[ML] Fix double-counting of inference memory in the assignment rebalancer (elastic#133919)

The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation.

This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.

Author: valeriy42

docs/changelog/133919.yaml

@@ -0,0 +1,5 @@
+pr: 133919
+summary: Fix double-counting of inference memory in the assignment rebalancer
+area: Machine Learning
+type: bug
+issues: []

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

@@ -307,9 +307,7 @@ private Map<List<String>, List<AssignmentPlan.Node>> createNodesByZoneMap() {
                         nodes.add(
                             new AssignmentPlan.Node(
                                 discoveryNode.getId(),
-                                // We subtract native inference memory as the planner expects available memory for
-                                // native inference including current assignments.
-                                getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference(load),
+                                load.getFreeMemoryExcludingPerNodeOverhead(),
                                 MlProcessors.get(discoveryNode, allocatedProcessorsScale).roundUp()
                             )
                         );
@@ -326,10 +324,6 @@ private Map<List<String>, List<AssignmentPlan.Node>> createNodesByZoneMap() {
         }));
     }
 
-    private static long getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference(NodeLoad load) {
-        return load.getFreeMemoryExcludingPerNodeOverhead() - load.getAssignedNativeInferenceMemory();
-    }
-
     private TrainedModelAssignmentMetadata.Builder buildAssignmentsFromPlan(AssignmentPlan assignmentPlan) {
         TrainedModelAssignmentMetadata.Builder builder = TrainedModelAssignmentMetadata.Builder.empty();
         for (AssignmentPlan.Deployment deployment : assignmentPlan.models()) {