig.ChannelMode)
 		go httpRemoteRestService.SendNCSnapShotPeriodically(rootCtx, cnsconfig.TelemetrySettings.SnapshotIntervalInMins)
 	}