Fix hang in reload during predict unit test

The issue was improper order of destructor calls which could occur with
specific timing that caused caling method returnStream() on object that
was removed. This resulted in program hanged on notify_one(). This was
specific to that one test that did not mimmick correctly predict path
behaviour.

In test it could occur that OVInferRequestsQueue destructor could be
called first, and then its method returnStream().

*Remove unnecesary *1 suffix

JIRA:CVS-34909

See merge request vmc-eip/OVMS/ovms-c!146

Author: atobiszei

src/test/prediction_service_test.cpp

@@ -141,15 +141,15 @@ void TestPredict::performPredict(const std::string modelName,
                                  std::unique_ptr<std::future<void>> waitBeforePerformInference) {
     // only validation is skipped
     std::shared_ptr<ovms::ModelInstance> modelInstance;
-    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard1;
+    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard;
 
     if (waitBeforeGettingModelInstance) {
         std::cout << "Waiting before getModelInstance. Batch size:" << batchSize << std::endl;
         waitBeforeGettingModelInstance->get();
     }
-    ASSERT_EQ(getModelInstance(manager, modelName, modelVersion, modelInstance, modelInstancePredictRequestsHandlesCountGuard1),
+    ASSERT_EQ(getModelInstance(manager, modelName, modelVersion, modelInstance, modelInstancePredictRequestsHandlesCountGuard),
               ovms::StatusCode::OK);
-    ASSERT_EQ(assureModelInstanceLoadedWithProperBatchSize(*modelInstance, batchSize, modelInstancePredictRequestsHandlesCountGuard1),
+    ASSERT_EQ(assureModelInstanceLoadedWithProperBatchSize(*modelInstance, batchSize, modelInstancePredictRequestsHandlesCountGuard),
               ovms::StatusCode::OK);
     ovms::OVInferRequestsQueue& inferRequestsQueue = modelInstance->getInferRequestsQueue();
     ovms::ExecutingStreamIdGuard executingStreamIdGuard(inferRequestsQueue);
@@ -166,7 +166,6 @@ void TestPredict::performPredict(const std::string modelName,
     auto status = performInference(inferRequestsQueue, executingInferId, inferRequest);
     ASSERT_EQ(status, ovms::StatusCode::OK);
     serializeAndCheck(DUMMY_MODEL_OUTPUT_SIZE * batchSize, inferRequest);
-    modelInstancePredictRequestsHandlesCountGuard1.reset();
 }
 
 TEST_F(TestPredict, SuccesfullOnDummyModel) {
@@ -205,11 +204,11 @@ TEST_F(TestPredict, SuccesfullReloadWhen1InferRequestJustBeforePredict) {
         });
     std::this_thread::sleep_for(std::chrono::seconds(1));
     std::shared_ptr<ovms::ModelInstance> modelInstance;
-    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard1;
-    ASSERT_EQ(getModelInstance(manager, config.getName(), config.getVersion(), modelInstance, modelInstancePredictRequestsHandlesCountGuard1),
+    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard;
+    ASSERT_EQ(getModelInstance(manager, config.getName(), config.getVersion(), modelInstance, modelInstancePredictRequestsHandlesCountGuard),
               ovms::StatusCode::OK);
 
-    std::thread t2([modelInstance, newBatchSize, &modelInstancePredictRequestsHandlesCountGuard1](){modelInstance->reloadModel(newBatchSize, modelInstancePredictRequestsHandlesCountGuard1);});
+    std::thread t2([modelInstance, newBatchSize, &modelInstancePredictRequestsHandlesCountGuard](){modelInstance->reloadModel(newBatchSize, modelInstancePredictRequestsHandlesCountGuard);});
     std::this_thread::sleep_for(std::chrono::milliseconds(100));
     releaseWaitBeforePerformInference.set_value();
     t2.join();
@@ -247,11 +246,11 @@ TEST_F(TestPredict, SuccesfullReloadWhen1InferRequestJustBeforeGettingModelInsta
         });
     std::this_thread::sleep_for(std::chrono::seconds(1));
     std::shared_ptr<ovms::ModelInstance> modelInstance;
-    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard1;
-    ASSERT_EQ(getModelInstance(manager, config.getName(), config.getVersion(), modelInstance, modelInstancePredictRequestsHandlesCountGuard1),
+    std::unique_ptr<ovms::ModelInstancePredictRequestsHandlesCountGuard> modelInstancePredictRequestsHandlesCountGuard;
+    ASSERT_EQ(getModelInstance(manager, config.getName(), config.getVersion(), modelInstance, modelInstancePredictRequestsHandlesCountGuard),
               ovms::StatusCode::OK);
-    std::thread assureProperBSLoadedThread([modelInstance, newBatchSize, &modelInstancePredictRequestsHandlesCountGuard1](){
-            auto status = assureModelInstanceLoadedWithProperBatchSize(*modelInstance, newBatchSize, modelInstancePredictRequestsHandlesCountGuard1);
+    std::thread assureProperBSLoadedThread([modelInstance, newBatchSize, &modelInstancePredictRequestsHandlesCountGuard](){
+            auto status = assureModelInstanceLoadedWithProperBatchSize(*modelInstance, newBatchSize, modelInstancePredictRequestsHandlesCountGuard);
             ASSERT_EQ(status, ovms::StatusCode::OK);
     });
     std::this_thread::sleep_for(std::chrono::milliseconds(100));
@@ -260,8 +259,9 @@ TEST_F(TestPredict, SuccesfullReloadWhen1InferRequestJustBeforeGettingModelInsta
     assureProperBSLoadedThread.join();
 
     ovms::OVInferRequestsQueue& inferRequestsQueue = modelInstance->getInferRequestsQueue();
-    ovms::ExecutingStreamIdGuard executingStreamIdGuard(inferRequestsQueue);
-    int executingInferId = executingStreamIdGuard.getId();
+    // exception from keeping the same as predict path - using unique_ptr to keep order of destructors the same
+    auto executingStreamIdGuard = std::make_unique<ovms::ExecutingStreamIdGuard>(inferRequestsQueue);
+    int executingInferId = executingStreamIdGuard->getId();
     InferenceEngine::InferRequest& inferRequest = inferRequestsQueue.getInferRequest(executingInferId);
 
     std::vector<float> input(DUMMY_MODEL_INPUT_SIZE * newBatchSize);
@@ -273,7 +273,8 @@ TEST_F(TestPredict, SuccesfullReloadWhen1InferRequestJustBeforeGettingModelInsta
     serializeAndCheck(DUMMY_MODEL_OUTPUT_SIZE * newBatchSize, inferRequest);
 
     std::cout << "Now releasing blockade from reloading." << std::endl;
-    modelInstancePredictRequestsHandlesCountGuard1.reset();
+    executingStreamIdGuard.reset();
+    modelInstancePredictRequestsHandlesCountGuard.reset();
     secondPredictRequest.join();
 }