Avoid overshooting watermarks during relocation (elastic#46079)

Today the `DiskThresholdDecider` attempts to account for already-relocating
shards when deciding how to allocate or relocate a shard. Its goal is to stop
relocating shards onto a node before that node exceeds the low watermark, and
to stop relocating shards away from a node as soon as the node drops below the
high watermark.

The decider handles multiple data paths by only accounting for relocating
shards that affect the appropriate data path. However, this mechanism does not
correctly account for _new_ relocating shards, which are unwittingly ignored.
This means that we may evict far too many shards from a node above the high
watermark, and may relocate far too many shards onto a node causing it to blow
right past the low watermark and potentially other watermarks too.

There are in fact two distinct issues that this PR fixes. New incoming shards
have an unknown data path until the `ClusterInfoService` refreshes its
statistics. New outgoing shards have a known data path, but we fail to account
for the change of the corresponding `ShardRouting` from `STARTED` to
`RELOCATING`, meaning that we fail to find the correct data path and treat the
path as unknown here too.

This PR also reworks the `MockDiskUsagesIT` test to avoid using fake data paths
for all shards. With the changes here, the data paths are handled in tests as
they are in production, except that their sizes are fake.

Fixes elastic#45177

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

@@ -47,9 +47,8 @@
 import org.elasticsearch.threadpool.ThreadPool;
 import org.elasticsearch.transport.ReceiveTimeoutTransportException;
 
-import java.util.ArrayList;
-import java.util.Collections;
 import java.util.List;
+import java.util.concurrent.CopyOnWriteArrayList;
 import java.util.concurrent.CountDownLatch;
 import java.util.concurrent.TimeUnit;
 import java.util.function.Consumer;
@@ -88,7 +87,7 @@ public class InternalClusterInfoService implements ClusterInfoService, LocalNode
     private final ClusterService clusterService;
     private final ThreadPool threadPool;
     private final NodeClient client;
-    private final List<Consumer<ClusterInfo>> listeners = Collections.synchronizedList(new ArrayList<>(1));
+    private final List<Consumer<ClusterInfo>> listeners = new CopyOnWriteArrayList<>();
 
     public InternalClusterInfoService(Settings settings, ClusterService clusterService, ThreadPool threadPool, NodeClient client) {
         this.leastAvailableSpaceUsages = ImmutableOpenMap.of();
@@ -275,6 +274,11 @@ private void maybeRefresh() {
         }
     }
 
+    // allow tests to adjust the node stats on receipt
+    List<NodeStats> adjustNodesStats(List<NodeStats> nodeStats) {
+        return nodeStats;
+    }
+
     /**
      * Refreshes the ClusterInfo in a blocking fashion
      */
@@ -284,12 +288,13 @@ public final ClusterInfo refresh() {
         }
         final CountDownLatch nodeLatch = updateNodeStats(new ActionListener<NodesStatsResponse>() {
             @Override
-            public void onResponse(NodesStatsResponse nodeStatses) {
-                ImmutableOpenMap.Builder<String, DiskUsage> newLeastAvaiableUsages = ImmutableOpenMap.builder();
-                ImmutableOpenMap.Builder<String, DiskUsage> newMostAvaiableUsages = ImmutableOpenMap.builder();
-                fillDiskUsagePerNode(logger, nodeStatses.getNodes(), newLeastAvaiableUsages, newMostAvaiableUsages);
-                leastAvailableSpaceUsages = newLeastAvaiableUsages.build();
-                mostAvailableSpaceUsages = newMostAvaiableUsages.build();
+            public void onResponse(NodesStatsResponse nodesStatsResponse) {
+                ImmutableOpenMap.Builder<String, DiskUsage> leastAvailableUsagesBuilder = ImmutableOpenMap.builder();
+                ImmutableOpenMap.Builder<String, DiskUsage> mostAvailableUsagesBuilder = ImmutableOpenMap.builder();
+                fillDiskUsagePerNode(logger, adjustNodesStats(nodesStatsResponse.getNodes()),
+                    leastAvailableUsagesBuilder, mostAvailableUsagesBuilder);
+                leastAvailableSpaceUsages = leastAvailableUsagesBuilder.build();
+                mostAvailableSpaceUsages = mostAvailableUsagesBuilder.build();
             }
 
             @Override
@@ -402,7 +407,7 @@ static void fillDiskUsagePerNode(Logger logger, List<NodeStats> nodeStatsArray,
                     if (leastAvailablePath == null) {
                         assert mostAvailablePath == null;
                         mostAvailablePath = leastAvailablePath = info;
-                    } else if (leastAvailablePath.getAvailable().getBytes() > info.getAvailable().getBytes()){
+                    } else if (leastAvailablePath.getAvailable().getBytes() > info.getAvailable().getBytes()) {
                         leastAvailablePath = info;
                     } else if (mostAvailablePath.getAvailable().getBytes() < info.getAvailable().getBytes()) {
                         mostAvailablePath = info;

server/src/main/java/org/elasticsearch/cluster/routing/allocation/DiskThresholdMonitor.java

@@ -221,6 +221,8 @@ public void onNewInfo(ClusterInfo info) {
             .collect(Collectors.toSet());
 
         if (indicesToAutoRelease.isEmpty() == false) {
+            logger.info("releasing read-only block on indices " + indicesToAutoRelease
+                + " since they are now allocated to nodes with sufficient disk space");
             updateIndicesReadOnly(indicesToAutoRelease, listener, false);
         } else {
             listener.onResponse(null);

server/src/main/java/org/elasticsearch/cluster/routing/allocation/decider/DiskThresholdDecider.java

@@ -90,16 +90,36 @@ static long sizeOfRelocatingShards(RoutingNode node, RoutingAllocation allocatio
                                        boolean subtractShardsMovingAway, String dataPath) {
         ClusterInfo clusterInfo = allocation.clusterInfo();
         long totalSize = 0;
-        for (ShardRouting routing : node.shardsWithState(ShardRoutingState.RELOCATING, ShardRoutingState.INITIALIZING)) {
-            String actualPath = clusterInfo.getDataPath(routing);
-            if (dataPath.equals(actualPath)) {
-                if (routing.initializing() && routing.relocatingNodeId() != null) {
-                    totalSize += getExpectedShardSize(routing, allocation, 0);
-                } else if (subtractShardsMovingAway && routing.relocating()) {
+
+        for (ShardRouting routing : node.shardsWithState(ShardRoutingState.INITIALIZING)) {
+            if (routing.relocatingNodeId() == null) {
+                // in practice the only initializing-but-not-relocating shards with a nonzero expected shard size will be ones created
+                // by a resize (shrink/split/clone) operation which we expect to happen using hard links, so they shouldn't be taking
+                // any additional space and can be ignored here
+                continue;
+            }
+
+            final String actualPath = clusterInfo.getDataPath(routing);
+            // if we don't yet know the actual path of the incoming shard then conservatively assume it's going to the path with the least
+            // free space
+            if (actualPath == null || actualPath.equals(dataPath)) {
+                totalSize += getExpectedShardSize(routing, allocation, 0);
+            }
+        }
+
+        if (subtractShardsMovingAway) {
+            for (ShardRouting routing : node.shardsWithState(ShardRoutingState.RELOCATING)) {
+                String actualPath = clusterInfo.getDataPath(routing);
+                if (actualPath == null) {
+                    // we might know the path of this shard from before when it was relocating
+                    actualPath = clusterInfo.getDataPath(routing.cancelRelocation());
+                }
+                if (dataPath.equals(actualPath)) {
                     totalSize -= getExpectedShardSize(routing, allocation, 0);
                 }
             }
         }
+
         return totalSize;
     }
 

server/src/test/java/org/elasticsearch/cluster/routing/allocation/decider/DiskThresholdDeciderTests.java

@@ -26,7 +26,6 @@
 import org.elasticsearch.cluster.ClusterState;
 import org.elasticsearch.cluster.DiskUsage;
 import org.elasticsearch.cluster.ESAllocationTestCase;
-import org.elasticsearch.cluster.MockInternalClusterInfoService.DevNullClusterInfo;
 import org.elasticsearch.cluster.metadata.IndexMetaData;
 import org.elasticsearch.cluster.metadata.MetaData;
 import org.elasticsearch.cluster.node.DiscoveryNode;
@@ -1002,4 +1001,20 @@ public void logShardStates(ClusterState state) {
                 rn.shardsWithState(RELOCATING),
                 rn.shardsWithState(STARTED));
     }
+
+    /**
+     * ClusterInfo that always reports /dev/null for the shards' data paths.
+     */
+    static class DevNullClusterInfo extends ClusterInfo {
+        DevNullClusterInfo(ImmutableOpenMap<String, DiskUsage> leastAvailableSpaceUsage,
+                           ImmutableOpenMap<String, DiskUsage> mostAvailableSpaceUsage,
+                           ImmutableOpenMap<String, Long> shardSizes) {
+            super(leastAvailableSpaceUsage, mostAvailableSpaceUsage, shardSizes, null);
+        }
+
+        @Override
+        public String getDataPath(ShardRouting shardRouting) {
+            return "/dev/null";
+        }
+    }
 }

server/src/test/java/org/elasticsearch/cluster/routing/allocation/decider/DiskThresholdDeciderUnitTests.java

@@ -25,7 +25,7 @@
 import org.elasticsearch.cluster.ClusterState;
 import org.elasticsearch.cluster.DiskUsage;
 import org.elasticsearch.cluster.ESAllocationTestCase;
-import org.elasticsearch.cluster.MockInternalClusterInfoService.DevNullClusterInfo;
+import org.elasticsearch.cluster.routing.allocation.decider.DiskThresholdDeciderTests.DevNullClusterInfo;
 import org.elasticsearch.cluster.metadata.IndexMetaData;
 import org.elasticsearch.cluster.metadata.MetaData;
 import org.elasticsearch.cluster.node.DiscoveryNode;