Reduce ML_ONLY heap size, so that direct memory is accounted for.

Author: jan-elastic

distribution/tools/server-cli/src/main/java/org/elasticsearch/server/cli/MachineDependentHeap.java

@@ -98,10 +98,19 @@ protected int getHeapSizeMb(Settings nodeSettings, MachineNodeRole role, long av
              * could result in ML processes crashing with OOM errors or repeated autoscaling up and down.
              */
             case ML_ONLY -> {
+                /*
+                 * An ML node used to have 40% of the total memory for the Java heap, and the remainder
+                 * for ML and overhead (the operating system). This did not account for the Java direct
+                 * memory, which equals half of the heap size (see JvmErgonomics).
+                 * Right now, a factor of 2/3 is applied to the heap size here, leaving the ML memory
+                 * formula the same. That means the formula now also correctly accounts for direct memory,
+                 * since the heap (2/3 * 40% of the memory) plus the direct memory (1/3 * 40% of the memory)
+                 * equals the original 40% of the total memory.
+                 */
                 if (availableMemory <= (GB * 16)) {
-                    yield mb((long) (availableMemory * .4), 4);
+                    yield mb((long) (availableMemory * .4 * 2/3), 4);
                 } else {
-                    yield mb((long) min((GB * 16) * .4 + (availableMemory - GB * 16) * .1, MAX_HEAP_SIZE), 4);
+                    yield mb((long) min(((GB * 16) * .4 + (availableMemory - GB * 16) * .1) * 2/3, MAX_HEAP_SIZE), 4);
                 }
             }
             /*

distribution/tools/server-cli/src/test/java/org/elasticsearch/server/cli/MachineDependentHeapTests.java

@@ -56,13 +56,13 @@ public void testMasterOnlyOptions() throws Exception {
     }
 
     public void testMlOnlyOptions() throws Exception {
-        assertHeapOptions(1, containsInAnyOrder("-Xmx408m", "-Xms408m"), "ml");
-        assertHeapOptions(4, containsInAnyOrder("-Xmx1636m", "-Xms1636m"), "ml");
-        assertHeapOptions(32, containsInAnyOrder("-Xmx8192m", "-Xms8192m"), "ml");
-        assertHeapOptions(64, containsInAnyOrder("-Xmx11468m", "-Xms11468m"), "ml");
+        assertHeapOptions(1, containsInAnyOrder("-Xmx272m", "-Xms272m"), "ml");
+        assertHeapOptions(4, containsInAnyOrder("-Xmx1092m", "-Xms1092m"), "ml");
+        assertHeapOptions(32, containsInAnyOrder("-Xmx5460m", "-Xms5460m"), "ml");
+        assertHeapOptions(64, containsInAnyOrder("-Xmx7644m", "-Xms7644m"), "ml");
         // We'd never see a node this big in Cloud, but this assertion proves that the 31GB absolute maximum
         // eventually kicks in (because 0.4 * 16 + 0.1 * (263 - 16) > 31)
-        assertHeapOptions(263, containsInAnyOrder("-Xmx31744m", "-Xms31744m"), "ml");
+        assertHeapOptions(263, containsInAnyOrder("-Xmx21228m", "-Xms21228m"), "ml");
     }
 
     public void testDataNodeOptions() throws Exception {