[ML] Add option to disable inference process cache by default (#108784)

* Add option to disable inference process cache by default

* Add test

* improve tests

* Update docs and improve code

---------

Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>

docs/reference/ml/trained-models/apis/start-trained-model-deployment.asciidoc

@@ -25,9 +25,9 @@ Currently only `pytorch` models are supported for deployment. Once deployed
 the model can be used by the <<inference-processor,{infer-cap} processor>>
 in an ingest pipeline or directly in the <<infer-trained-model>> API.
 
-A model can be deployed multiple times by using deployment IDs. A deployment ID 
-must be unique and should not match any other deployment ID or model ID, unless 
-it is the same as the ID of the model being deployed. If `deployment_id` is not 
+A model can be deployed multiple times by using deployment IDs. A deployment ID
+must be unique and should not match any other deployment ID or model ID, unless
+it is the same as the ID of the model being deployed. If `deployment_id` is not
 set, it defaults to the `model_id`.
 
 Scaling inference performance can be achieved by setting the parameters
@@ -61,7 +61,7 @@ include::{es-ref-dir}/ml/ml-shared.asciidoc[tag=model-id]
 `cache_size`::
 (Optional, <<byte-units,byte value>>)
 The inference cache size (in memory outside the JVM heap) per node for the
-model. The default value is the size of the model as reported by the
+model. In serverless, the cache is disabled by default. Otherwise, the default value is the size of the model as reported by the
 `model_size_bytes` field in the <<get-trained-models-stats>>. To disable the
 cache, `0b` can be provided.
 
@@ -165,8 +165,8 @@ The API returns the following results:
 [[start-trained-model-deployment-deployment-id-example]]
 === Using deployment IDs
 
-The following example starts a new deployment for the `my_model` trained model 
-with the ID `my_model_for_ingest`. The deployment ID an be used in {infer} API 
+The following example starts a new deployment for the `my_model` trained model
+with the ID `my_model_for_ingest`. The deployment ID an be used in {infer} API
 calls or in {infer} processors.
 
 [source,console]
@@ -181,4 +181,4 @@ The `my_model` trained model can be deployed again with a different ID:
 --------------------------------------------------
 POST _ml/trained_models/my_model/deployment/_start?deployment_id=my_model_for_search
 --------------------------------------------------
-// TEST[skip:TBD]
+// TEST[skip:TBD]

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MachineLearning.java

@@ -1475,7 +1475,7 @@ public List<RestHandler> getRestHandlers(
                 restHandlers.add(new RestCatDataFrameAnalyticsAction());
             }
             if (machineLearningExtension.get().isNlpEnabled()) {
-                restHandlers.add(new RestStartTrainedModelDeploymentAction());
+                restHandlers.add(new RestStartTrainedModelDeploymentAction(machineLearningExtension.get().disableInferenceProcessCache()));
                 restHandlers.add(new RestStopTrainedModelDeploymentAction());
                 restHandlers.add(new RestInferTrainedModelDeploymentAction());
                 restHandlers.add(new RestUpdateTrainedModelDeploymentAction());

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/MachineLearningExtension.java

@@ -29,6 +29,10 @@ default boolean isLearningToRankEnabled() {
         return false;
     }
 
+    default boolean disableInferenceProcessCache() {
+        return false;
+    }
+
     String[] getAnalyticsDestIndexAllowedSettings();
 
     AbstractNodeAvailabilityZoneMapper getNodeAvailabilityZoneMapper(Settings settings, ClusterSettings clusterSettings);

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/rest/inference/RestStartTrainedModelDeploymentAction.java

@@ -36,6 +36,18 @@
 @ServerlessScope(Scope.PUBLIC)
 public class RestStartTrainedModelDeploymentAction extends BaseRestHandler {
 
+    public RestStartTrainedModelDeploymentAction(boolean disableInferenceProcessCache) {
+        super();
+        if (disableInferenceProcessCache) {
+            this.defaultCacheSize = ByteSizeValue.ZERO;
+        } else {
+            // Don't set the default cache size yet
+            defaultCacheSize = null;
+        }
+    }
+
+    private final ByteSizeValue defaultCacheSize;
+
     @Override
     public String getName() {
         return "xpack_ml_start_trained_models_deployment_action";
@@ -98,6 +110,8 @@ protected RestChannelConsumer prepareRequest(RestRequest restRequest, NodeClient
                 request.setCacheSize(
                     ByteSizeValue.parseBytesSizeValue(restRequest.param(CACHE_SIZE.getPreferredName()), CACHE_SIZE.getPreferredName())
                 );
+            } else if (defaultCacheSize != null) {
+                request.setCacheSize(defaultCacheSize);
             }
             request.setQueueCapacity(restRequest.paramAsInt(QUEUE_CAPACITY.getPreferredName(), request.getQueueCapacity()));
             request.setPriority(

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/rest/inference/RestStartTrainedModelDeploymentActionTests.java

@@ -0,0 +1,70 @@
+/*
+ * 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.ml.rest.inference;
+
+import org.apache.lucene.util.SetOnce;
+import org.elasticsearch.common.unit.ByteSizeValue;
+import org.elasticsearch.rest.RestRequest;
+import org.elasticsearch.test.rest.FakeRestRequest;
+import org.elasticsearch.test.rest.RestActionTestCase;
+import org.elasticsearch.xpack.core.ml.action.CreateTrainedModelAssignmentAction;
+import org.elasticsearch.xpack.core.ml.action.StartTrainedModelDeploymentAction;
+import org.elasticsearch.xpack.core.ml.inference.assignment.TrainedModelAssignmentTests;
+
+import static org.hamcrest.CoreMatchers.is;
+import static org.hamcrest.Matchers.equalTo;
+import static org.hamcrest.Matchers.instanceOf;
+
+public class RestStartTrainedModelDeploymentActionTests extends RestActionTestCase {
+
+    public void testCacheDisabled() {
+        final boolean disableInferenceProcessCache = true;
+        controller().registerHandler(new RestStartTrainedModelDeploymentAction(disableInferenceProcessCache));
+        SetOnce<Boolean> executeCalled = new SetOnce<>();
+        verifyingClient.setExecuteVerifier(((actionType, actionRequest) -> {
+            assertThat(actionRequest, instanceOf(StartTrainedModelDeploymentAction.Request.class));
+
+            var request = (StartTrainedModelDeploymentAction.Request) actionRequest;
+            assertThat(request.getCacheSize(), is(ByteSizeValue.ZERO));
+
+            executeCalled.set(true);
+            return createResponse();
+        }));
+
+        RestRequest inferenceRequest = new FakeRestRequest.Builder(xContentRegistry()).withMethod(RestRequest.Method.POST)
+            .withPath("_ml/trained_models/test_id/deployment/_start")
+            .build();
+        dispatchRequest(inferenceRequest);
+        assertThat(executeCalled.get(), equalTo(true));
+    }
+
+    public void testCacheEnabled() {
+        final boolean disableInferenceProcessCache = false;
+        controller().registerHandler(new RestStartTrainedModelDeploymentAction(disableInferenceProcessCache));
+        SetOnce<Boolean> executeCalled = new SetOnce<>();
+        verifyingClient.setExecuteVerifier(((actionType, actionRequest) -> {
+            assertThat(actionRequest, instanceOf(StartTrainedModelDeploymentAction.Request.class));
+
+            var request = (StartTrainedModelDeploymentAction.Request) actionRequest;
+            assertNull(request.getCacheSize());
+
+            executeCalled.set(true);
+            return createResponse();
+        }));
+
+        RestRequest inferenceRequest = new FakeRestRequest.Builder(xContentRegistry()).withMethod(RestRequest.Method.POST)
+            .withPath("_ml/trained_models/test_id/deployment/_start")
+            .build();
+        dispatchRequest(inferenceRequest);
+        assertThat(executeCalled.get(), equalTo(true));
+    }
+
+    private static CreateTrainedModelAssignmentAction.Response createResponse() {
+        return new CreateTrainedModelAssignmentAction.Response(TrainedModelAssignmentTests.randomInstance());
+    }
+}