refactor(otel): Add X-Ray context extraction with trace ID priority

otel-plugin/pom.xml

@@ -11,7 +11,7 @@
     </parent>
 
     <artifactId>aws-durable-execution-sdk-java-otel</artifactId>
-    <name>AWS Lambda Durable Execution SDK - OpenTelemetry Plugin</name>
+    <name>AWS Lambda Durable Execution SDK OpenTelemetry Plugin</name>
     <description>OpenTelemetry instrumentation plugin for AWS Lambda Durable Execution SDK</description>
 
     <properties>
@@ -48,13 +48,6 @@
             <version>${opentelemetry.version}</version>
         </dependency>
 
-        <!-- AWS X-Ray Propagator (official OTel contrib) -->
-        <dependency>
-            <groupId>io.opentelemetry.contrib</groupId>
-            <artifactId>opentelemetry-aws-xray-propagator</artifactId>
-            <version>1.57.0-alpha</version>
-        </dependency>
-
         <!-- SLF4J for logging -->
         <dependency>
             <groupId>org.slf4j</groupId>

otel-plugin/src/main/java/software/amazon/lambda/durable/otel/ContextExtractor.java

@@ -2,13 +2,11 @@
 // SPDX-License-Identifier: Apache-2.0
 package software.amazon.lambda.durable.otel;
 
-import io.opentelemetry.context.Context;
-
 /**
- * Extracts OTel trace context from the Lambda runtime environment.
+ * Extracts trace context from the Lambda runtime environment.
  *
  * <p>Implementations read trace context from various sources (X-Ray trace header, W3C traceparent, etc.) and return an
- * OTel {@link Context} that can be used as the parent for invocation spans.
+ * {@link ExtractedContext} containing the trace ID and optional parent span ID.
  *
  * <p>Called once per invocation in {@code onInvocationStart} to establish the parent trace context.
  *
@@ -21,7 +19,7 @@ public interface ContextExtractor {
     /**
      * Extracts trace context from the runtime environment.
      *
-     * @return the extracted OTel context, or {@link Context#root()} if no context is available
+     * @return the extracted context, or {@code null} if no context is available
      */
-    Context extract();
+    ExtractedContext extract();
 }

otel-plugin/src/main/java/software/amazon/lambda/durable/otel/DeterministicIdGenerator.java

@@ -11,11 +11,18 @@
 /**
  * Generates deterministic trace and span IDs for durable execution observability.
  *
- * <p>All invocations of the same execution share a single trace ID (derived from the execution ARN). Operations get
- * stable span IDs derived from the execution ARN + operation ID, ensuring the same operation produces the same span
- * across invocations.
+ * <p>Trace ID resolution order:
  *
- * <p>When no pending operation ID is set, falls back to random generation (standard OTel behavior).
+ * <ol>
+ *   <li>If an extracted trace ID is set (from {@code _X_AMZN_TRACE_ID}), use it. The durable execution backend
+ *       propagates the same Root to all invocations, so this naturally unifies the trace.
+ *   <li>If no extracted trace ID is available (local tests, non-Lambda environments), derive a deterministic trace ID
+ *       from the execution ARN using SHA-256.
+ *   <li>If neither is set, fall back to random generation.
+ * </ol>
+ *
+ * <p>Span IDs for operations are deterministic (derived from execution ARN + operation ID), ensuring the same operation
+ * produces the same span across invocations. When no pending operation ID is set, falls back to random generation.
  *
  * @deprecated This is a preview API that is experimental and may be changed or removed in future releases.
  */
@@ -24,18 +31,30 @@ public class DeterministicIdGenerator implements IdGenerator {
 
     private static final IdGenerator RANDOM = IdGenerator.random();
 
-    private final AtomicReference<String> executionTraceId = new AtomicReference<>(null);
+    private final AtomicReference<String> extractedTraceId = new AtomicReference<>(null);
+    private final AtomicReference<String> arnDerivedTraceId = new AtomicReference<>(null);
     private final ThreadLocal<String> pendingSpanOperationId = new ThreadLocal<>();
     private final AtomicReference<String> durableExecutionArn = new AtomicReference<>(null);
 
     /**
-     * Sets the execution ARN used for generating deterministic IDs.
+     * Sets an externally extracted trace ID (e.g., from the X-Ray trace header). This takes highest priority for trace
+     * ID generation.
+     *
+     * @param traceId 32-char lowercase hex trace ID
+     */
+    public void setExtractedTraceId(String traceId) {
+        this.extractedTraceId.set(traceId);
+    }
+
+    /**
+     * Sets the execution ARN used for generating deterministic IDs. Computes and caches an ARN-derived trace ID as
+     * fallback when no extracted trace ID is available.
      *
      * @param arn the durable execution ARN
      */
     public void setDurableExecutionArn(String arn) {
         this.durableExecutionArn.set(arn);
-        this.executionTraceId.set(generateTraceIdFromArn(arn));
+        this.arnDerivedTraceId.set(generateTraceIdFromArn(arn));
     }
 
     /**
@@ -50,9 +69,6 @@ public void setNextSpanOperationId(String operationId) {
     /**
      * Generates a deterministic span ID for a given operation ID without consuming the ThreadLocal state.
      *
-     * <p>Used for creating non-recording placeholder spans when a parent operation's span context is needed but hasn't
-     * been exported yet.
-     *
      * @param operationId the operation ID to derive the span ID from
      * @return a deterministic 16-char hex span ID
      */
@@ -62,10 +78,17 @@ public String generateSpanIdForOperation(String operationId) {
 
     @Override
     public String generateTraceId() {
-        var cached = executionTraceId.get();
-        if (cached != null) {
-            return cached;
+        // Priority 1: extracted from X-Ray header (backend propagates same Root across invocations)
+        var extracted = extractedTraceId.get();
+        if (extracted != null) {
+            return extracted;
+        }
+        // Priority 2: deterministic from execution ARN (local tests, non-Lambda)
+        var arnDerived = arnDerivedTraceId.get();
+        if (arnDerived != null) {
+            return arnDerived;
         }
+        // Priority 3: random fallback
         return RANDOM.generateTraceId();
     }
 
@@ -79,27 +102,19 @@ public String generateSpanId() {
         return RANDOM.generateSpanId();
     }
 
-    /**
-     * Generates a deterministic trace ID from an execution ARN.
-     *
-     * <p>Uses SHA-256 hash truncated to 32 hex chars (128 bits) for the trace ID.
-     */
+    /** Generates a deterministic trace ID from an execution ARN using SHA-256 truncated to 32 hex chars. */
     private String generateTraceIdFromArn(String arn) {
         var hash = sha256(arn);
-        // Trace ID is 32 hex chars (16 bytes)
         return hash.substring(0, 32);
     }
 
     /**
-     * Generates a deterministic span ID from the execution ARN + operation ID.
-     *
-     * <p>Uses SHA-256 hash truncated to 16 hex chars (64 bits) for the span ID.
+     * Generates a deterministic span ID from the execution ARN + operation ID using SHA-256 truncated to 16 hex chars.
      */
     private String generateSpanIdFromOperation(String operationId) {
         var arn = durableExecutionArn.get();
         var input = arn != null ? arn + ":" + operationId : operationId;
         var hash = sha256(input);
-        // Span ID is 16 hex chars (8 bytes)
         return hash.substring(0, 16);
     }
 

otel-plugin/src/main/java/software/amazon/lambda/durable/otel/ExtractedContext.java

@@ -0,0 +1,17 @@
+// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0
+package software.amazon.lambda.durable.otel;
+
+/**
+ * Trace context extracted from the Lambda runtime environment.
+ *
+ * <p>Contains the trace ID (always present) and an optional parent span ID. When the durable execution backend
+ * propagates the same X-Ray Root across all invocations, the trace ID will be consistent, enabling spans from different
+ * invocations to be stitched into a single trace.
+ *
+ * @param traceId 32-character lowercase hex trace ID (OTel format, no dashes)
+ * @param parentSpanId 16-character lowercase hex parent span ID (may be null if no parent available)
+ * @deprecated This is a preview API that is experimental and may be changed or removed in future releases.
+ */
+@Deprecated
+public record ExtractedContext(String traceId, String parentSpanId) {}

otel-plugin/src/main/java/software/amazon/lambda/durable/otel/OpenTelemetryDurablePlugin.java

@@ -39,7 +39,22 @@
  *   <li><b>Attempt span</b> — one per user function execution (step attempt, child context run)
  * </ul>
  *
- * <p>Uses deterministic span/trace IDs so all invocations of the same execution share a single trace.
+ * <p>Trace ID resolution:
+ *
+ * <ol>
+ *   <li>Uses the X-Ray trace ID from {@code _X_AMZN_TRACE_ID} when available. The durable execution backend propagates
+ *       the same Root to all invocations of the same execution, naturally unifying the trace.
+ *   <li>Falls back to a deterministic trace ID derived from the execution ARN (for local tests or non-Lambda
+ *       environments).
+ * </ol>
+ *
+ * <p>Requires the ADOT Lambda Layer for trace export. Configure with:
+ *
+ * <ul>
+ *   <li>Lambda Layer: {@code aws-otel-java-agent} (ADOT Java auto-instrumentation layer)
+ *   <li>Env var: {@code AWS_LAMBDA_EXEC_WRAPPER=/opt/otel-handler}
+ *   <li>Tracing: Active (to populate {@code _X_AMZN_TRACE_ID})
+ * </ul>
  *
  * <p>Thread-safe: uses {@link ConcurrentHashMap} for span/scope storage since the SDK runs user code on multiple
  * threads.
@@ -75,6 +90,17 @@ public class OpenTelemetryDurablePlugin implements DurableExecutionPlugin {
     /**
      * Creates an OTel plugin with default settings: X-Ray context extraction, MDC enabled.
      *
+     * <p>Uses the provided tracer provider builder. Customers configure exporters and span processors on the builder —
+     * the plugin handles ID generation.
+     *
+     * <p>For ADOT layer usage, configure with an OTLP exporter:
+     *
+     * <pre>{@code
+     * var otlpExporter = OtlpGrpcSpanExporter.getDefault(); // sends to localhost:4317
+     * var plugin = new OpenTelemetryDurablePlugin(
+     *     SdkTracerProvider.builder().addSpanProcessor(SimpleSpanProcessor.create(otlpExporter)));
+     * }</pre>
+     *
      * @param tracerProviderBuilder the tracer provider builder (ID generator will be overridden)
      */
     public OpenTelemetryDurablePlugin(SdkTracerProviderBuilder tracerProviderBuilder) {
@@ -95,10 +121,6 @@ public OpenTelemetryDurablePlugin(
     /**
      * Creates an OTel plugin with full configuration.
      *
-     * <p>The plugin internally creates a {@link DeterministicIdGenerator} and sets it on the provided builder before
-     * building the tracer provider. Customers configure exporters, span processors, and samplers on the builder — the
-     * plugin handles ID generation.
-     *
      * @param tracerProviderBuilder the tracer provider builder (ID generator will be overridden)
      * @param contextExtractor extracts parent trace context from the Lambda environment
      * @param enableMdc if true, injects trace_id/span_id into SLF4J MDC for log correlation
@@ -117,14 +139,36 @@ public OpenTelemetryDurablePlugin(
     @Override
     public void onInvocationStart(InvocationInfo info) {
         this.durableExecutionArn = info.durableExecutionArn();
+
+        // Set execution ARN for deterministic span ID generation
         idGenerator.setDurableExecutionArn(info.durableExecutionArn());
 
-        // Extract parent context from Lambda environment (X-Ray, W3C, etc.)
-        var extractedParentContext = contextExtractor.extract();
+        // Extract trace context from environment (X-Ray header)
+        var extractedContext = contextExtractor.extract();
 
-        // Create invocation span as child of extracted context
+        if (extractedContext != null) {
+            // Use the X-Ray trace ID — backend propagates same Root across all invocations
+            idGenerator.setExtractedTraceId(extractedContext.traceId());
+        }
+        // If no extracted context, idGenerator falls back to ARN-derived trace ID
+
+        // Determine parent context for the invocation span
+        Context parentContext;
+        if (extractedContext != null && extractedContext.parentSpanId() != null) {
+            // Create a remote parent span context from the X-Ray Parent field
+            var parentSpanContext = SpanContext.createFromRemoteParent(
+                    extractedContext.traceId(),
+                    extractedContext.parentSpanId(),
+                    TraceFlags.getSampled(),
+                    TraceState.getDefault());
+            parentContext = Context.root().with(Span.wrap(parentSpanContext));
+        } else {
+            parentContext = Context.root();
+        }
+
+        // Create invocation span as child of Lambda's X-Ray segment (via Parent field)
         var spanBuilder = tracer.spanBuilder("durable.invocation")
-                .setParent(extractedParentContext)
+                .setParent(parentContext)
                 .setAttribute(DURABLE_EXECUTION_ARN, info.durableExecutionArn())
                 .setAttribute(DURABLE_FIRST_INVOCATION, info.isFirstInvocation());