From 664aa795c1c1299ea3cff8d075cc650182c7d216 Mon Sep 17 00:00:00 2001
From: owensmallwood <owen.smallwood@grafana.com>
Date: Tue, 27 Sep 2022 12:25:56 -0600
Subject: [PATCH] PublicDashboards: Add metric for public dashboard datasource
 query success (#55701)

Adds metric for which datasource queries are successful or not with pubdash.
---
 pkg/infra/metrics/metrics.go                  | 11 ++++++
 .../publicdashboards/models/models.go         |  5 +++
 .../publicdashboards/service/logging.go       | 36 +++++++++++++++++++
 .../publicdashboards/service/service.go       |  6 ++--
 4 files changed, 55 insertions(+), 3 deletions(-)
 create mode 100644 pkg/services/publicdashboards/service/logging.go

diff --git a/pkg/infra/metrics/metrics.go b/pkg/infra/metrics/metrics.go
index c3dfce10fcc..952c4bf112e 100644
--- a/pkg/infra/metrics/metrics.go
+++ b/pkg/infra/metrics/metrics.go
@@ -4,6 +4,7 @@ import (
 	"runtime"
 
 	"github.com/grafana/grafana/pkg/infra/metrics/metricutil"
+	pubdash "github.com/grafana/grafana/pkg/services/publicdashboards/models"
 	"github.com/grafana/grafana/pkg/setting"
 	"github.com/prometheus/client_golang/prometheus"
 )
@@ -104,6 +105,9 @@ var (
 
 	// MPublicDashboardRequestCount is a metric counter for public dashboards requests
 	MPublicDashboardRequestCount prometheus.Counter
+
+	// MPublicDashboardDatasourceQuerySuccess is a metric counter for successful queries labelled by datasource
+	MPublicDashboardDatasourceQuerySuccess *prometheus.CounterVec
 )
 
 // Timers
@@ -419,6 +423,12 @@ func init() {
 		Namespace: ExporterName,
 	})
 
+	MPublicDashboardDatasourceQuerySuccess = metricutil.NewCounterVecStartingAtZero(prometheus.CounterOpts{
+		Name:      "public_dashboard_datasource_query_success",
+		Help:      "counter for queries to public dashboard datasources labelled by datasource type and success status success/failed",
+		Namespace: ExporterName,
+	}, []string{"datasource", "status"}, map[string][]string{"status": pubdash.QueryResultStatuses})
+
 	MStatTotalDashboards = prometheus.NewGauge(prometheus.GaugeOpts{
 		Name:      "stat_totals_dashboard",
 		Help:      "total amount of dashboards",
@@ -664,5 +674,6 @@ func initMetricVars() {
 		StatsTotalDataKeys,
 		MStatTotalPublicDashboards,
 		MPublicDashboardRequestCount,
+		MPublicDashboardDatasourceQuerySuccess,
 	)
 }
diff --git a/pkg/services/publicdashboards/models/models.go b/pkg/services/publicdashboards/models/models.go
index fed864bd974..0ba668f2727 100644
--- a/pkg/services/publicdashboards/models/models.go
+++ b/pkg/services/publicdashboards/models/models.go
@@ -24,6 +24,11 @@ func (e PublicDashboardErr) Error() string {
 	return "Dashboard Error"
 }
 
+const QuerySuccess = "success"
+const QueryFailure = "failure"
+
+var QueryResultStatuses = []string{QuerySuccess, QueryFailure}
+
 var (
 	ErrPublicDashboardFailedGenerateUniqueUid = PublicDashboardErr{
 		Reason:     "failed to generate unique public dashboard id",
diff --git a/pkg/services/publicdashboards/service/logging.go b/pkg/services/publicdashboards/service/logging.go
new file mode 100644
index 00000000000..80f787ff123
--- /dev/null
+++ b/pkg/services/publicdashboards/service/logging.go
@@ -0,0 +1,36 @@
+package service
+
+import (
+	"github.com/grafana/grafana/pkg/infra/log"
+	"github.com/grafana/grafana/pkg/infra/metrics"
+	"github.com/grafana/grafana/pkg/services/publicdashboards/models"
+)
+
+/**
+A place to record relevant logs and metrics within the service layer of public dashboards
+*/
+
+func LogQuerySuccess(datasources []string, log log.Logger) {
+	log.Info("Successfully queried datasources for public dashboard", "datasources", datasources)
+	label := getLabelName(datasources)
+	metrics.MPublicDashboardDatasourceQuerySuccess.WithLabelValues(label, models.QuerySuccess).Inc()
+}
+
+func LogQueryFailure(datasources []string, log log.Logger, err error) {
+	log.Error("Error querying datasources for public dashboard", "error", err.Error(), "datasources", datasources)
+	label := getLabelName(datasources)
+	metrics.MPublicDashboardDatasourceQuerySuccess.WithLabelValues(label, models.QueryFailure).Inc()
+}
+
+func getLabelName(datasources []string) string {
+	size := len(datasources)
+
+	switch size {
+	case 0:
+		return "none"
+	case 1:
+		return datasources[0]
+	default:
+		return "mixed"
+	}
+}
diff --git a/pkg/services/publicdashboards/service/service.go b/pkg/services/publicdashboards/service/service.go
index 6862fe87f74..377489f1247 100644
--- a/pkg/services/publicdashboards/service/service.go
+++ b/pkg/services/publicdashboards/service/service.go
@@ -207,15 +207,15 @@ func (pd *PublicDashboardServiceImpl) GetQueryDataResponse(ctx context.Context,
 
 	res, err := pd.QueryDataService.QueryDataMultipleSources(ctx, anonymousUser, skipCache, metricReq, true)
 
-	// We want to track which datasources were successful and which were not
 	reqDatasources := metricReq.GetUniqueDatasourceTypes()
 	if err != nil {
-		pd.log.Error("Error querying datasources for public dashboard", "error", err.Error(), "datasources", reqDatasources)
+		LogQueryFailure(reqDatasources, pd.log, err)
 		return nil, err
 	}
+	LogQuerySuccess(reqDatasources, pd.log)
 
-	pd.log.Info("Successfully queried datasources for public dashboard", "datasources", reqDatasources)
 	queries.SanitizeMetadataFromQueryData(res)
+
 	return res, nil
 }