Revert "[ML] Remove scale to zero feature flag (#114323)"

test/test-clusters/src/main/java/org/elasticsearch/test/cluster/FeatureFlag.java

@@ -20,6 +20,7 @@ public enum FeatureFlag {
     FAILURE_STORE_ENABLED("es.failure_store_feature_flag_enabled=true", Version.fromString("8.12.0"), null),
     SUB_OBJECTS_AUTO_ENABLED("es.sub_objects_auto_feature_flag_enabled=true", Version.fromString("8.16.0"), null),
     CHUNKING_SETTINGS_ENABLED("es.inference_chunking_settings_feature_flag_enabled=true", Version.fromString("8.16.0"), null),
+    INFERENCE_SCALE_TO_ZERO("es.inference_scale_to_zero_feature_flag_enabled=true", Version.fromString("8.16.0"), null),
     INFERENCE_DEFAULT_ELSER("es.inference_default_elser_feature_flag_enabled=true", Version.fromString("8.16.0"), null);
 
     public final String systemProperty;

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/adaptiveallocations/AdaptiveAllocationsScaler.java

@@ -170,7 +170,8 @@ Integer scale() {
         if (maxNumberOfAllocations != null) {
             numberOfAllocations = Math.min(numberOfAllocations, maxNumberOfAllocations);
         }
-        if ((minNumberOfAllocations == null || minNumberOfAllocations == 0)
+        if (ScaleToZeroFeatureFlag.isEnabled()
+            && (minNumberOfAllocations == null || minNumberOfAllocations == 0)
             && timeWithoutRequestsSeconds > SCALE_TO_ZERO_AFTER_NO_REQUESTS_TIME_SECONDS) {
             logger.debug("[{}] adaptive allocations scaler: scaling down to zero, because of no requests.", deploymentId);
             numberOfAllocations = 0;

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/adaptiveallocations/AdaptiveAllocationsScalerService.java

@@ -410,7 +410,8 @@ private void processDeploymentStats(GetDeploymentStatsAction.Response statsRespo
     }
 
     public boolean maybeStartAllocation(TrainedModelAssignment assignment) {
-        if (assignment.getAdaptiveAllocationsSettings() != null
+        if (ScaleToZeroFeatureFlag.isEnabled()
+            && assignment.getAdaptiveAllocationsSettings() != null
             && assignment.getAdaptiveAllocationsSettings().getEnabled() == Boolean.TRUE) {
             lastScaleUpTimesMillis.put(assignment.getDeploymentId(), System.currentTimeMillis());
             updateNumberOfAllocations(assignment.getDeploymentId(), 1);

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/inference/adaptiveallocations/ScaleToZeroFeatureFlag.java

@@ -0,0 +1,20 @@
+/*
+ * 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.inference.adaptiveallocations;
+
+import org.elasticsearch.common.util.FeatureFlag;
+
+public class ScaleToZeroFeatureFlag {
+    private ScaleToZeroFeatureFlag() {}
+
+    private static final FeatureFlag FEATURE_FLAG = new FeatureFlag("inference_scale_to_zero");
+
+    public static boolean isEnabled() {
+        return FEATURE_FLAG.isEnabled();
+    }
+}

x-pack/plugin/ml/src/test/java/org/elasticsearch/xpack/ml/inference/adaptiveallocations/AdaptiveAllocationsScalerTests.java

@@ -148,6 +148,8 @@ public void testAutoscaling_maxAllocationsSafeguard() {
     }
 
     public void testAutoscaling_scaleDownToZeroAllocations() {
+        assumeTrue("Should only run if adaptive allocations feature flag is enabled", ScaleToZeroFeatureFlag.isEnabled());
+
         AdaptiveAllocationsScaler adaptiveAllocationsScaler = new AdaptiveAllocationsScaler("test-deployment", 1);
         // 1 hour with 1 request per 1 seconds, so don't scale.
         for (int i = 0; i < 3600; i++) {
@@ -178,6 +180,8 @@ public void testAutoscaling_scaleDownToZeroAllocations() {
     }
 
     public void testAutoscaling_dontScaleDownToZeroAllocationsWhenMinAllocationsIsSet() {
+        assumeTrue("Should only run if adaptive allocations feature flag is enabled", ScaleToZeroFeatureFlag.isEnabled());
+
         AdaptiveAllocationsScaler adaptiveAllocationsScaler = new AdaptiveAllocationsScaler("test-deployment", 1);
         adaptiveAllocationsScaler.setMinMaxNumberOfAllocations(1, null);