add latency trace interceptor

Author: Cecile Terpin

dd-trace-api/src/main/java/datadog/trace/api/ConfigDefaults.java

@@ -69,6 +69,7 @@ public final class ConfigDefaults {
   static final boolean DEFAULT_DB_CLIENT_HOST_SPLIT_BY_INSTANCE_TYPE_SUFFIX = false;
   static final boolean DEFAULT_DB_CLIENT_HOST_SPLIT_BY_HOST = false;
   static final String DEFAULT_DB_DBM_PROPAGATION_MODE_MODE = "disabled";
+  static final int DEFAULT_TRACE_LATENCY_INTERCEPTOR_VALUE = -1;
   static final int DEFAULT_SCOPE_DEPTH_LIMIT = 100;
   static final int DEFAULT_SCOPE_ITERATION_KEEP_ALIVE = 30; // in seconds
   static final int DEFAULT_PARTIAL_FLUSH_MIN_SPANS = 1000;

dd-trace-api/src/main/java/datadog/trace/api/config/TracerConfig.java

@@ -73,7 +73,7 @@ public final class TracerConfig {
   public static final String TRACE_HTTP_CLIENT_ERROR_STATUSES = "trace.http.client.error.statuses";
 
   public static final String SPLIT_BY_TAGS = "trace.split-by-tags";
-
+  public static final String TRACE_LATENCY_INTERCEPTOR_VALUE = "trace.latency.interceptor.value";
   public static final String SCOPE_DEPTH_LIMIT = "trace.scope.depth.limit";
   public static final String SCOPE_STRICT_MODE = "trace.scope.strict.mode";
   public static final String SCOPE_ITERATION_KEEP_ALIVE = "trace.scope.iteration.keep.alive";

dd-trace-api/src/main/java/datadog/trace/api/interceptor/AbstractTraceInterceptor.java

@@ -22,6 +22,9 @@ public enum Priority {
     DD_INTAKE(2),
     GIT_METADATA(3),
 
+    // trace custom sampling
+    ROOT_SPAN_LATENCY(Integer.MAX_VALUE - 2),
+
     // trace data collection
     CI_VISIBILITY_TELEMETRY(Integer.MAX_VALUE - 1),
     SERVICE_NAME_COLLECTING(Integer.MAX_VALUE);

dd-trace-core/src/main/java/datadog/trace/core/CoreTracer.java

@@ -88,6 +88,7 @@
 import datadog.trace.core.scopemanager.ContinuableScopeManager;
 import datadog.trace.core.taginterceptor.RuleFlags;
 import datadog.trace.core.taginterceptor.TagInterceptor;
+import datadog.trace.core.traceinterceptor.LatencyTraceInterceptor;
 import datadog.trace.lambda.LambdaHandler;
 import datadog.trace.relocate.api.RatelimitedLogger;
 import datadog.trace.util.AgentTaskScheduler;
@@ -745,6 +746,10 @@ private CoreTracer(
       addTraceInterceptor(GitMetadataTraceInterceptor.INSTANCE);
     }
 
+    if (config.isTraceLatencyInterceptorEnabled()) {
+      addTraceInterceptor(LatencyTraceInterceptor.INSTANCE);
+    }
+
     this.instrumentationGateway = instrumentationGateway;
     callbackProviderAppSec = instrumentationGateway.getCallbackProvider(RequestContextSlot.APPSEC);
     callbackProviderIast = instrumentationGateway.getCallbackProvider(RequestContextSlot.IAST);

dd-trace-core/src/main/java/datadog/trace/core/traceinterceptor/LatencyTraceInterceptor.java

@@ -0,0 +1,36 @@
+package datadog.trace.core.traceinterceptor;
+
+import datadog.trace.api.Config;
+import datadog.trace.api.DDTags;
+import datadog.trace.api.interceptor.AbstractTraceInterceptor;
+import datadog.trace.api.interceptor.MutableSpan;
+import datadog.trace.api.interceptor.TraceInterceptor;
+import java.util.Collection;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+public class LatencyTraceInterceptor extends AbstractTraceInterceptor {
+  private static final Logger log = LoggerFactory.getLogger(LatencyTraceInterceptor.class);
+  // duration configured in ms, need to be converted in nano seconds
+  private static final int LATENCY = Config.get().getTraceLatencyInterceptorValue() * 1000000;
+
+  public static final TraceInterceptor INSTANCE =
+      new LatencyTraceInterceptor(Priority.ROOT_SPAN_LATENCY);
+
+  protected LatencyTraceInterceptor(Priority priority) {
+    super(priority);
+  }
+
+  @Override
+  public Collection<? extends MutableSpan> onTraceComplete(
+      Collection<? extends MutableSpan> latencyTrace) {
+    if (latencyTrace.isEmpty()) {
+      return latencyTrace;
+    }
+    MutableSpan rootSpan = latencyTrace.iterator().next().getLocalRootSpan();
+    if (rootSpan != null && rootSpan.getDurationNano() > LATENCY) {
+      rootSpan.setTag(DDTags.MANUAL_KEEP, true);
+    }
+    return latencyTrace;
+  }
+}

dd-trace-core/src/test/groovy/datadog/trace/core/traceinterceptor/LatencyTraceInterceptorTest.groovy

@@ -0,0 +1,46 @@
+package datadog.trace.core.traceinterceptor
+
+import datadog.trace.api.DDTags
+import datadog.trace.common.writer.ListWriter
+
+import datadog.trace.core.test.DDCoreSpecification
+
+import spock.lang.Timeout
+
+@Timeout(10)
+class LatencyTraceInterceptorTest  extends DDCoreSpecification {
+
+
+  def "test set sampling priority according to latency"() {
+    setup:
+    def writer = new ListWriter()
+    def props = new Properties()
+    props.setProperty("trace.partial.flush.enabled", partialFlushEnabled)
+    props.setProperty("trace.latency.interceptor.value", latencyThreshold)
+
+    def tracer = tracerBuilder().withProperties(props).writer(writer).build()
+
+    def spanSetup = tracer.buildSpan("test","my_operation_name").withTag(tagname, true).start()
+    sleep(duration)
+    spanSetup.finish()
+
+    expect:
+    def trace = writer.firstTrace()
+    trace.size() == 1
+    def span = trace[0]
+
+    span.context().getSamplingPriority() == expected
+
+    tracer.close()
+    where:
+    partialFlushEnabled | latencyThreshold | tagname                      | duration    | expected
+    "true"              |       "102"     | DDTags.MANUAL_KEEP            |   100       |  2
+    "true"              |       "102"    | DDTags.MANUAL_DROP             |   100       |  -1
+    "true"              |       "102"     | DDTags.MANUAL_KEEP            |   105       |  2
+    "true"              |       "102"     | DDTags.MANUAL_DROP            |   105       |  -1
+    // "false"             |       "102"     | DDTags.MANUAL_KEEP            |   100       |  2
+    // "false"             |       "102"     | DDTags.MANUAL_DROP            |   100       |  -1
+    // "false"             |       "102"     | DDTags.MANUAL_KEEP            |   105       |  2
+    // "false"             |       "102"     | DDTags.MANUAL_DROP            |   105       |  2
+  }
+}

internal-api/src/main/java/datadog/trace/api/Config.java

@@ -176,6 +176,8 @@ public static String getHostName() {
   private final boolean scopeStrictMode;
   private final int scopeIterationKeepAlive;
   private final int partialFlushMinSpans;
+  private final int traceLatencyInterceptorValue;
+  private final boolean traceLatencyInterceptorEnabled;
   private final boolean traceStrictWritesEnabled;
   private final boolean logExtractHeaderNames;
   private final Set<PropagationStyle> propagationStylesToExtract;
@@ -860,6 +862,12 @@ private Config(final ConfigProvider configProvider, final InstrumenterConfig ins
             ? 0
             : configProvider.getInteger(PARTIAL_FLUSH_MIN_SPANS, DEFAULT_PARTIAL_FLUSH_MIN_SPANS);
 
+    traceLatencyInterceptorValue =
+        configProvider.getInteger(
+            TRACE_LATENCY_INTERCEPTOR_VALUE, DEFAULT_TRACE_LATENCY_INTERCEPTOR_VALUE);
+
+    traceLatencyInterceptorEnabled = !partialFlushEnabled && (traceLatencyInterceptorValue >= 0);
+
     traceStrictWritesEnabled = configProvider.getBoolean(TRACE_STRICT_WRITES_ENABLED, false);
 
     logExtractHeaderNames =
@@ -2075,6 +2083,14 @@ public int getPartialFlushMinSpans() {
     return partialFlushMinSpans;
   }
 
+  public int getTraceLatencyInterceptorValue() {
+    return traceLatencyInterceptorValue;
+  }
+
+  public boolean isTraceLatencyInterceptorEnabled() {
+    return traceLatencyInterceptorEnabled;
+  }
+
   public boolean isTraceStrictWritesEnabled() {
     return traceStrictWritesEnabled;
   }
@@ -4158,6 +4174,10 @@ public String toString() {
         + scopeIterationKeepAlive
         + ", partialFlushMinSpans="
         + partialFlushMinSpans
+        + ", traceLatencyInterceptorEnabled="
+        + traceLatencyInterceptorEnabled
+        + ", traceLatencyInterceptorValue="
+        + traceLatencyInterceptorValue
         + ", traceStrictWritesEnabled="
         + traceStrictWritesEnabled
         + ", tracePropagationStylesToExtract="