[BUG] Basalt VIO race condition #1510
Description
Currently this line in the Basalt VIO node:
depthai-core/src/basalt/BasaltVIO.cpp
Line 85 in 16319cb
can cause a segfault due to a race condition when updating the shared_ptr reference for leftImg from both the pipeline thread and the host node thread. (at least I think that's what is happening, see stack trace below).
Either the passthrough should be removed, or access to leftImg should be protected.
oakd-1 | [component_container-1] Stack trace (most recent call last) in thread 30649:
oakd-1 | [component_container-1] #6 Object "/usr/lib/aarch64-linux-gnu/ld-linux-aarch64.so.1", at 0xffffffffffffffff, in
oakd-1 | [component_container-1] #5 Object "/usr/lib/aarch64-linux-gnu/libc.so.6", at 0xffffb6f09e9b, in
oakd-1 | [component_container-1] #4 Object "/usr/lib/aarch64-linux-gnu/libc.so.6", at 0xffffb6ea0397, in
oakd-1 | [component_container-1] #3 Object "/usr/lib/aarch64-linux-gnu/libstdc++.so.6.0.30", at 0xffffb70d29cb, in
oakd-1 | [component_container-1] #2 | Source "/usr/include/c++/11/bits/std_thread.h", line 211, in operator()
oakd-1 | [component_container-1] | 210: void
oakd-1 | [component_container-1] | > 211: _M_run() { _M_func(); }
oakd-1 | [component_container-1] | 212: };
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/std_thread.h", line 266, in _M_invoke<0>
oakd-1 | [component_container-1] | 264: using _Indices
oakd-1 | [component_container-1] | 265: = typename _Build_index_tuple<tuple_size<_Tuple>::value>::__type;
oakd-1 | [component_container-1] | > 266: return _M_invoke(_Indices());
oakd-1 | [component_container-1] | 267: }
oakd-1 | [component_container-1] | 268: };
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/std_thread.h", line 259, in __invoke<dai::ThreadedNode::start()::<lambda()> >
oakd-1 | [component_container-1] | 257: typename __result<_Tuple>::type
oakd-1 | [component_container-1] | 258: _M_invoke(_Index_tuple<_Ind...>)
oakd-1 | [component_container-1] | > 259: { return std::__invoke(std::get<_Ind>(std::move(_M_t))...); }
oakd-1 | [component_container-1] | 260:
oakd-1 | [component_container-1] | 261: typename __result<_Tuple>::type
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/invoke.h", line 96, in __invoke_impl<void, dai::ThreadedNode::start()::<lambda()> >
oakd-1 | [component_container-1] | 94: using __type = typename __result::type;
oakd-1 | [component_container-1] | 95: using __tag = typename __result::__invoke_type;
oakd-1 | [component_container-1] | > 96: return std::__invoke_impl<__type>(__tag{}, std::forward<_Callable>(__fn),
oakd-1 | [component_container-1] | 97: std::forward<_Args>(__args)...);
oakd-1 | [component_container-1] | 98: }
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/invoke.h", line 61, in operator()
oakd-1 | [component_container-1] | 59: constexpr _Res
oakd-1 | [component_container-1] | 60: __invoke_impl(__invoke_other, _Fn&& __f, _Args&&... __args)
oakd-1 | [component_container-1] | > 61: { return std::forward<_Fn>(__f)(std::forward<_Args>(__args)...); }
oakd-1 | [component_container-1] | 62:
oakd-1 | [component_container-1] | 63: template<typename _Res, typename _MemFun, typename _Tp, typename... _Args>
oakd-1 | [component_container-1] Source "/opt/ros_ws/src/depthai-core/src/pipeline/ThreadedNode.cpp", line 35, in _M_run [0xffff7d7c24ab]
oakd-1 | [component_container-1] 32: running = true;
oakd-1 | [component_container-1] 33: thread = std::thread([this]() {
oakd-1 | [component_container-1] 34: try {
oakd-1 | [component_container-1] > 35: run();
oakd-1 | [component_container-1] 36: } catch(const MessageQueue::QueueException& ex) {
oakd-1 | [component_container-1] 37: // catch the exception and stop the node
oakd-1 | [component_container-1] 38: auto expStr = fmt::format("Node stopped with a queue exception: {}", ex.what());
oakd-1 | [component_container-1] #1 | Source "/opt/ros_ws/src/depthai-core/src/basalt/BasaltVIO.cpp", line 85, in ~shared_ptr
oakd-1 | [component_container-1] | 83: auto out = std::make_shared<TransformData>(trans.x(), trans.y(), trans.z(), rot.x(), rot.y(), rot.z(), rot.w());
oakd-1 | [component_container-1] | 84: transform.send(out);
oakd-1 | [component_container-1] | > 85: passthrough.send(leftImg);
oakd-1 | [component_container-1] | 86: }
oakd-1 | [component_container-1] | 87: }
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/shared_ptr.h", line 122, in ~__shared_ptr
oakd-1 | [component_container-1] | 120: */
oakd-1 | [component_container-1] | 121: template<typename _Tp>
oakd-1 | [component_container-1] | > 122: class shared_ptr : public __shared_ptr<_Tp>
oakd-1 | [component_container-1] | 123: {
oakd-1 | [component_container-1] | 124: template<typename... _Args>
oakd-1 | [component_container-1] | Source "/usr/include/c++/11/bits/shared_ptr_base.h", line 1154, in ~__shared_count
oakd-1 | [component_container-1] | 1152: __shared_ptr(const __shared_ptr&) noexcept = default;
oakd-1 | [component_container-1] | 1153: __shared_ptr& operator=(const __shared_ptr&) noexcept = default;
oakd-1 | [component_container-1] | >1154: ~__shared_ptr() = default;
oakd-1 | [component_container-1] | 1155:
oakd-1 | [component_container-1] | 1156: template<typename _Yp, typename = _Compatible<_Yp>>
oakd-1 | [component_container-1] Source "/usr/include/c++/11/bits/shared_ptr_base.h", line 705, in run [0xffff7d9e59c7]
oakd-1 | [component_container-1] 702: ~__shared_count() noexcept
oakd-1 | [component_container-1] 703: {
oakd-1 | [component_container-1] 704: if (_M_pi != nullptr)
oakd-1 | [component_container-1] > 705: _M_pi->_M_release();
oakd-1 | [component_container-1] 706: }
oakd-1 | [component_container-1] 707:
oakd-1 | [component_container-1] 708: __shared_count(const __shared_count& __r) noexcept
oakd-1 | [component_container-1] #0 Source "/usr/include/c++/11/bits/shared_ptr_base.h", line 168, in __stack_chk_fail [0xaaaaba7b27c8]
oakd-1 | [component_container-1] 165: if (__gnu_cxx::__exchange_and_add_dispatch(&_M_use_count, -1) == 1)
oakd-1 | [component_container-1] 166: {
oakd-1 | [component_container-1] 167: _GLIBCXX_SYNCHRONIZATION_HAPPENS_AFTER(&_M_use_count);
oakd-1 | [component_container-1] > 168: _M_dispose();
oakd-1 | [component_container-1] 169: // There must be a memory barrier between dispose() and destroy()
oakd-1 | [component_container-1] 170: // to ensure that the effects of dispose() are observed in the
oakd-1 | [component_container-1] 171: // thread that runs destroy().
oakd-1 | [component_container-1] Segmentation fault (Address not mapped to object [0xffff2808f])