BackendMemory::Create must release all errors

[lhrios]

This PR changes BackendMemory::Create to make it release errors even if the allocation request succeeded. BackendMemory::Create allows callers to try different allocation types. However, it was not properly releasing the error associated with attempts that failed prior to the one that succeeded.

Valgrind generates the following error while debugging the Triton Server:

==76== 182,619 (70,920 direct, 111,699 indirect) bytes in 1,773 blocks are definitely lost in loss record 1,321 of 1,321
==76==    at 0x4849013: operator new(unsigned long) (in /usr/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
==76==    by 0x5319270: TRITONSERVER_ErrorNew (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x5184A53: TRITONBACKEND_MemoryManagerAllocate (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x15F16833: triton::backend::BackendMemory::Create(TRITONBACKEND_MemoryManager*, triton::backend::BackendMemory::AllocationType, long, unsigned long, triton::backend::BackendMemory**) (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x15F17B5A: triton::backend::BackendMemory::Create(TRITONBACKEND_MemoryManager*, std::vector<triton::backend::BackendMemory::AllocationType, std::allocator<triton::backend::BackendMemory::AllocationType> > const&, long, unsigned long, triton::backend::BackendMemory**) (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x15EC4945: triton::backend::onnxruntime::ModelInstanceState::SetStringInputTensor(TRITONBACKEND_Request**, unsigned int, std::vector<TRITONBACKEND_Response*, std::allocator<TRITONBACKEND_Response*> >*, char const*, std::vector<char const*, std::allocator<char const*> >*, bool*) (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x15EC748A: triton::backend::onnxruntime::ModelInstanceState::SetInputTensors(unsigned long, TRITONBACKEND_Request**, unsigned int, std::vector<TRITONBACKEND_Response*, std::allocator<TRITONBACKEND_Response*> >*, triton::backend::BackendInputCollector*, std::vector<char const*, std::allocator<char const*> >*, bool*) (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x15ECDF6E: triton::backend::onnxruntime::ModelInstanceState::ProcessRequests(TRITONBACKEND_Request**, unsigned int) (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x15ED1B83: TRITONBACKEND_ModelInstanceExecute (in /opt/tritonserver/backends/onnxruntime/libtriton_onnxruntime.so)
==76==    by 0x51A3F13: triton::core::TritonModelInstance::Execute(std::vector<TRITONBACKEND_Request*, std::allocator<TRITONBACKEND_Request*> >&) (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x51A427A: triton::core::TritonModelInstance::Schedule(std::vector<std::unique_ptr<triton::core::InferenceRequest, std::default_delete<triton::core::InferenceRequest> >, std::allocator<std::unique_ptr<triton::core::InferenceRequest, std::default_delete<triton::core::InferenceRequest> > > >&&) (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x52B505C: triton::core::Payload::Execute(bool*) (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x51A86A3: triton::core::TritonModelInstance::TritonBackendThread::BackendThread() (in /opt/tritonserver/lib/libtritonserver.so)
==76==    by 0x6940252: ??? (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.30)
==76==    by 0x6B24AC2: start_thread (pthread_create.c:442)
==76==    by 0x6BB5A03: clone (clone.S:100)

[Tabrizian]

Thanks for your contribution @lhrios! Could you please sign the CLA as instructed here: https://github.com/triton-inference-server/server/blob/main/CONTRIBUTING.md

[casassg]

@Tabrizian we signed the CLA from Block (including Luis here as contributor) for this PR: triton-inference-server/onnxruntime_backend#195 Happy to re-send it over.

[Tabrizian]

Thanks @casassg. I think in that case we just need to run it through the CI and merge it.

[lhrios]

run it through the CI

Hey @Tabrizian . By running it through the CI you mean following these steps or has it been automated?

[Tabrizian]

At this point there are no actions required on your end. I'll run it through the CI to make sure there are no issues with that.

[lhrios]

At this point there are no actions required on your end. I'll run it through the CI to make sure there are no issues with that.

Hey @Tabrizian. Would you know if this will be released as part of the next Triton version (the one after 23.12)?

[Tabrizian]

@lhrios This should be part of the 24.02 release.