Fix testStopQueryLocal (#131130)

By default, ES|QL uses all workers in the esql_worker threadpool to
execute drivers on data nodes. If a node is both data and coordinator,
and all drivers are blocked by the allowEmitting latch, there are no
workers left to execute the final driver or fetch pages from remote
clusters. This can prevent remote clusters from being marked as
successful on the coordinator, even if they have completed. To avoid
this, we reserve at least one worker for the final driver and page
fetching. A single worker is enough, as these two tasks can be paused
and yielded.

Closes #121672

(cherry picked from commit b325f2b)

Author: dnhatn

muted-tests.yml

@@ -224,9 +224,6 @@ tests:
 - class: org.elasticsearch.packaging.test.BootstrapCheckTests
   method: test20RunWithBootstrapChecks
   issue: https://github.com/elastic/elasticsearch/issues/124940
-- class: org.elasticsearch.xpack.esql.action.CrossClusterAsyncQueryStopIT
-  method: testStopQueryLocal
-  issue: https://github.com/elastic/elasticsearch/issues/121672
 - class: org.elasticsearch.packaging.test.BootstrapCheckTests
   method: test10Install
   issue: https://github.com/elastic/elasticsearch/issues/124957

x-pack/plugin/esql/src/internalClusterTest/java/org/elasticsearch/xpack/esql/action/CrossClusterAsyncQueryStopIT.java

@@ -13,7 +13,10 @@
 import org.elasticsearch.logging.LogManager;
 import org.elasticsearch.logging.Logger;
 import org.elasticsearch.tasks.TaskInfo;
+import org.elasticsearch.transport.TransportService;
 import org.elasticsearch.xpack.core.async.AsyncStopRequest;
+import org.elasticsearch.xpack.esql.plugin.EsqlPlugin;
+import org.elasticsearch.xpack.esql.plugin.QueryPragmas;
 
 import java.util.Iterator;
 import java.util.List;
@@ -132,13 +135,21 @@ public void testStopQueryLocal() throws Exception {
 
         Tuple<Boolean, Boolean> includeCCSMetadata = randomIncludeCCSMetadata();
         boolean responseExpectMeta = includeCCSMetadata.v2();
-
-        final String asyncExecutionId = startAsyncQuery(
+        // By default, ES|QL uses all workers in the esql_worker threadpool to execute drivers on data nodes.
+        // If a node is both data and coordinator, and all drivers are blocked by the allowEmitting latch,
+        // there are no workers left to execute the final driver or fetch pages from remote clusters.
+        // This can prevent remote clusters from being marked as successful on the coordinator, even if they
+        // have completed. To avoid this, we reserve at least one worker for the final driver and page fetching.
+        // A single worker is enough, as these two tasks can be paused and yielded.
+        var threadpool = cluster(LOCAL_CLUSTER).getInstance(TransportService.class).getThreadPool();
+        int maxEsqlWorkers = threadpool.info(EsqlPlugin.ESQL_WORKER_THREAD_POOL_NAME).getMax();
+        LOGGER.info("--> Launching async query");
+        final String asyncExecutionId = startAsyncQueryWithPragmas(
             client(),
             "FROM blocking,*:logs-* | STATS total=sum(coalesce(const,v)) | LIMIT 1",
-            includeCCSMetadata.v1()
+            includeCCSMetadata.v1(),
+            Map.of(QueryPragmas.TASK_CONCURRENCY.getKey(), between(1, maxEsqlWorkers - 1))
         );
-
         try {
             // wait until we know that the local query against 'blocking' has started
             assertTrue(SimplePauseFieldPlugin.startEmitting.await(30, TimeUnit.SECONDS));