[processor/tailsampling] Improved "not sampled" decision cache usage (open-telemetry#37189)

Currently, a "not sampled" decision for a trace is only cached when late
spans arrive for it. Whereas a "sampled" decision for a trace is
immediately cached. This pull request immediately caches "not sampled"
decisions, late spans (or reprocessed spans) for "not sampled" traces
now take fewer resources to process.

Currently, traces are stored in memory until `num_traces` is reached,
then the oldest is deleted to make room. This is a clever mechanism to
limit the size* of the internal map, however, trace span counts and
their contents likely vary considerably. This pull request doesn't touch
this mechanism, but it does delete traces from the internal map after
they are released and are added to a decision cache.

---------

Signed-off-by: Sean Porter <[email protected]>

Author: portertech

.chloggen/tailsamplingprocessor-improved-not-sampled-decision-cache.yaml

@@ -0,0 +1,27 @@
+# Use this changelog template to create an entry for release notes.
+
+# One of 'breaking', 'deprecation', 'new_component', 'enhancement', 'bug_fix'
+change_type: enhancement
+
+# The name of the component, or a single word describing the area of concern, (e.g. filelogreceiver)
+component: tailsamplingprocessor
+
+# A brief description of the change.  Surround your text with quotes ("") if it needs to start with a backtick (`).
+note: Improved not sampled decision cache usage and deleting traces from the internal map when they are in a decision cache.
+
+# Mandatory: One or more tracking issues related to the change. You can use the PR number here if no issue exists.
+issues: [37189]
+
+# (Optional) One or more lines of additional information to render under the primary note.
+# These lines will be padded with 2 spaces and then inserted directly into the document.
+# Use pipe (|) for multiline entries.
+subtext:
+
+# If your change doesn't affect end users or the exported elements of any package,
+# you should instead start your pull request title with [chore] or use the "Skip Changelog" label.
+# Optional: The change log or logs in which this entry should be included.
+# e.g. '[user]' or '[user, api]'
+# Include 'user' if the change is relevant to end users.
+# Include 'api' if there is a change to a library API.
+# Default: '[user]'
+change_logs: [user]

processor/tailsamplingprocessor/processor.go

@@ -350,8 +350,11 @@ func (tsp *tailSamplingSpanProcessor) samplingPolicyOnTick() {
 		trace.ReceivedBatches = ptrace.NewTraces()
 		trace.Unlock()
 
-		if decision == sampling.Sampled {
+		switch decision {
+		case sampling.Sampled:
 			tsp.releaseSampledTrace(context.Background(), id, allSpans)
+		case sampling.NotSampled:
+			tsp.releaseNotSampledTrace(id)
 		}
 	}
 
@@ -517,7 +520,7 @@ func (tsp *tailSamplingSpanProcessor) processTraces(resourceSpans ptrace.Resourc
 				appendToTraces(traceTd, resourceSpans, spans)
 				tsp.releaseSampledTrace(tsp.ctx, id, traceTd)
 			case sampling.NotSampled:
-				tsp.nonSampledIDCache.Put(id, true)
+				tsp.releaseNotSampledTrace(id)
 			default:
 				tsp.logger.Warn("Encountered unexpected sampling decision",
 					zap.Int("decision", int(finalDecision)))
@@ -565,16 +568,30 @@ func (tsp *tailSamplingSpanProcessor) dropTrace(traceID pcommon.TraceID, deletio
 	tsp.telemetry.ProcessorTailSamplingSamplingTraceRemovalAge.Record(tsp.ctx, int64(deletionTime.Sub(trace.ArrivalTime)/time.Second))
 }
 
-// releaseSampledTrace sends the trace data to the next consumer.
-// It additionally adds the trace ID to the cache of sampled trace IDs.
-// It does not (yet) delete the spans from the internal map.
+// releaseSampledTrace sends the trace data to the next consumer. It
+// additionally adds the trace ID to the cache of sampled trace IDs. If the
+// trace ID is cached, it deletes the spans from the internal map.
 func (tsp *tailSamplingSpanProcessor) releaseSampledTrace(ctx context.Context, id pcommon.TraceID, td ptrace.Traces) {
 	tsp.sampledIDCache.Put(id, true)
 	if err := tsp.nextConsumer.ConsumeTraces(ctx, td); err != nil {
 		tsp.logger.Warn(
 			"Error sending spans to destination",
 			zap.Error(err))
 	}
+	_, ok := tsp.sampledIDCache.Get(id)
+	if ok {
+		tsp.dropTrace(id, time.Now())
+	}
+}
+
+// releaseNotSampledTrace adds the trace ID to the cache of not sampled trace
+// IDs. If the trace ID is cached, it deletes the spans from the internal map.
+func (tsp *tailSamplingSpanProcessor) releaseNotSampledTrace(id pcommon.TraceID) {
+	tsp.nonSampledIDCache.Put(id, true)
+	_, ok := tsp.nonSampledIDCache.Get(id)
+	if ok {
+		tsp.dropTrace(id, time.Now())
+	}
 }
 
 func appendToTraces(dest ptrace.Traces, rss ptrace.ResourceSpans, spanAndScopes []spanAndScope) {

processor/tailsamplingprocessor/processor_decisions_test.go

@@ -6,7 +6,6 @@ package tailsamplingprocessor
 import (
 	"context"
 	"testing"
-	"time"
 
 	"github.com/stretchr/testify/require"
 	"go.opentelemetry.io/collector/component/componenttest"
@@ -386,8 +385,7 @@ func TestLateArrivingSpanUsesDecisionCache(t *testing.T) {
 	// The final decision SHOULD be Sampled.
 	require.EqualValues(t, 1, nextConsumer.SpanCount())
 
-	// Drop the trace to force cache to make decision
-	tsp.dropTrace(traceID, time.Now())
+	// The trace should have been dropped after its id was added to the decision cache
 	_, ok := tsp.idToTrace.Load(traceID)
 	require.False(t, ok)
 
@@ -461,8 +459,7 @@ func TestLateSpanUsesNonSampledDecisionCache(t *testing.T) {
 	// The final decision SHOULD be NOT Sampled.
 	require.EqualValues(t, 0, nextConsumer.SpanCount())
 
-	// Drop the trace to force cache to make decision
-	tsp.dropTrace(traceID, time.Now())
+	// The trace should have been dropped after its id was added to the decision cache
 	_, ok := tsp.idToTrace.Load(traceID)
 	require.False(t, ok)