Rework MSE query throttling to take into account estimated number of …

…threads used by a query (#14847)

pinot-broker/src/main/java/org/apache/pinot/broker/requesthandler/MultiStageBrokerRequestHandler.java

@@ -238,10 +238,11 @@ protected BrokerResponse handleRequest(long requestId, String query, SqlNodeAndO
     }
 
     Timer queryTimer = new Timer(queryTimeoutMs);
+    int estimatedNumQueryThreads = dispatchableSubPlan.getEstimatedNumQueryThreads();
     try {
       // It's fine to block in this thread because we use a separate thread pool from the main Jersey server to process
       // these requests.
-      if (!_queryThrottler.tryAcquire(queryTimeoutMs, TimeUnit.MILLISECONDS)) {
+      if (!_queryThrottler.tryAcquire(estimatedNumQueryThreads, queryTimeoutMs, TimeUnit.MILLISECONDS)) {
         LOGGER.warn("Timed out waiting to execute request {}: {}", requestId, query);
         requestContext.setErrorCode(QueryException.EXECUTION_TIMEOUT_ERROR_CODE);
         return new BrokerResponseNative(QueryException.EXECUTION_TIMEOUT_ERROR);
@@ -322,7 +323,7 @@ protected BrokerResponse handleRequest(long requestId, String query, SqlNodeAndO
 
       return brokerResponse;
     } finally {
-      _queryThrottler.release();
+      _queryThrottler.release(estimatedNumQueryThreads);
     }
   }
 

pinot-broker/src/main/java/org/apache/pinot/broker/requesthandler/MultiStageQueryThrottler.java

@@ -37,8 +37,14 @@
 
 /**
  * This class helps limit the number of multi-stage queries being executed concurrently. Note that the cluster
- * configuration is a "per server" value and the broker currently simply assumes that a query will be across all
- * servers. Another assumption here is that queries are evenly distributed across brokers.
+ * configuration is a "per server" value and the broker currently computes the max server query threads as
+ * <em>CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS * numServers / numBrokers</em>. Note that the config value,
+ * number of servers, and number of brokers are all dynamically updated here.
+ * <p>
+ * Another assumption made here is that queries are evenly distributed across brokers.
+ * <p>
+ * This is designed to limit the number of multi-stage queries being concurrently executed across a cluster and is not
+ * intended to prevent individual large queries from being executed.
  */
 public class MultiStageQueryThrottler implements ClusterChangeHandler {
 
@@ -50,10 +56,10 @@ public class MultiStageQueryThrottler implements ClusterChangeHandler {
   private int _numBrokers;
   private int _numServers;
   /**
-   * If _maxConcurrentQueries is <= 0, it means that the cluster is not configured to limit the number of multi-stage
+   * If _maxServerQueryThreads is <= 0, it means that the cluster is not configured to limit the number of multi-stage
    * queries that can be executed concurrently. In this case, we should not block the query.
    */
-  private int _maxConcurrentQueries;
+  private int _maxServerQueryThreads;
   private AdjustableSemaphore _semaphore;
 
   @Override
@@ -63,11 +69,11 @@ public void init(HelixManager helixManager) {
     _helixConfigScope = new HelixConfigScopeBuilder(HelixConfigScope.ConfigScopeProperty.CLUSTER).forCluster(
         _helixManager.getClusterName()).build();
 
-    _maxConcurrentQueries = Integer.parseInt(
+    _maxServerQueryThreads = Integer.parseInt(
         _helixAdmin.getConfig(_helixConfigScope,
-                Collections.singletonList(CommonConstants.Helix.CONFIG_OF_MAX_CONCURRENT_MULTI_STAGE_QUERIES))
-            .getOrDefault(CommonConstants.Helix.CONFIG_OF_MAX_CONCURRENT_MULTI_STAGE_QUERIES,
-                CommonConstants.Helix.DEFAULT_MAX_CONCURRENT_MULTI_STAGE_QUERIES));
+                Collections.singletonList(CommonConstants.Helix.CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS))
+            .getOrDefault(CommonConstants.Helix.CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS,
+                CommonConstants.Helix.DEFAULT_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS));
 
     List<String> clusterInstances = _helixAdmin.getInstancesInCluster(_helixManager.getClusterName());
     _numBrokers = Math.max(1, (int) clusterInstances.stream()
@@ -77,36 +83,49 @@ public void init(HelixManager helixManager) {
         .filter(instance -> instance.startsWith(CommonConstants.Helix.PREFIX_OF_SERVER_INSTANCE))
         .count());
 
-    if (_maxConcurrentQueries > 0) {
-      _semaphore = new AdjustableSemaphore(Math.max(1, _maxConcurrentQueries * _numServers / _numBrokers), true);
+    if (_maxServerQueryThreads > 0) {
+      _semaphore = new AdjustableSemaphore(Math.max(1, _maxServerQueryThreads * _numServers / _numBrokers), true);
     }
   }
 
   /**
    * Returns true if the query can be executed (waiting until it can be executed if necessary), false otherwise.
    * <p>
-   * {@link #release()} should be called after the query is done executing. It is the responsibility of the caller to
-   * ensure that {@link #release()} is called exactly once for each call to this method.
+   * {@link #release(int)} should be called after the query is done executing. It is the responsibility of the caller to
+   * ensure that {@link #release(int)} is called exactly once for each call to this method.
    *
+   * @param numQueryThreads the estimated number of query server threads
    * @param timeout the maximum time to wait
    * @param unit the time unit of the timeout argument
+   *
    * @throws InterruptedException if the current thread is interrupted
+   * @throws RuntimeException if the query can never be dispatched due to the number of estimated query server threads
+   * being greater than the maximum number of server query threads calculated on the basis of
+   * <em>CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS * numServers / numBrokers</em>
    */
-  public boolean tryAcquire(long timeout, TimeUnit unit)
+  public boolean tryAcquire(int numQueryThreads, long timeout, TimeUnit unit)
       throws InterruptedException {
-    if (_maxConcurrentQueries <= 0) {
+    if (_maxServerQueryThreads <= 0) {
       return true;
     }
-    return _semaphore.tryAcquire(timeout, unit);
+
+    if (numQueryThreads > _semaphore.getTotalPermits()) {
+      throw new RuntimeException(
+          "Can't dispatch query because the estimated number of server threads for this query is too large for the "
+              + "configured value of '" + CommonConstants.Helix.CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS
+              + "'. Consider increasing the value of this configuration");
+    }
+
+    return _semaphore.tryAcquire(numQueryThreads, timeout, unit);
   }
 
   /**
    * Should be called after the query is done executing. It is the responsibility of the caller to ensure that this
-   * method is called exactly once for each call to {@link #tryAcquire(long, TimeUnit)}.
+   * method is called exactly once for each call to {@link #tryAcquire(int, long, TimeUnit)}.
    */
-  public void release() {
-    if (_maxConcurrentQueries > 0) {
-      _semaphore.release();
+  public void release(int numQueryThreads) {
+    if (_maxServerQueryThreads > 0) {
+      _semaphore.release(numQueryThreads);
     }
   }
 
@@ -128,34 +147,33 @@ public void processClusterChange(HelixConstants.ChangeType changeType) {
       if (numBrokers != _numBrokers || numServers != _numServers) {
         _numBrokers = numBrokers;
         _numServers = numServers;
-        if (_maxConcurrentQueries > 0) {
-          _semaphore.setPermits(Math.max(1, _maxConcurrentQueries * _numServers / _numBrokers));
+        if (_maxServerQueryThreads > 0) {
+          _semaphore.setPermits(Math.max(1, _maxServerQueryThreads * _numServers / _numBrokers));
         }
       }
     } else {
-      int maxConcurrentQueries = Integer.parseInt(
-          _helixAdmin.getConfig(_helixConfigScope,
-                  Collections.singletonList(CommonConstants.Helix.CONFIG_OF_MAX_CONCURRENT_MULTI_STAGE_QUERIES))
-              .getOrDefault(CommonConstants.Helix.CONFIG_OF_MAX_CONCURRENT_MULTI_STAGE_QUERIES,
-                  CommonConstants.Helix.DEFAULT_MAX_CONCURRENT_MULTI_STAGE_QUERIES));
+      int maxServerQueryThreads = Integer.parseInt(_helixAdmin.getConfig(_helixConfigScope,
+              Collections.singletonList(CommonConstants.Helix.CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS))
+          .getOrDefault(CommonConstants.Helix.CONFIG_OF_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS,
+              CommonConstants.Helix.DEFAULT_MULTI_STAGE_ENGINE_MAX_SERVER_QUERY_THREADS));
 
-      if (_maxConcurrentQueries == maxConcurrentQueries) {
+      if (_maxServerQueryThreads == maxServerQueryThreads) {
         return;
       }
 
-      if (_maxConcurrentQueries <= 0 && maxConcurrentQueries > 0
-          || _maxConcurrentQueries > 0 && maxConcurrentQueries <= 0) {
+      if (_maxServerQueryThreads <= 0 && maxServerQueryThreads > 0
+          || _maxServerQueryThreads > 0 && maxServerQueryThreads <= 0) {
         // This operation isn't safe to do while queries are running so we require a restart of the broker for this
         // change to take effect.
         LOGGER.warn("Enabling or disabling limitation of the maximum number of multi-stage queries running "
             + "concurrently requires a restart of the broker to take effect");
         return;
       }
 
-      if (maxConcurrentQueries > 0) {
-        _semaphore.setPermits(Math.max(1, maxConcurrentQueries * _numServers / _numBrokers));
+      if (maxServerQueryThreads > 0) {
+        _semaphore.setPermits(Math.max(1, maxServerQueryThreads * _numServers / _numBrokers));
       }
-      _maxConcurrentQueries = maxConcurrentQueries;
+      _maxServerQueryThreads = maxServerQueryThreads;
     }
   }