fn: introducing lb placer basic metrics (#1058)

* fn: introducing lb placer basic metrics

This change adds basic metrics to naive and consistent
hash LB placers. The stats show how many times we scanned
the full runner list, if runner pool failed to return a
runner list or if runner pool returned an empty list.

Placed and not placed status are also tracked along with
if TryExec returned an error or not. Most common error
code, Too-Busy is specifically tracked.

If client cancels/times out, this is also tracked as
a client cancel metric.

For placer latency, we would like to know how much time
the placer spent on searching for a runner until it
successfully places a call. This includes round-trip
times for NACK responses from the runners until a successful
TryExec() call. By excluding last successful TryExec() latency,
we try to exclude function execution & runner container
startup time from this metric in an attempt to isolate
Placer only latency.

* fn: latency and attempt tracker

Removing full scan metric. Tracking number of
runners attempted is a better metric for this
purpose.

Also, if rp.Runners() fail, this is an unrecoverable
error and we should bail out instead of retrying.

* fn: typo fix, ch placer finalize err return

* fn: enable LB placer metrics in WithAgentFromEnv if prometheus is enabled

api/runnerpool/placer_stats.go

@@ -0,0 +1,103 @@
+package runnerpool
+
+import (
+	"context"
+	"math"
+	"time"
+
+	"github.com/sirupsen/logrus"
+	"go.opencensus.io/stats"
+	"go.opencensus.io/stats/view"
+	"go.opencensus.io/tag"
+)
+
+var (
+	attemptCountMeasure      = stats.Int64("lb_placer_attempt_count", "LB Placer Number of Runners Attempted Count", "")
+	errorPoolCountMeasure    = stats.Int64("lb_placer_rp_error_count", "LB Placer RunnerPool RunnerList Error Count", "")
+	emptyPoolCountMeasure    = stats.Int64("lb_placer_rp_empty_count", "LB Placer RunnerPool RunnerList Empty Count", "")
+	cancelCountMeasure       = stats.Int64("lb_placer_client_cancelled_count", "LB Placer Client Cancel Count", "")
+	placedErrorCountMeasure  = stats.Int64("lb_placer_placed_error_count", "LB Placer Placed Call Count With Errors", "")
+	placedOKCountMeasure     = stats.Int64("lb_placer_placed_ok_count", "LB Placer Placed Call Count Without Errors", "")
+	retryTooBusyCountMeasure = stats.Int64("lb_placer_retry_busy_count", "LB Placer Retry Count - Too Busy", "")
+	retryErrorCountMeasure   = stats.Int64("lb_placer_retry_error_count", "LB Placer Retry Count - Errors", "")
+	placerLatencyMeasure     = stats.Int64("lb_placer_latency", "LB Placer Latency", "msecs")
+)
+
+// Helper struct for tracking LB Placer latency and attempt counts
+type attemptTracker struct {
+	ctx             context.Context
+	startTime       time.Time
+	lastAttemptTime time.Time
+	attemptCount    int64
+}
+
+func newAttemptTracker(ctx context.Context) *attemptTracker {
+	return &attemptTracker{
+		ctx:       ctx,
+		startTime: time.Now(),
+	}
+}
+
+func (data *attemptTracker) finalizeAttempts(isSuccess bool) {
+	stats.Record(data.ctx, attemptCountMeasure.M(data.attemptCount))
+
+	// IMPORTANT: here we use (lastAttemptTime - startTime). We want to exclude TryExec
+	// latency *if* TryExec() goes through with success. Placer latency metric only shows
+	// how much time are spending in Placer loop/retries. The metric includes rtt/latency of
+	// *all* unsuccessful NACK (retriable) responses from runners as well. For example, if
+	// Placer loop here retries 4 runners (which takes 5 msecs each) and then 5th runner
+	// succeeds (but takes 35 seconds to finish execution), we report 20 msecs as our LB
+	// latency.
+	endTime := data.lastAttemptTime
+	if !isSuccess {
+		endTime = time.Now()
+	}
+
+	stats.Record(data.ctx, placerLatencyMeasure.M(int64(endTime.Sub(data.startTime)/time.Millisecond)))
+}
+
+func (data *attemptTracker) recordAttempt() {
+	data.lastAttemptTime = time.Now()
+	if data.attemptCount != math.MaxInt64 {
+		data.attemptCount++
+	}
+}
+
+func makeKeys(names []string) []tag.Key {
+	var tagKeys []tag.Key
+	for _, name := range names {
+		key, err := tag.NewKey(name)
+		if err != nil {
+			logrus.WithError(err).Fatal("cannot create tag key for %v", name)
+		}
+		tagKeys = append(tagKeys, key)
+	}
+	return tagKeys
+}
+
+func createView(measure stats.Measure, agg *view.Aggregation, tagKeys []string) *view.View {
+	return &view.View{
+		Name:        measure.Name(),
+		Description: measure.Description(),
+		TagKeys:     makeKeys(tagKeys),
+		Measure:     measure,
+		Aggregation: agg,
+	}
+}
+
+func RegisterPlacerViews(tagKeys []string) {
+	err := view.Register(
+		createView(attemptCountMeasure, view.Distribution(0, 1, 2, 4, 8, 32, 64, 256), tagKeys),
+		createView(errorPoolCountMeasure, view.Count(), tagKeys),
+		createView(emptyPoolCountMeasure, view.Count(), tagKeys),
+		createView(cancelCountMeasure, view.Count(), tagKeys),
+		createView(placedErrorCountMeasure, view.Count(), tagKeys),
+		createView(placedOKCountMeasure, view.Count(), tagKeys),
+		createView(retryTooBusyCountMeasure, view.Count(), tagKeys),
+		createView(retryErrorCountMeasure, view.Count(), tagKeys),
+		createView(placerLatencyMeasure, view.Distribution(1, 10, 25, 50, 200, 1000, 10000, 60000), tagKeys),
+	)
+	if err != nil {
+		logrus.WithError(err).Fatal("cannot create view")
+	}
+}