[Inference API] Propagate product use case http header to EIS

docs/changelog/124025.yaml

@@ -0,0 +1,5 @@
+pr: 124025
+summary: "[Inference API] Propagate product use case http header to EIS"
+area: Machine Learning
+type: enhancement
+issues: []

server/src/main/java/org/elasticsearch/TransportVersions.java

@@ -185,6 +185,7 @@ static TransportVersion def(int id) {
     public static final TransportVersion INCLUDE_INDEX_MODE_IN_GET_DATA_STREAM = def(9_023_0_00);
     public static final TransportVersion MAX_OPERATION_SIZE_REJECTIONS_ADDED = def(9_024_0_00);
     public static final TransportVersion RETRY_ILM_ASYNC_ACTION_REQUIRE_ERROR = def(9_025_0_00);
+    public static final TransportVersion INFERENCE_CONTEXT = def(9_026_0_00);
 
     /*
      * STOP! READ THIS FIRST! No, really,

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/inference/InferenceContext.java

@@ -0,0 +1,50 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License
+ * 2.0; you may not use this file except in compliance with the Elastic License
+ * 2.0.
+ */
+
+package org.elasticsearch.xpack.core.inference;
+
+import org.elasticsearch.common.io.stream.StreamInput;
+import org.elasticsearch.common.io.stream.StreamOutput;
+import org.elasticsearch.common.io.stream.Writeable;
+import org.elasticsearch.xcontent.ToXContent;
+import org.elasticsearch.xcontent.XContentBuilder;
+
+import java.io.IOException;
+
+/**
+ * Record for storing context alongside an inference request, typically used for metadata.
+ * This is mainly used to pass along inference context on the transport layer without relying on {@link org.elasticsearch.common.util.concurrent.ThreadContext},
+ * which depending on the internal {@link org.elasticsearch.client.internal.Client} throws away parts of the context, when passed along the transport layer.
+ *
+ * @param productUseCase - for now mainly interesting to Elastic Inference Service
+ */
+public record InferenceContext(String productUseCase) implements Writeable, ToXContent {
+
+    public InferenceContext(StreamInput in) throws IOException {
+        this(in.readString());
+    }
+
+    public static InferenceContext empty() {
+        return new InferenceContext("");
+    }
+
+    @Override
+    public void writeTo(StreamOutput out) throws IOException {
+        out.writeString(productUseCase);
+    }
+
+    @Override
+    public XContentBuilder toXContent(XContentBuilder builder, Params params) throws IOException {
+        builder.startObject();
+
+        builder.field("product_use_case", productUseCase);
+
+        builder.endObject();
+
+        return builder;
+    }
+}

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/inference/action/BaseInferenceActionRequest.java

@@ -12,8 +12,10 @@
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
 import org.elasticsearch.inference.TaskType;
+import org.elasticsearch.xpack.core.inference.InferenceContext;
 
 import java.io.IOException;
+import java.util.Objects;
 
 /**
  * Base class for inference action requests. Tracks request routing state to prevent potential routing loops
@@ -23,8 +25,11 @@ public abstract class BaseInferenceActionRequest extends ActionRequest {
 
     private boolean hasBeenRerouted;
 
-    public BaseInferenceActionRequest() {
+    private final InferenceContext context;
+
+    public BaseInferenceActionRequest(InferenceContext context) {
         super();
+        this.context = context;
     }
 
     public BaseInferenceActionRequest(StreamInput in) throws IOException {
@@ -36,6 +41,12 @@ public BaseInferenceActionRequest(StreamInput in) throws IOException {
             // a version pre-node-local-rate-limiting as already rerouted to maintain pre-node-local-rate-limiting behavior.
             this.hasBeenRerouted = true;
         }
+
+        if (in.getTransportVersion().onOrAfter(TransportVersions.INFERENCE_CONTEXT)) {
+            this.context = new InferenceContext(in);
+        } else {
+            this.context = InferenceContext.empty();
+        }
     }
 
     public abstract boolean isStreaming();
@@ -52,11 +63,32 @@ public boolean hasBeenRerouted() {
         return hasBeenRerouted;
     }
 
+    public InferenceContext getContext() {
+        return context;
+    }
+
     @Override
     public void writeTo(StreamOutput out) throws IOException {
         super.writeTo(out);
         if (out.getTransportVersion().onOrAfter(TransportVersions.INFERENCE_REQUEST_ADAPTIVE_RATE_LIMITING)) {
             out.writeBoolean(hasBeenRerouted);
         }
+
+        if (out.getTransportVersion().onOrAfter(TransportVersions.INFERENCE_CONTEXT)) {
+            context.writeTo(out);
+        }
+    }
+
+    @Override
+    public boolean equals(Object o) {
+        if (this == o) return true;
+        if (o == null || getClass() != o.getClass()) return false;
+        BaseInferenceActionRequest that = (BaseInferenceActionRequest) o;
+        return hasBeenRerouted == that.hasBeenRerouted && Objects.equals(context, that.context);
+    }
+
+    @Override
+    public int hashCode() {
+        return Objects.hash(hasBeenRerouted, context);
     }
 }

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/inference/action/InferenceAction.java

@@ -28,6 +28,7 @@
 import org.elasticsearch.xcontent.ParseField;
 import org.elasticsearch.xcontent.ToXContent;
 import org.elasticsearch.xcontent.XContentParser;
+import org.elasticsearch.xpack.core.inference.InferenceContext;
 import org.elasticsearch.xpack.core.inference.results.LegacyTextEmbeddingResults;
 import org.elasticsearch.xpack.core.inference.results.SparseEmbeddingResults;
 import org.elasticsearch.xpack.core.ml.inference.results.TextExpansionResults;
@@ -74,12 +75,14 @@ public static class Request extends BaseInferenceActionRequest {
             InputType.UNSPECIFIED
         );
 
-        public static Builder parseRequest(String inferenceEntityId, TaskType taskType, XContentParser parser) throws IOException {
+        public static Builder parseRequest(String inferenceEntityId, TaskType taskType, InferenceContext context, XContentParser parser)
+            throws IOException {
             Request.Builder builder = PARSER.apply(parser, null);
             builder.setInferenceEntityId(inferenceEntityId);
             builder.setTaskType(taskType);
             // For rest requests we won't know what the input type is
             builder.setInputType(InputType.UNSPECIFIED);
+            builder.setContext(context);
             return builder;
         }
 
@@ -102,6 +105,21 @@ public Request(
             TimeValue inferenceTimeout,
             boolean stream
         ) {
+            this(taskType, inferenceEntityId, query, input, taskSettings, inputType, inferenceTimeout, stream, InferenceContext.empty());
+        }
+
+        public Request(
+            TaskType taskType,
+            String inferenceEntityId,
+            String query,
+            List<String> input,
+            Map<String, Object> taskSettings,
+            InputType inputType,
+            TimeValue inferenceTimeout,
+            boolean stream,
+            InferenceContext context
+        ) {
+            super(context);
             this.taskType = taskType;
             this.inferenceEntityId = inferenceEntityId;
             this.query = query;
@@ -241,19 +259,31 @@ static InputType getInputTypeToWrite(InputType inputType, TransportVersion versi
         public boolean equals(Object o) {
             if (this == o) return true;
             if (o == null || getClass() != o.getClass()) return false;
+            if (super.equals(o) == false) return false;
             Request request = (Request) o;
-            return taskType == request.taskType
+            return stream == request.stream
+                && taskType == request.taskType
                 && Objects.equals(inferenceEntityId, request.inferenceEntityId)
+                && Objects.equals(query, request.query)
                 && Objects.equals(input, request.input)
                 && Objects.equals(taskSettings, request.taskSettings)
-                && Objects.equals(inputType, request.inputType)
-                && Objects.equals(query, request.query)
+                && inputType == request.inputType
                 && Objects.equals(inferenceTimeout, request.inferenceTimeout);
         }
 
         @Override
         public int hashCode() {
-            return Objects.hash(taskType, inferenceEntityId, input, taskSettings, inputType, query, inferenceTimeout);
+            return Objects.hash(
+                super.hashCode(),
+                taskType,
+                inferenceEntityId,
+                query,
+                input,
+                taskSettings,
+                inputType,
+                inferenceTimeout,
+                stream
+            );
         }
 
         public static class Builder {
@@ -266,6 +296,7 @@ public static class Builder {
             private String query;
             private TimeValue timeout = DEFAULT_TIMEOUT;
             private boolean stream = false;
+            private InferenceContext context;
 
             private Builder() {}
 
@@ -313,8 +344,13 @@ public Builder setStream(boolean stream) {
                 return this;
             }
 
+            public Builder setContext(InferenceContext context) {
+                this.context = context;
+                return this;
+            }
+
             public Request build() {
-                return new Request(taskType, inferenceEntityId, query, input, taskSettings, inputType, timeout, stream);
+                return new Request(taskType, inferenceEntityId, query, input, taskSettings, inputType, timeout, stream, context);
             }
         }
 
@@ -333,6 +369,8 @@ public String toString() {
                 + this.getInputType()
                 + ", timeout="
                 + this.getInferenceTimeout()
+                + ", context="
+                + this.getContext()
                 + ")";
         }
     }

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/inference/action/InferenceActionProxy.java

@@ -7,6 +7,7 @@
 
 package org.elasticsearch.xpack.core.inference.action;
 
+import org.elasticsearch.TransportVersions;
 import org.elasticsearch.action.ActionRequest;
 import org.elasticsearch.action.ActionRequestValidationException;
 import org.elasticsearch.action.ActionType;
@@ -17,6 +18,7 @@
 import org.elasticsearch.core.TimeValue;
 import org.elasticsearch.inference.TaskType;
 import org.elasticsearch.xcontent.XContentType;
+import org.elasticsearch.xpack.core.inference.InferenceContext;
 
 import java.io.IOException;
 import java.util.Objects;
@@ -44,21 +46,24 @@ public static class Request extends ActionRequest {
         private final XContentType contentType;
         private final TimeValue timeout;
         private final boolean stream;
+        private final InferenceContext context;
 
         public Request(
             TaskType taskType,
             String inferenceEntityId,
             BytesReference content,
             XContentType contentType,
             TimeValue timeout,
-            boolean stream
+            boolean stream,
+            InferenceContext context
         ) {
             this.taskType = taskType;
             this.inferenceEntityId = inferenceEntityId;
             this.content = content;
             this.contentType = contentType;
             this.timeout = timeout;
             this.stream = stream;
+            this.context = context;
         }
 
         public Request(StreamInput in) throws IOException {
@@ -71,6 +76,12 @@ public Request(StreamInput in) throws IOException {
 
             // streaming is not supported yet for transport traffic
             this.stream = false;
+
+            if (in.getTransportVersion().onOrAfter(TransportVersions.INFERENCE_CONTEXT)) {
+                this.context = new InferenceContext(in);
+            } else {
+                this.context = InferenceContext.empty();
+            }
         }
 
         public TaskType getTaskType() {
@@ -97,6 +108,10 @@ public boolean isStreaming() {
             return stream;
         }
 
+        public InferenceContext getContext() {
+            return context;
+        }
+
         @Override
         public ActionRequestValidationException validate() {
             return null;
@@ -110,6 +125,10 @@ public void writeTo(StreamOutput out) throws IOException {
             out.writeBytesReference(content);
             XContentHelper.writeTo(out, contentType);
             out.writeTimeValue(timeout);
+
+            if (out.getTransportVersion().onOrAfter(TransportVersions.INFERENCE_CONTEXT)) {
+                context.writeTo(out);
+            }
         }
 
         @Override
@@ -122,12 +141,13 @@ public boolean equals(Object o) {
                 && Objects.equals(content, request.content)
                 && contentType == request.contentType
                 && timeout == request.timeout
-                && stream == request.stream;
+                && stream == request.stream
+                && context == request.context;
         }
 
         @Override
         public int hashCode() {
-            return Objects.hash(taskType, inferenceEntityId, content, contentType, timeout, stream);
+            return Objects.hash(taskType, inferenceEntityId, content, contentType, timeout, stream, context);
         }
     }
 }

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/inference/action/UnifiedCompletionAction.java

@@ -15,6 +15,7 @@
 import org.elasticsearch.inference.TaskType;
 import org.elasticsearch.inference.UnifiedCompletionRequest;
 import org.elasticsearch.xcontent.XContentParser;
+import org.elasticsearch.xpack.core.inference.InferenceContext;
 
 import java.io.IOException;
 import java.util.Objects;
@@ -28,10 +29,15 @@ public UnifiedCompletionAction() {
     }
 
     public static class Request extends BaseInferenceActionRequest {
-        public static Request parseRequest(String inferenceEntityId, TaskType taskType, TimeValue timeout, XContentParser parser)
-            throws IOException {
+        public static Request parseRequest(
+            String inferenceEntityId,
+            TaskType taskType,
+            TimeValue timeout,
+            InferenceContext context,
+            XContentParser parser
+        ) throws IOException {
             var unifiedRequest = UnifiedCompletionRequest.PARSER.apply(parser, null);
-            return new Request(inferenceEntityId, taskType, unifiedRequest, timeout);
+            return new Request(inferenceEntityId, taskType, unifiedRequest, context, timeout);
         }
 
         private final String inferenceEntityId;
@@ -40,6 +46,17 @@ public static Request parseRequest(String inferenceEntityId, TaskType taskType,
         private final TimeValue timeout;
 
         public Request(String inferenceEntityId, TaskType taskType, UnifiedCompletionRequest unifiedCompletionRequest, TimeValue timeout) {
+            this(inferenceEntityId, taskType, unifiedCompletionRequest, InferenceContext.empty(), timeout);
+        }
+
+        public Request(
+            String inferenceEntityId,
+            TaskType taskType,
+            UnifiedCompletionRequest unifiedCompletionRequest,
+            InferenceContext context,
+            TimeValue timeout
+        ) {
+            super(context);
             this.inferenceEntityId = Objects.requireNonNull(inferenceEntityId);
             this.taskType = Objects.requireNonNull(taskType);
             this.unifiedCompletionRequest = Objects.requireNonNull(unifiedCompletionRequest);