[Inference API] Fix testExecute_NotifiesNonRateLimitedTasksOfShutdown (elastic#137876)

Author: timgrein

muted-tests.yml

@@ -417,9 +417,6 @@ tests:
 - class: org.elasticsearch.xpack.inference.external.http.sender.RequestExecutorServiceTests
   method: testChangingCapacity_DoesNotRejectsOverflowTasks_BecauseOfQueueFull
   issue: https://github.com/elastic/elasticsearch/issues/137823
-- class: org.elasticsearch.xpack.inference.external.http.sender.RequestExecutorServiceTests
-  method: testExecute_NotifiesNonRateLimitedTasksOfShutdown
-  issue: https://github.com/elastic/elasticsearch/issues/137831
 - class: org.elasticsearch.xpack.inference.integration.AuthorizationTaskExecutorMultipleNodesIT
   method: testAuthorizationTaskGetsRelocatedToAnotherNode_WhenTheNodeThatIsRunningItShutsDown
   issue: https://github.com/elastic/elasticsearch/issues/137911

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/external/http/sender/RequestExecutorServiceTests.java

@@ -17,6 +17,7 @@
 import org.elasticsearch.core.Nullable;
 import org.elasticsearch.core.TimeValue;
 import org.elasticsearch.inference.InferenceServiceResults;
+import org.elasticsearch.inference.InputType;
 import org.elasticsearch.test.ESTestCase;
 import org.elasticsearch.threadpool.Scheduler;
 import org.elasticsearch.threadpool.ThreadPool;
@@ -330,7 +331,7 @@ public void testExecute_NotifiesNonRateLimitedTasksOfShutdown() {
 
         var requestManager = RequestManagerTests.createMockWithRateLimitingDisabled(mock(RequestSender.class), "id");
         var listener = new PlainActionFuture<InferenceServiceResults>();
-        service.execute(requestManager, new EmbeddingsInput(List.of(), InputTypeTests.randomWithNull()), null, listener);
+        service.execute(requestManager, new EmbeddingsInput(List.of(), InputType.SEARCH), null, listener);
 
         service.shutdown();
         service.start();