From d830b8dd161c4cd48de83faae135eaaf2e0b6986 Mon Sep 17 00:00:00 2001
From: Patrick Ohly <patrick.ohly@intel.com>
Date: Wed, 11 Nov 2020 11:00:04 +0100
Subject: [PATCH] metrics: optional process_start_time

The process_start_time as registered by csi-lib-utils and thus
Kubernetes component base conflicts with the one from the Prometheus
Golang collector, leading to this error:

gathered metric family process_start_time_seconds has help "[ALPHA] Start time of the process since unix epoch in seconds." but should have "Start time of the process since unix epoch in seconds."

Allowing processes to control where they get the metric from solves
this conflict.
---
 metrics/metrics.go | 35 ++++++++++++++++++++++++++++-------
 1 file changed, 28 insertions(+), 7 deletions(-)

diff --git a/metrics/metrics.go b/metrics/metrics.go
index 6ca2b9ae..4c81c46f 100644
--- a/metrics/metrics.go
+++ b/metrics/metrics.go
@@ -145,6 +145,22 @@ func WithLabels(labels map[string]string) MetricsManagerOption {
 	}
 }
 
+// WithProcessStartTime controlls whether process_start_time_seconds is registered
+// in the registry of the metrics manager. It's enabled by default out of convenience
+// (no need to do anything special in most sidecars) but should be disabled in more
+// complex scenarios (more than one metrics manager per process, metric already
+// provided elsewhere like via the Prometheus Golang collector).
+//
+// In particular, registering this metric via metric manager and thus the Kubernetes
+// component base conflicts with the Prometheus Golang collector (gathered metric family
+// process_start_time_seconds has help "[ALPHA] Start time of the process since unix epoch in seconds."
+// but should have "Start time of the process since unix epoch in seconds."
+func WithProcessStartTime(registerProcessStartTime bool) MetricsManagerOption {
+	return func(cmm *csiMetricsManager) {
+		cmm.registerProcessStartTime = registerProcessStartTime
+	}
+}
+
 // NewCSIMetricsManagerForSidecar creates and registers metrics for CSI Sidecars and
 // returns an object that can be used to trigger the metrics. It uses "csi_sidecar"
 // as subsystem.
@@ -177,18 +193,22 @@ func NewCSIMetricsManagerForPlugin(driverName string) CSIMetricsManager {
 //              If unknown, leave empty, and use SetDriverName method to update later.
 func NewCSIMetricsManagerWithOptions(driverName string, options ...MetricsManagerOption) CSIMetricsManager {
 	cmm := csiMetricsManager{
-		registry:       metrics.NewKubeRegistry(),
-		subsystem:      SubsystemSidecar,
-		stabilityLevel: metrics.ALPHA,
+		registry:                 metrics.NewKubeRegistry(),
+		subsystem:                SubsystemSidecar,
+		stabilityLevel:           metrics.ALPHA,
+		registerProcessStartTime: true,
 	}
 
-	// https://github.com/open-telemetry/opentelemetry-collector/issues/969
-	// Add process_start_time_seconds into the metric to let the start time be parsed correctly
-	metrics.RegisterProcessStartTime(cmm.registry.Register)
-
 	for _, option := range options {
 		option(&cmm)
 	}
+
+	if cmm.registerProcessStartTime {
+		// https://github.com/open-telemetry/opentelemetry-collector/issues/969
+		// Add process_start_time_seconds into the metric to let the start time be parsed correctly
+		metrics.RegisterProcessStartTime(cmm.registry.Register)
+	}
+
 	labels := []string{labelCSIDriverName, labelCSIOperationName, labelGrpcStatusCode}
 	labels = append(labels, cmm.additionalLabelNames...)
 	for _, label := range cmm.additionalLabels {
@@ -219,6 +239,7 @@ type csiMetricsManager struct {
 	additionalLabelNames       []string
 	additionalLabels           []label
 	csiOperationsLatencyMetric *metrics.HistogramVec
+	registerProcessStartTime   bool
 }
 
 type label struct {