[ML] Default endpoint allocations are now configurable

Min and max allocations for the default endpoints' adaptive allocation
settings are now configurable via a setting.

This is intended to help new clusters running on laptops and in
serverless.

This does not automatically increase or reduce the default endpoints to
those values - we still need a cluster reboot to set the values.

Relate #124653

Author: prwhelan

x-pack/plugin/core/src/main/java/org/elasticsearch/xpack/core/ml/inference/assignment/AdaptiveAllocationsSettings.java

@@ -11,6 +11,7 @@
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
 import org.elasticsearch.common.io.stream.Writeable;
+import org.elasticsearch.common.settings.Setting;
 import org.elasticsearch.xcontent.ObjectParser;
 import org.elasticsearch.xcontent.ParseField;
 import org.elasticsearch.xcontent.ToXContentObject;
@@ -22,6 +23,24 @@
 
 public class AdaptiveAllocationsSettings implements ToXContentObject, Writeable {
 
+    public static final Setting<Integer> DEFAULT_MIN_ALLOCATIONS = Setting.intSetting(
+        "xpack.ml.models.default_allocations.min",
+        0,
+        0,
+        32,
+        Setting.Property.Dynamic,
+        Setting.Property.NodeScope
+    );
+
+    public static final Setting<Integer> DEFAULT_MAX_ALLOCATIONS = Setting.intSetting(
+        "xpack.ml.models.default_allocations.max",
+        32,
+        1,
+        32,
+        Setting.Property.Dynamic,
+        Setting.Property.NodeScope
+    );
+
     public static final AdaptiveAllocationsSettings RESET_PLACEHOLDER = new AdaptiveAllocationsSettings(false, -1, -1);
 
     public static final ParseField ENABLED = new ParseField("enabled");

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticRerankerServiceSettings.java

@@ -9,19 +9,27 @@
 
 import org.elasticsearch.common.ValidationException;
 import org.elasticsearch.common.io.stream.StreamInput;
+import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings;
 
 import java.io.IOException;
 import java.util.Map;
 
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MAX_ALLOCATIONS;
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MIN_ALLOCATIONS;
 import static org.elasticsearch.xpack.inference.services.elasticsearch.ElasticsearchInternalService.RERANKER_ID;
 
 public class ElasticRerankerServiceSettings extends ElasticsearchInternalServiceSettings {
 
     public static final String NAME = "elastic_reranker_service_settings";
 
-    public static ElasticRerankerServiceSettings defaultEndpointSettings() {
-        return new ElasticRerankerServiceSettings(null, 1, RERANKER_ID, new AdaptiveAllocationsSettings(Boolean.TRUE, 0, 32));
+    public static ElasticRerankerServiceSettings defaultEndpointSettings(Settings settings) {
+        return new ElasticRerankerServiceSettings(
+            null,
+            1,
+            RERANKER_ID,
+            new AdaptiveAllocationsSettings(Boolean.TRUE, DEFAULT_MIN_ALLOCATIONS.get(settings), DEFAULT_MAX_ALLOCATIONS.get(settings))
+        );
     }
 
     public ElasticRerankerServiceSettings(ElasticsearchInternalServiceSettings other) {

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElasticsearchInternalService.java

@@ -923,22 +923,22 @@ private List<Model> defaultConfigs(boolean useLinuxOptimizedModel) {
             DEFAULT_ELSER_ID,
             TaskType.SPARSE_EMBEDDING,
             NAME,
-            ElserInternalServiceSettings.defaultEndpointSettings(useLinuxOptimizedModel),
+            ElserInternalServiceSettings.defaultEndpointSettings(useLinuxOptimizedModel, settings),
             ElserMlNodeTaskSettings.DEFAULT,
             ChunkingSettingsBuilder.DEFAULT_SETTINGS
         );
         var defaultE5 = new MultilingualE5SmallModel(
             DEFAULT_E5_ID,
             TaskType.TEXT_EMBEDDING,
             NAME,
-            MultilingualE5SmallInternalServiceSettings.defaultEndpointSettings(useLinuxOptimizedModel),
+            MultilingualE5SmallInternalServiceSettings.defaultEndpointSettings(useLinuxOptimizedModel, settings),
             ChunkingSettingsBuilder.DEFAULT_SETTINGS
         );
         var defaultRerank = new ElasticRerankerModel(
             DEFAULT_RERANK_ID,
             TaskType.RERANK,
             NAME,
-            ElasticRerankerServiceSettings.defaultEndpointSettings(),
+            ElasticRerankerServiceSettings.defaultEndpointSettings(settings),
             RerankTaskSettings.DEFAULT_SETTINGS
         );
         return List.of(defaultElser, defaultE5, defaultRerank);

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/ElserInternalServiceSettings.java

@@ -10,11 +10,14 @@
 import org.elasticsearch.TransportVersion;
 import org.elasticsearch.TransportVersions;
 import org.elasticsearch.common.io.stream.StreamInput;
+import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.inference.MinimalServiceSettings;
 import org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings;
 
 import java.io.IOException;
 
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MAX_ALLOCATIONS;
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MIN_ALLOCATIONS;
 import static org.elasticsearch.xpack.inference.services.elasticsearch.ElserModels.ELSER_V2_MODEL;
 import static org.elasticsearch.xpack.inference.services.elasticsearch.ElserModels.ELSER_V2_MODEL_LINUX_X86;
 
@@ -26,12 +29,12 @@ public static MinimalServiceSettings minimalServiceSettings() {
         return MinimalServiceSettings.sparseEmbedding(ElasticsearchInternalService.NAME);
     }
 
-    public static ElserInternalServiceSettings defaultEndpointSettings(boolean useLinuxOptimizedModel) {
+    public static ElserInternalServiceSettings defaultEndpointSettings(boolean useLinuxOptimizedModel, Settings settings) {
         return new ElserInternalServiceSettings(
             null,
             1,
             useLinuxOptimizedModel ? ELSER_V2_MODEL_LINUX_X86 : ELSER_V2_MODEL,
-            new AdaptiveAllocationsSettings(Boolean.TRUE, 0, 32)
+            new AdaptiveAllocationsSettings(Boolean.TRUE, DEFAULT_MIN_ALLOCATIONS.get(settings), DEFAULT_MAX_ALLOCATIONS.get(settings))
         );
     }
 

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elasticsearch/MultilingualE5SmallInternalServiceSettings.java

@@ -9,6 +9,7 @@
 
 import org.elasticsearch.common.ValidationException;
 import org.elasticsearch.common.io.stream.StreamInput;
+import org.elasticsearch.common.settings.Settings;
 import org.elasticsearch.index.mapper.vectors.DenseVectorFieldMapper;
 import org.elasticsearch.inference.MinimalServiceSettings;
 import org.elasticsearch.inference.SimilarityMeasure;
@@ -18,6 +19,8 @@
 import java.util.Arrays;
 import java.util.Map;
 
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MAX_ALLOCATIONS;
+import static org.elasticsearch.xpack.core.ml.inference.assignment.AdaptiveAllocationsSettings.DEFAULT_MIN_ALLOCATIONS;
 import static org.elasticsearch.xpack.inference.services.elasticsearch.ElasticsearchInternalService.MULTILINGUAL_E5_SMALL_MODEL_ID;
 import static org.elasticsearch.xpack.inference.services.elasticsearch.ElasticsearchInternalService.MULTILINGUAL_E5_SMALL_MODEL_ID_LINUX_X86;
 
@@ -37,12 +40,12 @@ public static MinimalServiceSettings minimalServiceSettings() {
         );
     }
 
-    public static MultilingualE5SmallInternalServiceSettings defaultEndpointSettings(boolean useLinuxOptimizedModel) {
+    public static MultilingualE5SmallInternalServiceSettings defaultEndpointSettings(boolean useLinuxOptimizedModel, Settings settings) {
         return new MultilingualE5SmallInternalServiceSettings(
             null,
             1,
             useLinuxOptimizedModel ? MULTILINGUAL_E5_SMALL_MODEL_ID_LINUX_X86 : MULTILINGUAL_E5_SMALL_MODEL_ID,
-            new AdaptiveAllocationsSettings(Boolean.TRUE, 0, 32)
+            new AdaptiveAllocationsSettings(Boolean.TRUE, DEFAULT_MIN_ALLOCATIONS.get(settings), DEFAULT_MAX_ALLOCATIONS.get(settings))
         );
     }