cleanedup unwanted metrics

opsramp · Apr 8, 2024 · dcd9704 · dcd9704
1 parent 7411cf6
commit dcd9704
Show file tree

Hide file tree

Showing 4 changed files with 3 additions and 57 deletions.
diff --git a/collect/collect.go b/collect/collect.go
@@ -107,9 +107,6 @@ func (i *InMemCollector) Start() error {
 	// listen for config reloads
 	i.Config.RegisterReloadCallback(i.sendReloadSignal)
 
-	i.Metrics.Register("trace_duration_ms", "histogram")
-	i.Metrics.Register("trace_spans_count_total", "histogram")
-	i.Metrics.Register("collector_tosend_queue", "histogram")
 	i.Metrics.Register("collector_incoming_queue", "histogram")
 	i.Metrics.Register("collector_peer_queue", "histogram")
 	i.Metrics.Register("collector_cache_size", "gauge")
@@ -125,12 +122,6 @@ func (i *InMemCollector) Start() error {
 	i.Metrics.Register(TraceSendEjectedFull, "counter")
 	i.Metrics.Register(TraceSendEjectedMemsize, "counter")
 
-	i.Metrics.RegisterWithDescriptionLabels(
-		"trace_operations_latency_ms",
-		"gauge",
-		"Trace latency wrt each trace operation",
-		[]string{"service_name", "operation", "app", "instance", "transaction_type", "transaction_category", "transaction_sub_category", "language"},
-	)
 	i.Metrics.RegisterWithDescriptionLabels(
 		"trace_operations_failed",
 		"counter",
@@ -149,24 +140,6 @@ func (i *InMemCollector) Start() error {
 		"Total Number of events in spans wrt each trace operation",
 		[]string{"service_name", "operation", "app", "instance", "transaction_type", "transaction_category", "transaction_sub_category", "language"},
 	)
-	i.Metrics.RegisterWithDescriptionLabels(
-		"trace_root_span",
-		"counter",
-		"Number of root spans in an operation",
-		[]string{"service_name", "operation", "app", "instance", "transaction_type", "transaction_category", "transaction_sub_category", "language"},
-	)
-	i.Metrics.RegisterWithDescriptionLabels(
-		"trace_spans_count",
-		"counter",
-		"Number of spans in an operation",
-		[]string{"service_name", "operation", "app", "instance", "transaction_type", "transaction_category", "transaction_sub_category", "language"},
-	)
-	i.Metrics.RegisterWithDescriptionLabels(
-		"trace_root_operation_latency_ms",
-		"gauge",
-		"Trace latency wrt each root trace operation",
-		[]string{"service_name", "operation", "app", "instance", "transaction_type", "transaction_category", "transaction_sub_category", "language"},
-	)
 	i.Metrics.RegisterWithDescriptionLabels(
 		"trace_root_operations_failed",
 		"counter",
@@ -662,9 +635,6 @@ func (i *InMemCollector) send(trace *types.Trace, reason string) {
 	}
 	trace.Sent = true
 
-	traceDur := time.Since(trace.ArrivalTime)
-	i.Metrics.Histogram("trace_duration_ms", float64(traceDur.Milliseconds()))
-	i.Metrics.Histogram("trace_spans_count_total", float64(trace.DescendantCount()))
 	if trace.RootSpan != nil {
 		i.Metrics.Increment("trace_send_has_root")
 	} else {
@@ -732,7 +702,6 @@ func (i *InMemCollector) send(trace *types.Trace, reason string) {
 
 		durationMsString, ok := span.Data["durationMs"]
 		if ok && durationMsString != nil {
-			i.Metrics.GaugeWithLabels("trace_operations_latency_ms", labels, metrics.ConvertNumeric(durationMsString))
 
 			// getting the latency from end and start time
 			i.Metrics.HistogramWithLabels(
@@ -766,11 +735,7 @@ func (i *InMemCollector) send(trace *types.Trace, reason string) {
 				(metrics.ConvertNumeric(span.Data["endTime"])-metrics.ConvertNumeric(span.Data["startTime"]))/float64(time.Millisecond),
 			)
 
-			i.Metrics.GaugeWithLabels("trace_root_operation_latency_ms", labels, metrics.ConvertNumeric(durationMsString))
-			i.Metrics.IncrementWithLabels("trace_root_span", labels)
 		}
-		i.Metrics.IncrementWithLabels("trace_spans_count", labels)
-
 		errorStatus, ok := span.Data["error"]
 		if ok && errorStatus != nil && errorStatus.(bool) {
 			i.Metrics.IncrementWithLabels("trace_operations_failed", labels)

diff --git a/metrics/opsramp.go b/metrics/opsramp.go
@@ -562,7 +562,7 @@ func (p *OpsRampMetrics) calculateTraceOperationError(metricFamilySlice []*io_pr
 		if !p.re.MatchString(metricFamily.GetName()) {
 			continue
 		}
-		if metricFamily.GetName() == "trace_operations_failed" || metricFamily.GetName() == "trace_spans_count" {
+		if metricFamily.GetName() == "trace_operations_failed" {
 			for _, metric := range metricFamily.GetMetric() {
 				var labels []prompb.Label
 				for _, label := range metric.GetLabel() {
@@ -571,19 +571,15 @@ func (p *OpsRampMetrics) calculateTraceOperationError(metricFamilySlice []*io_pr
 						Value: label.GetValue(),
 					})
 				}
-				key := "trace_operations_failed&trace_spans_count&"
+				key := "trace_operations_failed&"
 				labelSlice := metric.GetLabel()
 				sort.Slice(labelSlice, func(i, j int) bool {
 					return labelSlice[i].GetName()+labelSlice[i].GetValue() > labelSlice[j].GetName()+labelSlice[i].GetValue()
 				})
 				for _, label := range labelSlice {
 					key += label.GetName() + label.GetValue()
 				}
-				if metricFamily.GetName() == "trace_operations_failed" {
-					uniqueFailedMap[key] = *metric.Counter.Value
-				} else {
-					uniqueSpansMap[key] = *metric.Counter.Value
-				}
+				uniqueFailedMap[key] = *metric.Counter.Value
 				uniqueLabelsMap[key] = labels
 			}
 		}

diff --git a/pkg/libtrace/transmission/transmission.go b/pkg/libtrace/transmission/transmission.go
@@ -290,10 +290,8 @@ func (h *TraceProxy) Flush() (err error) {
 // work can be enqueued.
 func (h *TraceProxy) Add(ev *Event) {
 	if h.tryAdd(ev) {
-		h.Metrics.Increment("messages_queued")
 		return
 	}
-	h.Metrics.Increment("queue_overflow")
 	r := Response{
 		Err:      errors.New("queue overflow"),
 		Metadata: ev.Metadata,
@@ -309,12 +307,6 @@ func (h *TraceProxy) tryAdd(ev *Event) bool {
 	h.musterLock.RLock()
 	defer h.musterLock.RUnlock()
 
-	// Even though this queue is locked against changing h.Muster, the Work queue length
-	// could change due to actions on the worker side, so make sure we only measure it once.
-	qlen := len(h.muster.Work)
-	h.Logger.Debug().Logf("adding event to transmission; queue length %d", qlen)
-	h.Metrics.Gauge("queue_length", qlen)
-
 	if h.BlockOnSend {
 		h.muster.Work <- ev
 		return true
@@ -774,9 +766,6 @@ func (b *batchAgg) exportProtoMsgBatch(events []*Event) {
 			if st, ok := status.FromError(err); ok {
 				if st.Code() != codes.OK {
 					b.logger.Error().Logf("sending failed. error: %s", st.String())
-					b.metrics.Increment("send_errors")
-				} else {
-					b.metrics.Increment("batches_sent")
 				}
 			}
 
@@ -793,15 +782,12 @@ func (b *batchAgg) exportProtoMsgBatch(events []*Event) {
 			if st, ok := status.FromError(err); ok {
 				if st.Code() != codes.OK {
 					b.logger.Error().Logf("sending failed. error: %s", st.String())
-					b.metrics.Increment("send_errors")
 					if strings.Contains(strings.ToUpper(err.Error()), "TRACE MANAGEMENT WAS NOT ENABLED") {
 						b.logger.Error().Logf("Enable Trace Management For Tenant and Restart Tracing Proxy")
 						m.Lock()
 						SendTraces = false
 						m.Unlock()
 					}
-				} else {
-					b.metrics.Increment("batches_sent")
 				}
 			}
 			b.logger.Debug().Logf("trace proxy response msg: %s", r.GetMessage())

diff --git a/route/route.go b/route/route.go
@@ -135,7 +135,6 @@ func (r *Router) LnS(incomingOrPeer string) {
 		return
 	}
 
-	r.Metrics.Register(r.incomingOrPeer+"_router_proxied", "counter")
 	r.Metrics.Register(r.incomingOrPeer+"_router_event", "counter")
 	r.Metrics.Register(r.incomingOrPeer+"_router_batch", "counter")
 	r.Metrics.Register(r.incomingOrPeer+"_router_nonspan", "counter")