feat: add ability to include reason in count metrics

Adds a configuration flag to enable including the `reason` for a TaskRun or PipelineRun status on their count metrics. This allows for more fine-grained monitoring and alerting of run failures.

Signed-off-by: Marcus Noble <[email protected]>

config/config-observability.yaml

@@ -58,3 +58,4 @@ data:
     metrics.taskrun.duration-type: "histogram"
     metrics.pipelinerun.level: "pipeline"
     metrics.pipelinerun.duration-type: "histogram"
+    metrics.count.reason: "false"

docs/metrics.md

@@ -15,10 +15,10 @@ We expose several kinds of exporters, including Prometheus, Google Stackdriver,
 | ---------- | ----------- | ----------- | ----------- |
 | `tekton_pipelines_controller_pipelinerun_duration_seconds_[bucket, sum, count]` | Histogram/LastValue(Gauge) | `*pipeline`=&lt;pipeline_name&gt; <br> `*pipelinerun`=&lt;pipelinerun_name&gt; <br> `status`=&lt;status&gt; <br> `namespace`=&lt;pipelinerun-namespace&gt; | experimental |
 | `tekton_pipelines_controller_pipelinerun_taskrun_duration_seconds_[bucket, sum, count]` | Histogram/LastValue(Gauge) | `*pipeline`=&lt;pipeline_name&gt; <br> `*pipelinerun`=&lt;pipelinerun_name&gt; <br> `status`=&lt;status&gt; <br> `*task`=&lt;task_name&gt; <br> `*taskrun`=&lt;taskrun_name&gt;<br> `namespace`=&lt;pipelineruns-taskruns-namespace&gt;| experimental |
-| `tekton_pipelines_controller_pipelinerun_count` | Counter | `status`=&lt;status&gt; | experimental |
+| `tekton_pipelines_controller_pipelinerun_count` | Counter | `status`=&lt;status&gt; <br> `*reason`=&lt;reason&gt; | experimental |
 | `tekton_pipelines_controller_running_pipelineruns_count` | Gauge | | experimental |
 | `tekton_pipelines_controller_taskrun_duration_seconds_[bucket, sum, count]` | Histogram/LastValue(Gauge) | `status`=&lt;status&gt; <br> `*task`=&lt;task_name&gt; <br> `*taskrun`=&lt;taskrun_name&gt;<br> `namespace`=&lt;pipelineruns-taskruns-namespace&gt; | experimental |
-| `tekton_pipelines_controller_taskrun_count` | Counter | `status`=&lt;status&gt; | experimental |
+| `tekton_pipelines_controller_taskrun_count` | Counter | `status`=&lt;status&gt; <br> `*reason`=&lt;reason&gt; | experimental |
 | `tekton_pipelines_controller_running_taskruns_count` | Gauge | | experimental |
 | `tekton_pipelines_controller_running_taskruns_throttled_by_quota_count` | Gauge | | experimental |
 | `tekton_pipelines_controller_running_taskruns_throttled_by_node_count`  | Gauge | | experimental |
@@ -37,6 +37,7 @@ A sample config-map has been provided as [config-observability](./../config/conf
     metrics.taskrun.duration-type: "histogram"
     metrics.pipelinerun.level: "pipeline"
     metrics.pipelinerun.duration-type: "histogram"
+    metrics.count.reason: "false"
 ```
 
 Following values are available in the configmap:
@@ -53,6 +54,7 @@ Following values are available in the configmap:
 | metrics.taskrun.duration-type | `lastvalue` | `tekton_pipelines_controller_pipelinerun_taskrun_duration_seconds` and  `tekton_pipelines_controller_taskrun_duration_seconds` is of type gauge or lastvalue |
 | metrics.pipelinerun.duration-type | `histogram` | `tekton_pipelines_controller_pipelinerun_duration_seconds` is of type histogram |
 | metrics.pipelinerun.duration-type | `lastvalue` | `tekton_pipelines_controller_pipelinerun_duration_seconds` is of type gauge or lastvalue |
+| metrics.count.reason | `false` | Sets if the `reason` label should be included on count metrics |
 
 Histogram value isn't available when pipelinerun or taskrun labels are selected. The Lastvalue or Gauge will be provided. Histogram would serve no purpose because it would generate a single bar. TaskRun and PipelineRun level metrics aren't recommended because they lead to an unbounded cardinality which degrades the observability database.
 

pkg/apis/config/metrics.go

@@ -36,6 +36,9 @@ const (
 	// metrics to use for aggregating duration for pipelinerun
 	metricsDurationPipelinerunType = "metrics.pipelinerun.duration-type"
 
+	// countWithReasonKey sets if the reason label should be included on count metrics
+	countWithReasonKey = "metrics.count.reason"
+
 	// DefaultTaskrunLevel determines to what level to aggregate metrics
 	// when it isn't specified in configmap
 	DefaultTaskrunLevel = TaskrunLevelAtTask
@@ -92,6 +95,7 @@ type Metrics struct {
 	PipelinerunLevel        string
 	DurationTaskrunType     string
 	DurationPipelinerunType string
+	CountWithReason         bool
 }
 
 // GetMetricsConfigName returns the name of the configmap containing all
@@ -113,7 +117,8 @@ func (cfg *Metrics) Equals(other *Metrics) bool {
 	return other.TaskrunLevel == cfg.TaskrunLevel &&
 		other.PipelinerunLevel == cfg.PipelinerunLevel &&
 		other.DurationTaskrunType == cfg.DurationTaskrunType &&
-		other.DurationPipelinerunType == cfg.DurationPipelinerunType
+		other.DurationPipelinerunType == cfg.DurationPipelinerunType &&
+		other.CountWithReason == cfg.CountWithReason
 }
 
 // newMetricsFromMap returns a Config given a map corresponding to a ConfigMap
@@ -123,6 +128,7 @@ func newMetricsFromMap(cfgMap map[string]string) (*Metrics, error) {
 		PipelinerunLevel:        DefaultPipelinerunLevel,
 		DurationTaskrunType:     DefaultDurationTaskrunType,
 		DurationPipelinerunType: DefaultDurationPipelinerunType,
+		CountWithReason:         false,
 	}
 
 	if taskrunLevel, ok := cfgMap[metricsTaskrunLevelKey]; ok {
@@ -138,6 +144,11 @@ func newMetricsFromMap(cfgMap map[string]string) (*Metrics, error) {
 	if durationPipelinerun, ok := cfgMap[metricsDurationPipelinerunType]; ok {
 		tc.DurationPipelinerunType = durationPipelinerun
 	}
+
+	if countWithReason, ok := cfgMap[countWithReasonKey]; ok && countWithReason != "false" {
+		tc.CountWithReason = true
+	}
+
 	return &tc, nil
 }
 

pkg/pipelinerunmetrics/metrics.go

@@ -43,6 +43,7 @@ var (
 	pipelineTag    = tag.MustNewKey("pipeline")
 	namespaceTag   = tag.MustNewKey("namespace")
 	statusTag      = tag.MustNewKey("status")
+	reasonTag      = tag.MustNewKey("reason")
 
 	prDuration = stats.Float64(
 		"pipelinerun_duration_seconds",
@@ -149,11 +150,15 @@ func viewRegister(cfg *config.Metrics) error {
 		TagKeys:     append([]tag.Key{statusTag, namespaceTag}, prunTag...),
 	}
 
+	prCountViewTags := []tag.Key{statusTag}
+	if cfg.CountWithReason {
+		prCountViewTags = append(prCountViewTags, reasonTag)
+	}
 	prCountView = &view.View{
 		Description: prCount.Description(),
 		Measure:     prCount,
 		Aggregation: view.Count(),
-		TagKeys:     []tag.Key{statusTag},
+		TagKeys:     prCountViewTags,
 	}
 	runningPRsCountView = &view.View{
 		Description: runningPRsCount.Description(),
@@ -230,13 +235,15 @@ func (r *Recorder) DurationAndCount(pr *v1.PipelineRun, beforeCondition *apis.Co
 		}
 	}
 
+	cond := pr.Status.GetCondition(apis.ConditionSucceeded)
 	status := "success"
-	if cond := pr.Status.GetCondition(apis.ConditionSucceeded); cond.Status == corev1.ConditionFalse {
+	if cond.Status == corev1.ConditionFalse {
 		status = "failed"
 		if cond.Reason == ReasonCancelled {
 			status = "cancelled"
 		}
 	}
+	reason := cond.Reason
 
 	pipelineName := "anonymous"
 	if pr.Spec.PipelineRef != nil && pr.Spec.PipelineRef.Name != "" {
@@ -245,7 +252,7 @@ func (r *Recorder) DurationAndCount(pr *v1.PipelineRun, beforeCondition *apis.Co
 	ctx, err := tag.New(
 		context.Background(),
 		append([]tag.Mutator{tag.Insert(namespaceTag, pr.Namespace),
-			tag.Insert(statusTag, status)}, r.insertTag(pipelineName, pr.Name)...)...)
+			tag.Insert(statusTag, status), tag.Insert(reasonTag, reason)}, r.insertTag(pipelineName, pr.Name)...)...)
 	if err != nil {
 		return err
 	}

pkg/taskrunmetrics/metrics.go

@@ -50,6 +50,7 @@ var (
 	taskTag        = tag.MustNewKey("task")
 	namespaceTag   = tag.MustNewKey("namespace")
 	statusTag      = tag.MustNewKey("status")
+	reasonTag      = tag.MustNewKey("reason")
 	podTag         = tag.MustNewKey("pod")
 
 	trDurationView                      *view.View
@@ -198,11 +199,16 @@ func viewRegister(cfg *config.Metrics) error {
 		Aggregation: distribution,
 		TagKeys:     append([]tag.Key{statusTag, namespaceTag}, append(trunTag, prunTag...)...),
 	}
+
+	trCountViewTags := []tag.Key{statusTag}
+	if cfg.CountWithReason {
+		trCountViewTags = append(trCountViewTags, reasonTag)
+	}
 	trCountView = &view.View{
 		Description: trCount.Description(),
 		Measure:     trCount,
 		Aggregation: view.Count(),
-		TagKeys:     []tag.Key{statusTag},
+		TagKeys:     trCountViewTags,
 	}
 	runningTRsCountView = &view.View{
 		Description: runningTRsCount.Description(),
@@ -316,13 +322,15 @@ func (r *Recorder) DurationAndCount(ctx context.Context, tr *v1.TaskRun, beforeC
 		taskName = tr.Spec.TaskRef.Name
 	}
 
+	cond := tr.Status.GetCondition(apis.ConditionSucceeded)
 	status := "success"
-	if cond := tr.Status.GetCondition(apis.ConditionSucceeded); cond.Status == corev1.ConditionFalse {
+	if cond.Status == corev1.ConditionFalse {
 		status = "failed"
 	}
+	reason := cond.Reason
 
 	durationStat := trDuration
-	tags := []tag.Mutator{tag.Insert(namespaceTag, tr.Namespace), tag.Insert(statusTag, status)}
+	tags := []tag.Mutator{tag.Insert(namespaceTag, tr.Namespace), tag.Insert(statusTag, status), tag.Insert(reasonTag, reason)}
 	if ok, pipeline, pipelinerun := IsPartOfPipeline(tr); ok {
 		durationStat = prTRDuration
 		tags = append(tags, r.insertPipelineTag(pipeline, pipelinerun)...)