DesiredBalanceReconciler always returns AllocationStats (#122458)

Ensures that the DesiredBalanceReconciler always returns a non-empty AllocationStats object, eliminating edge cases where the stats available to DesiredBalanceMetrics may not be updated due to some kind of throttling or the balancer being disabled via cluster settings. Adds documentation around AllocationDecider#canRebalance(RoutingAllocation) Closes ES-10581

Author: DiannaHohensee

docs/changelog/122458.yaml

@@ -0,0 +1,5 @@
+pr: 122458
+summary: '`DesiredBalanceReconciler` always returns `AllocationStats`'
+area: Allocation
+type: bug
+issues: []

server/src/main/java/org/elasticsearch/cluster/routing/RoutingNode.java

@@ -112,6 +112,9 @@ public String nodeId() {
  return this.nodeId;
  }
 
+ /**
+ * Number of shards assigned to this node. Includes relocating shards. Use {@link #numberOfOwningShards()} to exclude relocating shards.
+ */
  public int size() {
  return shards.size();
  }

server/src/main/java/org/elasticsearch/cluster/routing/allocation/allocator/DesiredBalanceReconciler.java

@@ -515,11 +515,6 @@ private void moveShards() {
  }
 
  private DesiredBalanceMetrics.AllocationStats balance() {
- // Check if rebalancing is disabled.
- if (allocation.deciders().canRebalance(allocation).type() != Decision.Type.YES) {
- return DesiredBalanceMetrics.EMPTY_ALLOCATION_STATS;
- }
-
  int unassignedShards = routingNodes.unassigned().size() + routingNodes.unassigned().ignored().size();
  int totalAllocations = 0;
  int undesiredAllocationsExcludingShuttingDownNodes = 0;
@@ -549,9 +544,15 @@ private DesiredBalanceMetrics.AllocationStats balance() {
  }
 
  if (allocation.metadata().nodeShutdowns().contains(shardRouting.currentNodeId()) == false) {
+ // shard is not on a shutting down node, nor is it on a desired node per the previous check.
  undesiredAllocationsExcludingShuttingDownNodes++;
  }
 
+ if (allocation.deciders().canRebalance(allocation).type() != Decision.Type.YES) {
+ // Rebalancing is disabled, we're just here to collect the AllocationStats to return.
+ continue;
+ }
+
  if (allocation.deciders().canRebalance(shardRouting, allocation).type() != Decision.Type.YES) {
  // rebalancing disabled for this shard
  continue;

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

@@ -75,9 +75,9 @@ public Decision shouldAutoExpandToNode(IndexMetadata indexMetadata, DiscoveryNod
  }
 
  /**
- * Returns a {@link Decision} whether the cluster can execute
- * re-balanced operations at all.
- * {@link Decision#ALWAYS}.
+ * Returns a {@link Decision} on whether the cluster is allowed to rebalance shards to improve relative node shard weights and
+ * performance.
+ * @return {@link Decision#ALWAYS} is returned by default if not overridden.
  */
  public Decision canRebalance(RoutingAllocation allocation) {
  return Decision.ALWAYS;

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

@@ -82,6 +82,10 @@ public Decision canAllocate(ShardRouting shardRouting, RoutingNode node, Routing
  );
  }
 
+ /**
+ * Returns whether rebalancing (move shards to improve relative node weights and performance) is allowed right now.
+ * Rebalancing can be disabled via cluster settings, or throttled by cluster settings (e.g. max concurrent shard moves).
+ */
  public Decision canRebalance(RoutingAllocation allocation) {
  return withDeciders(
  allocation,

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

@@ -150,6 +150,10 @@ public Decision canRebalance(ShardRouting shardRouting, RoutingAllocation alloca
  + "]"
  );
 
+ /**
+ * Rebalancing may be enabled, disabled, or only allowed after all primaries have started, depending on the cluster setting
+ * {@link #CLUSTER_ROUTING_ALLOCATION_ALLOW_REBALANCE_SETTING}.
+ */
  @SuppressWarnings("fallthrough")
  @Override
  public Decision canRebalance(RoutingAllocation allocation) {

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

@@ -61,6 +61,11 @@ public Decision canRebalance(ShardRouting shardRouting, RoutingAllocation alloca
  return canRebalance(allocation);
  }
 
+ /**
+ * We allow a limited number of concurrent shard relocations, per the cluster setting
+ * {@link #CLUSTER_ROUTING_ALLOCATION_CLUSTER_CONCURRENT_REBALANCE_SETTING}.
+ * Returns a {@link Decision#THROTTLE} decision if the limit is exceeded, otherwise returns {@link Decision#YES}.
+ */
  @Override
  public Decision canRebalance(RoutingAllocation allocation) {
  int relocatingShards = allocation.routingNodes().getRelocatingShardCount();

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

@@ -146,6 +146,11 @@ public Decision canAllocate(ShardRouting shardRouting, RoutingAllocation allocat
  };
  }
 
+ /**
+ * Rebalancing is limited by the {@link Rebalance} value set on the cluster setting {@link #CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING}.
+ * We might allow movement only of primary shards, or replica shards, or none, or all.
+ * This method only concerns itself with whether {@link Rebalance#NONE} is set: rebalancing is allowed for all other setting values.
+ */
  @Override
  public Decision canRebalance(RoutingAllocation allocation) {
  if (allocation.ignoreDisable()) {
@@ -243,7 +248,7 @@ public String toString() {
  }
 
  /**
- * Rebalance values or rather their string representation to be used used with
+ * Rebalance values or rather their string representation to be used with
  * {@link EnableAllocationDecider#CLUSTER_ROUTING_REBALANCE_ENABLE_SETTING} /
  * {@link EnableAllocationDecider#INDEX_ROUTING_REBALANCE_ENABLE_SETTING}
  * via cluster / index settings.