Fix StressMemLeaksTests with several models
#19986
Conversation
| @@ -97,14 +95,14 @@ infer_request_inference(const std::string &model, const std::string &target_devi | |||
|
|
|||
|
|
|||
| std::function<void()> reinfer_request_inference(std::shared_ptr<InferApiBase> &ie_wrapper) { | |||
| return [&] { | |||
| return [=] { | |||
| ie_wrapper->infer(); | |||
There was a problem hiding this comment.
I am not sure if it's right here, from my understand of the test code, maybe I am wrong, even if use [=] here, when test_params.numthreads > 1, there will be multi threads use same ie_wrapper to call infer(); one of them will do infer and other threads will throw request busy. so please help to set test_params.numthreads=10 and check if the test case is ok, thanks.
There was a problem hiding this comment.
Thank you for detailed review! The original issue was in consecutive inference, each inference request in separate thread (code snippet may be found in CVS-111490). So it shouldn't cause issues with ie_wrapper
test_params.numthreads is a legacy from StressUnitTests, StressMemLeakTests weren't designed to test parallel execution because it is non-trivial
* Fix `StressMemLeaksTests` with several models * Fix OMZ branch name in `get_testdata.py`
Details:
ie_wrapperwas captured by reference, tests actually execute only last model instead of several models