Setting for estimated shard heap allocation decider

modules/data-streams/src/test/java/org/elasticsearch/datastreams/LookAHeadTimeTests.java

@@ -118,7 +118,8 @@ public void testLookAheadTimeSettingHigherThanTimeSeriesPollIntervalSetting() {
         updateIndexSettings(indexSettings);
     }
 
-    private void updateClusterSettings(Settings settings) {
+    @Override
+    protected void updateClusterSettings(Settings settings) {
         clusterAdmin().updateSettings(
             new ClusterUpdateSettingsRequest(TEST_REQUEST_TIMEOUT, TEST_REQUEST_TIMEOUT).persistentSettings(settings)
         ).actionGet();

server/src/internalClusterTest/java/org/elasticsearch/index/shard/IndexShardIT.java

@@ -264,14 +264,33 @@ public void testExpectedShardSizeIsPresent() throws InterruptedException {
     public void testHeapUsageEstimateIsPresent() {
         InternalClusterInfoService clusterInfoService = (InternalClusterInfoService) getInstanceFromNode(ClusterInfoService.class);
         ClusterInfoServiceUtils.refresh(clusterInfoService);
-        ClusterState state = getInstanceFromNode(ClusterService.class).state();
         Map<String, ShardHeapUsage> shardHeapUsages = clusterInfoService.getClusterInfo().getShardHeapUsages();
         assertNotNull(shardHeapUsages);
-        assertEquals(state.nodes().size(), shardHeapUsages.size());
-        for (DiscoveryNode node : state.nodes()) {
-            assertTrue(shardHeapUsages.containsKey(node.getId()));
-            ShardHeapUsage shardHeapUsage = shardHeapUsages.get(node.getId());
-            assertThat(shardHeapUsage.estimatedFreeBytes(), lessThanOrEqualTo(shardHeapUsage.totalBytes()));
+        // Not collecting yet because it is disabled
+        assertTrue(shardHeapUsages.isEmpty());
+
+        // Enable collection for shard heap usages
+        updateClusterSettings(
+            Settings.builder()
+                .put(InternalClusterInfoService.CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED.getKey(), true)
+                .build()
+        );
+        try {
+            ClusterInfoServiceUtils.refresh(clusterInfoService);
+            ClusterState state = getInstanceFromNode(ClusterService.class).state();
+            shardHeapUsages = clusterInfoService.getClusterInfo().getShardHeapUsages();
+            assertEquals(state.nodes().size(), shardHeapUsages.size());
+            for (DiscoveryNode node : state.nodes()) {
+                assertTrue(shardHeapUsages.containsKey(node.getId()));
+                ShardHeapUsage shardHeapUsage = shardHeapUsages.get(node.getId());
+                assertThat(shardHeapUsage.estimatedFreeBytes(), lessThanOrEqualTo(shardHeapUsage.totalBytes()));
+            }
+        } finally {
+            updateClusterSettings(
+                Settings.builder()
+                    .putNull(InternalClusterInfoService.CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED.getKey())
+                    .build()
+            );
         }
     }
 

server/src/main/java/org/elasticsearch/cluster/ClusterInfoSimulator.java

@@ -33,6 +33,7 @@ public class ClusterInfoSimulator {
     private final CopyOnFirstWriteMap<String, Long> shardSizes;
     private final Map<ShardId, Long> shardDataSetSizes;
     private final Map<NodeAndShard, String> dataPath;
+    private final Map<String, ShardHeapUsage> shardHeapUsages;
 
     public ClusterInfoSimulator(RoutingAllocation allocation) {
         this.allocation = allocation;
@@ -41,6 +42,7 @@ public ClusterInfoSimulator(RoutingAllocation allocation) {
         this.shardSizes = new CopyOnFirstWriteMap<>(allocation.clusterInfo().shardSizes);
         this.shardDataSetSizes = Map.copyOf(allocation.clusterInfo().shardDataSetSizes);
         this.dataPath = Map.copyOf(allocation.clusterInfo().dataPath);
+        this.shardHeapUsages = allocation.clusterInfo().getShardHeapUsages();
     }
 
     /**
@@ -154,7 +156,7 @@ public ClusterInfo getClusterInfo() {
             shardDataSetSizes,
             dataPath,
             Map.of(),
-            Map.of()
+            shardHeapUsages
         );
     }
 }

server/src/main/java/org/elasticsearch/cluster/InternalClusterInfoService.java

@@ -83,7 +83,15 @@ public class InternalClusterInfoService implements ClusterInfoService, ClusterSt
         Property.NodeScope
     );
 
+    public static final Setting<Boolean> CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED = Setting.boolSetting(
+        "cluster.routing.allocation.shard_heap.threshold_enabled",
+        false,
+        Property.Dynamic,
+        Property.NodeScope
+    );
+
     private volatile boolean diskThresholdEnabled;
+    private volatile boolean shardHeapThresholdEnabled;
     private volatile TimeValue updateFrequency;
     private volatile TimeValue fetchTimeout;
 
@@ -130,12 +138,20 @@ public InternalClusterInfoService(
             DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING,
             this::setDiskThresholdEnabled
         );
+        clusterSettings.initializeAndWatch(
+            CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED,
+            this::setShardHeapThresholdEnabled
+        );
     }
 
     private void setDiskThresholdEnabled(boolean diskThresholdEnabled) {
         this.diskThresholdEnabled = diskThresholdEnabled;
     }
 
+    private void setShardHeapThresholdEnabled(boolean shardHeapThresholdEnabled) {
+        this.shardHeapThresholdEnabled = shardHeapThresholdEnabled;
+    }
+
     private void setFetchTimeout(TimeValue fetchTimeout) {
         this.fetchTimeout = fetchTimeout;
     }
@@ -185,20 +201,44 @@ void execute() {
             logger.trace("starting async refresh");
 
             try (var ignoredRefs = fetchRefs) {
-                if (diskThresholdEnabled) {
-                    try (var ignored = threadPool.getThreadContext().clearTraceContext()) {
-                        fetchIndicesStats();
-                    }
-                } else {
-                    logger.trace("skipping collecting disk usage info from cluster, notifying listeners with empty cluster info");
-                    indicesStatsSummary = IndicesStatsSummary.EMPTY;
+                maybeFetchIndicesStats(diskThresholdEnabled);
+                maybeFetchNodeStats(diskThresholdEnabled || shardHeapThresholdEnabled);
+                maybeFetchNodesHeapUsage(shardHeapThresholdEnabled);
+            }
+        }
+
+        private void maybeFetchIndicesStats(boolean shouldFetch) {
+            if (shouldFetch) {
+                try (var ignored = threadPool.getThreadContext().clearTraceContext()) {
+                    fetchIndicesStats();
                 }
+            } else {
+                logger.trace("skipping collecting disk usage info from cluster, notifying listeners with empty indices stats");
+                indicesStatsSummary = IndicesStatsSummary.EMPTY;
+            }
+        }
+
+        private void maybeFetchNodeStats(boolean shouldFetch) {
+            if (shouldFetch) {
                 try (var ignored = threadPool.getThreadContext().clearTraceContext()) {
                     fetchNodeStats();
                 }
+            } else {
+                logger.trace("skipping collecting node stats from cluster, notifying listeners with empty node stats");
+                leastAvailableSpaceUsages = Map.of();
+                mostAvailableSpaceUsages = Map.of();
+                maxHeapPerNode = Map.of();
+            }
+        }
+
+        private void maybeFetchNodesHeapUsage(boolean shouldFetch) {
+            if (shouldFetch) {
                 try (var ignored = threadPool.getThreadContext().clearTraceContext()) {
                     fetchNodesHeapUsage();
                 }
+            } else {
+                logger.trace("skipping collecting shard heap usage from cluster, notifying listeners with empty shard heap usage");
+                shardHeapUsagePerNode = Map.of();
             }
         }
 

server/src/main/java/org/elasticsearch/cluster/ShardHeapUsage.java

@@ -45,6 +45,10 @@ public double estimatedFreeBytesAsPercentage() {
     }
 
     public double estimatedUsageAsPercentage() {
-        return 100.0 * estimatedUsageBytes / (double) totalBytes;
+        return 100.0 * estimatedUsageAsRatio();
+    }
+
+    public double estimatedUsageAsRatio() {
+        return estimatedUsageBytes / (double) totalBytes;
     }
 }

server/src/main/java/org/elasticsearch/common/settings/ClusterSettings.java

@@ -292,6 +292,7 @@ public void apply(Settings value, Settings current, Settings previous) {
         DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_FLOOD_STAGE_FROZEN_MAX_HEADROOM_SETTING,
         DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_DISK_THRESHOLD_ENABLED_SETTING,
         DiskThresholdSettings.CLUSTER_ROUTING_ALLOCATION_REROUTE_INTERVAL_SETTING,
+        InternalClusterInfoService.CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED,
         SameShardAllocationDecider.CLUSTER_ROUTING_ALLOCATION_SAME_HOST_SETTING,
         InternalClusterInfoService.INTERNAL_CLUSTER_INFO_UPDATE_INTERVAL_SETTING,
         InternalClusterInfoService.INTERNAL_CLUSTER_INFO_TIMEOUT_SETTING,

server/src/test/java/org/elasticsearch/cluster/InternalClusterInfoServiceSchedulingTests.java

@@ -53,7 +53,9 @@ public void testScheduling() {
         final DiscoveryNodes noMaster = DiscoveryNodes.builder().add(discoveryNode).localNodeId(discoveryNode.getId()).build();
         final DiscoveryNodes localMaster = noMaster.withMasterNodeId(discoveryNode.getId());
 
-        final Settings.Builder settingsBuilder = Settings.builder().put(Node.NODE_NAME_SETTING.getKey(), discoveryNode.getName());
+        final Settings.Builder settingsBuilder = Settings.builder()
+            .put(Node.NODE_NAME_SETTING.getKey(), discoveryNode.getName())
+            .put(InternalClusterInfoService.CLUSTER_ROUTING_ALLOCATION_SHARD_HEAP_THRESHOLD_DECIDER_ENABLED.getKey(), true);
         if (randomBoolean()) {
             settingsBuilder.put(INTERNAL_CLUSTER_INFO_UPDATE_INTERVAL_SETTING.getKey(), randomIntBetween(10000, 60000) + "ms");
         }

test/framework/src/main/java/org/elasticsearch/test/ESSingleNodeTestCase.java

@@ -536,4 +536,8 @@ protected final void deletePipeline(String id) {
             )
         );
     }
+
+    protected void updateClusterSettings(Settings settings) {
+        safeGet(clusterAdmin().prepareUpdateSettings(TEST_REQUEST_TIMEOUT, TEST_REQUEST_TIMEOUT).setPersistentSettings(settings).execute());
+    }
 }