SOLR-17648: multiThreaded=true: changed queue implementation (#3155)

from unlimited to 1000 max, after which the caller thread will execute.
Didn't need the RejectedExecutionException avoidance hack anymore; Lucene 9.12 has it.
Configurable size: solr.search.multiThreaded.queueSize

Author: dsmiley

solr/CHANGES.txt

@@ -162,7 +162,7 @@ Improvements
   other v2 APIs.  SolrJ now offers (experimental) SolrRequest implementations for all v2 configset APIs in
   `org.apache.solr.client.solrj.request.ConfigsetsApi`. (Jason Gerlowski)
 
-Optimizations                                           
+Optimizations
 ---------------------
 * SOLR-17578: Remove ZkController internal core supplier, for slightly faster reconnection after Zookeeper session loss. (Pierre Salagnac)
 
@@ -213,6 +213,9 @@ Other Changes
 
 * SOLR-17623: SimpleOrderedMap (a NamedList) now implements java.util.Map. (Renato Haeberli, David Smiley)
 
+* SOLR-17648: multiThreaded=true: changed queue implementation from unlimited to 1000 max, after
+  which the caller thread will execute.  (David Smiley)
+
 ==================  9.8.0 ==================
 New Features
 ---------------------

solr/core/src/java/org/apache/solr/search/SolrIndexSearcher.java

@@ -34,7 +34,7 @@
 import java.util.Set;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.ExecutorService;
-import java.util.concurrent.SynchronousQueue;
+import java.util.concurrent.LinkedBlockingQueue;
 import java.util.concurrent.TimeUnit;
 import java.util.concurrent.atomic.AtomicLong;
 import java.util.concurrent.atomic.AtomicReference;
@@ -96,6 +96,7 @@
 import org.apache.solr.common.SolrException.ErrorCode;
 import org.apache.solr.common.params.ModifiableSolrParams;
 import org.apache.solr.common.util.CollectionUtil;
+import org.apache.solr.common.util.EnvUtils;
 import org.apache.solr.common.util.ExecutorUtil.MDCAwareThreadPoolExecutor;
 import org.apache.solr.common.util.ObjectReleaseTracker;
 import org.apache.solr.common.util.SolrNamedThreadFactory;
@@ -228,25 +229,15 @@ public static ExecutorService initCollectorExecutor(NodeConfig cfg) {
       return null;
     }
 
+    // note that Lucene will catch a RejectedExecutionException to just run the task.
+    //  Therefore, we shouldn't worry too much about the queue size.
     return new MDCAwareThreadPoolExecutor(
         indexSearcherExecutorThreads,
         indexSearcherExecutorThreads,
         0L,
         TimeUnit.MILLISECONDS,
-        new SynchronousQueue<>(true) { // fairness
-          // a hack to force ThreadPoolExecutor to block if threads are busy
-          // -- otherwise it will throw RejectedExecutionException; unacceptable
-          @Override
-          public boolean offer(Runnable runnable) { // is supposed to not block, but we do anyway
-            try {
-              put(runnable); // blocks
-            } catch (InterruptedException e) {
-              Thread.currentThread().interrupt();
-              throw new RuntimeException("interrupted submitting to search multi-threaded pool", e);
-            }
-            return true;
-          }
-        },
+        new LinkedBlockingQueue<>(
+            EnvUtils.getPropertyAsInteger("solr.search.multiThreaded.queueSize", 1000)),
         new SolrNamedThreadFactory("searcherCollector")) {
 
       @Override