[metric] #4 Initialize count metrics on reconciler start

The Count views won't show up in query results until data is sent when error
condition happens.

This change initialize these metrics and adjust the e2e test to check
for Rate aggregation.

The public documentation should also be revised to recommend the Rate
aggregation for this type of metrics.

e2e/nomostest/prometheus_metrics.go

@@ -344,56 +344,54 @@ func metricResourceFightsHasValueAtLeast(nt *NT, syncLabels prometheusmodel.Labe
 		labels := prometheusmodel.LabelSet{
 			prometheusmodel.LabelName(ocmetrics.KeyComponent.Name()): prometheusmodel.LabelValue(ocmetrics.OtelCollectorName),
 		}.Merge(syncLabels)
-		// ResourceFightsView counts the total number of ResourceFights, so we don't need to aggregate
-		query := fmt.Sprintf("%s%s", metricName, labels)
 		if value == 0 {
-			// Tolerate missing metrics when expecting a zero value.
-			// Don't allow any value other than zero.
-			return metricExistsWithValueOrDoesNotExist(ctx, nt, v1api, query, float64(value))
+			query := fmt.Sprintf("rate(%s%s[1m])", metricName, labels)
+			return metricExistsWithValue(ctx, nt, v1api, query, 0)
 		}
+		query := fmt.Sprintf("%s%s", metricName, labels)
 		return metricExistsWithValueAtLeast(ctx, nt, v1api, query, float64(value))
 	}
 }
 
 // metricResourceConflictsHasValueAtLeast returns a MetricsPredicate that
-// validates that ResourceConflicts has at least the expected value.
-// If the expected value is zero, the metric must be zero or not found.
+// validates whether ResourceConflicts has been recorded recently
+// If the expected value is zero, the rate at which ResourceConflicts occur is also zero.
 func metricResourceConflictsHasValueAtLeast(nt *NT, syncLabels prometheusmodel.LabelSet, commitHash string, value int) MetricsPredicate {
 	return func(ctx context.Context, v1api prometheusv1.API) error {
 		metricName := ocmetrics.ResourceConflictsView.Name
 		metricName = fmt.Sprintf("%s%s", prometheusConfigSyncMetricPrefix, metricName)
+		if value == 0 {
+			labels := prometheusmodel.LabelSet{
+				prometheusmodel.LabelName(ocmetrics.KeyComponent.Name()): prometheusmodel.LabelValue(ocmetrics.OtelCollectorName),
+				prometheusmodel.LabelName(ocmetrics.KeyCommit.Name()):    ocmetrics.Initialization,
+			}.Merge(syncLabels)
+			query := fmt.Sprintf("rate(%s%s[30s])", metricName, labels)
+			return metricExistsWithValue(ctx, nt, v1api, query, 0)
+		}
 		labels := prometheusmodel.LabelSet{
 			prometheusmodel.LabelName(ocmetrics.KeyComponent.Name()): prometheusmodel.LabelValue(ocmetrics.OtelCollectorName),
 			prometheusmodel.LabelName(ocmetrics.KeyCommit.Name()):    prometheusmodel.LabelValue(commitHash),
 		}.Merge(syncLabels)
-		// ResourceConflictsView counts the total number of ResourceConflicts, so we don't need to aggregate
 		query := fmt.Sprintf("%s%s", metricName, labels)
-		if value == 0 {
-			// Tolerate missing metrics when expecting a zero value.
-			// Don't allow any value other than zero.
-			return metricExistsWithValueOrDoesNotExist(ctx, nt, v1api, query, float64(value))
-		}
 		return metricExistsWithValueAtLeast(ctx, nt, v1api, query, float64(value))
 	}
 }
 
 // metricInternalErrorsHasValueAtLeast returns a MetricsPredicate that validates
-// that InternalErrors has at least the expected value.
-// If the expected value is zero, the metric must be zero or not found.
+// whether InternalErrors has been recorded recently.
+// If the expected value is zero, the rate at which InternalErrors occur is also zero.
 func metricInternalErrorsHasValueAtLeast(nt *NT, syncLabels prometheusmodel.LabelSet, value int) MetricsPredicate {
 	return func(ctx context.Context, v1api prometheusv1.API) error {
 		metricName := ocmetrics.InternalErrorsView.Name
 		metricName = fmt.Sprintf("%s%s", prometheusConfigSyncMetricPrefix, metricName)
 		labels := prometheusmodel.LabelSet{
 			prometheusmodel.LabelName(ocmetrics.KeyComponent.Name()): prometheusmodel.LabelValue(ocmetrics.OtelCollectorName),
 		}.Merge(syncLabels)
-		// InternalErrorsView counts the total number of InternalErrors, so we don't need to aggregate
-		query := fmt.Sprintf("%s%s", metricName, labels)
 		if value == 0 {
-			// Tolerate missing metrics when expecting a zero value.
-			// Don't allow any value other than zero.
-			return metricExistsWithValueOrDoesNotExist(ctx, nt, v1api, query, float64(value))
+			query := fmt.Sprintf("rate(%s%s[1m])", metricName, labels)
+			return metricExistsWithValue(ctx, nt, v1api, query, 0)
 		}
+		query := fmt.Sprintf("%s%s", metricName, labels)
 		return metricExistsWithValueAtLeast(ctx, nt, v1api, query, float64(value))
 	}
 }

pkg/metrics/record.go

@@ -27,6 +27,9 @@ import (
 	"kpt.dev/configsync/pkg/status"
 )
 
+// Initialization is the tag value to initiate a Count
+const Initialization = "cs_counter_initialization"
+
 func record(ctx context.Context, ms ...stats.Measurement) {
 	stats.Record(ctx, ms...)
 	if klog.V(5).Enabled() {

pkg/parse/run.go

@@ -78,6 +78,12 @@ type RunFunc func(ctx context.Context, p Parser, trigger string, state *reconcil
 
 // Run keeps checking whether a parse-apply-watch loop is necessary and starts a loop if needed.
 func Run(ctx context.Context, p Parser, nsControllerState *namespacecontroller.State, runOpts RunOpts) {
+	// Pre-initialize count metrics to ensure visibility, even if errors haven't occurred.
+	// Special label indicates initial state. Metrics are cumulative and used for rate aggregation
+	// to detect ongoing errors. Zero value means no recent errors.
+	metrics.RecordResourceFight(ctx, metrics.Initialization)
+	metrics.RecordInternalError(ctx, metrics.Initialization)
+	metrics.RecordResourceConflict(ctx, metrics.Initialization)
 	opts := p.options()
 	// Use timers, not tickers.
 	// Tickers can cause memory leaks and continuous execution, when execution