search

ros / class_loader

ROS-independent library for dynamic class (i.e. plugin) introspection and loading from runtime libraries
http://www.ros.org/wiki/class_loader
35 stars 95 forks source link

make the meta-object alive in the lifecycle of the registered plugin #201

Closed iuhilnehc-ynos closed 2 years ago

iuhilnehc-ynos commented 2 years ago

This PR is to address the memory leak about the meta-object is not destroyed during a process exit if using ClassLoader.

memory leak log

```shell chenlh test $ valgrind --leak-check=full ./class_loader_unique_ptr_test --gtest_filter="*.basicLoad" ==63164== Memcheck, a memory error detector ==63164== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al. ==63164== Using Valgrind-3.15.0 and LibVEX; rerun with -h for copyright info ==63164== Command: ./class_loader_unique_ptr_test --gtest_filter=*.basicLoad ==63164== Note: Google Test filter = *.basicLoad [==========] Running 1 test from 1 test suite. [----------] Global test environment set-up. [----------] 1 test from ClassLoaderUniquePtrTest [ RUN ] ClassLoaderUniquePtrTest.basicLoad Meow ******************************************************************************* ***** class_loader impl DEBUG INFORMATION ***** ******************************************************************************* OPEN LIBRARIES IN MEMORY: -------------------------------------------------------------------------------- Open library 0 = libclass_loader_TestPlugins1.so (handle = 0x4ede240) METAOBJECTS (i.e. FACTORIES) IN MEMORY: -------------------------------------------------------------------------------- Metaobject 0 (ptr = 0x4edf050): TypeId = N12class_loader4impl10MetaObjectI3Cat4BaseEE Associated Library = libclass_loader_TestPlugins1.so Associated Loader 0 = 0x1ffeff8000 -------------------------------------------------------------------------------- Metaobject 1 (ptr = 0x4edf6b0): TypeId = N12class_loader4impl10MetaObjectI3Cow4BaseEE Associated Library = libclass_loader_TestPlugins1.so Associated Loader 0 = 0x1ffeff8000 -------------------------------------------------------------------------------- Metaobject 2 (ptr = 0x4edec70): TypeId = N12class_loader4impl10MetaObjectI3Dog4BaseEE Associated Library = libclass_loader_TestPlugins1.so Associated Loader 0 = 0x1ffeff8000 -------------------------------------------------------------------------------- Metaobject 3 (ptr = 0x4edf380): TypeId = N12class_loader4impl10MetaObjectI4Duck4BaseEE Associated Library = libclass_loader_TestPlugins1.so Associated Loader 0 = 0x1ffeff8000 -------------------------------------------------------------------------------- Metaobject 4 (ptr = 0x4edf9e0): TypeId = N12class_loader4impl10MetaObjectI5Sheep4BaseEE Associated Library = libclass_loader_TestPlugins1.so Associated Loader 0 = 0x1ffeff8000 -------------------------------------------------------------------------------- ********************************** END DEBUG ********************************** ******************************************************************************* [ OK ] ClassLoaderUniquePtrTest.basicLoad (161 ms) [----------] 1 test from ClassLoaderUniquePtrTest (167 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test suite ran. (196 ms total) [ PASSED ] 1 test. ==63164== ==63164== HEAP SUMMARY: ==63164== in use at exit: 1,040 bytes in 20 blocks ==63164== total heap usage: 502 allocs, 482 frees, 129,843 bytes allocated ==63164== ==63164== 208 (16 direct, 192 indirect) bytes in 1 blocks are definitely lost in loss record 16 of 20 ==63164== at 0x483DE63: operator new(unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so) ==63164== by 0x48A81A0: ??? ==63164== by 0x48A7667: ??? ==63164== by 0x48A7E35: ??? ==63164== by 0x48A7E7F: ??? ==63164== by 0x4011B99: call_init.part.0 (dl-init.c:72) ==63164== by 0x4011CA0: call_init (dl-init.c:30) ==63164== by 0x4011CA0: _dl_init (dl-init.c:119) ==63164== by 0x4C8E984: _dl_catch_exception (dl-error-skeleton.c:182) ==63164== by 0x40160CE: dl_open_worker (dl-open.c:758) ==63164== by 0x4C8E927: _dl_catch_exception (dl-error-skeleton.c:208) ==63164== by 0x4015609: _dl_open (dl-open.c:837) ==63164== by 0x4EB034B: dlopen_doit (dlopen.c:66) ==63164== ==63164== 208 (16 direct, 192 indirect) bytes in 1 blocks are definitely lost in loss record 17 of 20 ==63164== at 0x483DE63: operator new(unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so) ==63164== by 0x48A8436: ??? ==63164== by 0x48A7815: ??? ==63164== by 0x48A7E41: ??? ==63164== by 0x48A7E7F: ??? ==63164== by 0x4011B99: call_init.part.0 (dl-init.c:72) ==63164== by 0x4011CA0: call_init (dl-init.c:30) ==63164== by 0x4011CA0: _dl_init (dl-init.c:119) ==63164== by 0x4C8E984: _dl_catch_exception (dl-error-skeleton.c:182) ==63164== by 0x40160CE: dl_open_worker (dl-open.c:758) ==63164== by 0x4C8E927: _dl_catch_exception (dl-error-skeleton.c:208) ==63164== by 0x4015609: _dl_open (dl-open.c:837) ==63164== by 0x4EB034B: dlopen_doit (dlopen.c:66) ==63164== ==63164== 208 (16 direct, 192 indirect) bytes in 1 blocks are definitely lost in loss record 18 of 20 ==63164== at 0x483DE63: operator new(unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so) ==63164== by 0x48A86CC: ??? ==63164== by 0x48A79C3: ??? ==63164== by 0x48A7E4D: ??? ==63164== by 0x48A7E7F: ??? ==63164== by 0x4011B99: call_init.part.0 (dl-init.c:72) ==63164== by 0x4011CA0: call_init (dl-init.c:30) ==63164== by 0x4011CA0: _dl_init (dl-init.c:119) ==63164== by 0x4C8E984: _dl_catch_exception (dl-error-skeleton.c:182) ==63164== by 0x40160CE: dl_open_worker (dl-open.c:758) ==63164== by 0x4C8E927: _dl_catch_exception (dl-error-skeleton.c:208) ==63164== by 0x4015609: _dl_open (dl-open.c:837) ==63164== by 0x4EB034B: dlopen_doit (dlopen.c:66) ==63164== ==63164== 208 (16 direct, 192 indirect) bytes in 1 blocks are definitely lost in loss record 19 of 20 ==63164== at 0x483DE63: operator new(unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so) ==63164== by 0x48A8962: ??? ==63164== by 0x48A7B71: ??? ==63164== by 0x48A7E59: ??? ==63164== by 0x48A7E7F: ??? ==63164== by 0x4011B99: call_init.part.0 (dl-init.c:72) ==63164== by 0x4011CA0: call_init (dl-init.c:30) ==63164== by 0x4011CA0: _dl_init (dl-init.c:119) ==63164== by 0x4C8E984: _dl_catch_exception (dl-error-skeleton.c:182) ==63164== by 0x40160CE: dl_open_worker (dl-open.c:758) ==63164== by 0x4C8E927: _dl_catch_exception (dl-error-skeleton.c:208) ==63164== by 0x4015609: _dl_open (dl-open.c:837) ==63164== by 0x4EB034B: dlopen_doit (dlopen.c:66) ==63164== ==63164== 208 (16 direct, 192 indirect) bytes in 1 blocks are definitely lost in loss record 20 of 20 ==63164== at 0x483DE63: operator new(unsigned long) (in /usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so) ==63164== by 0x48A8BF8: ??? ==63164== by 0x48A7D1F: ??? ==63164== by 0x48A7E65: ??? ==63164== by 0x48A7E7F: ??? ==63164== by 0x4011B99: call_init.part.0 (dl-init.c:72) ==63164== by 0x4011CA0: call_init (dl-init.c:30) ==63164== by 0x4011CA0: _dl_init (dl-init.c:119) ==63164== by 0x4C8E984: _dl_catch_exception (dl-error-skeleton.c:182) ==63164== by 0x40160CE: dl_open_worker (dl-open.c:758) ==63164== by 0x4C8E927: _dl_catch_exception (dl-error-skeleton.c:208) ==63164== by 0x4015609: _dl_open (dl-open.c:837) ==63164== by 0x4EB034B: dlopen_doit (dlopen.c:66) ==63164== ==63164== LEAK SUMMARY: ==63164== definitely lost: 80 bytes in 5 blocks ==63164== indirectly lost: 960 bytes in 15 blocks ==63164== possibly lost: 0 bytes in 0 blocks ==63164== still reachable: 0 bytes in 0 blocks ==63164== suppressed: 0 bytes in 0 blocks ==63164== ==63164== For lists of detected and suppressed errors, rerun with: -s ==63164== ERROR SUMMARY: 5 errors from 5 contexts (suppressed: 0 from 0) ```

Signed-off-by: Chen Lihui lihui.chen@sony.com

gbiggs commented 2 years ago

Can you please provide more information about the bug this PR will address?

iuhilnehc-ynos commented 2 years ago

@gbiggs

Thanks for your reply. I have updated the description in https://github.com/ros/class_loader/pull/201#issue-1384752935. In order to not make the relative issue complicated, I used the existing test case to show the problem.

iuhilnehc-ynos commented 2 years ago

TRY CI BUT WITHOUT MACOS:

iuhilnehc-ynos commented 2 years ago
gbiggs commented 2 years ago

Thanks for the contribution!

tylerjw commented 2 years ago

@iuhilnehc-ynos I was hopeful that your much simpler change here would fix the bugs found by asan... but it unfortunately does not: https://github.com/tylerjw/class_loader/actions/runs/3298507574/jobs/5440716254

iuhilnehc-ynos commented 2 years ago

@tylerjw

Thanks for your information, but I think this PR is to address the memory leak about the meta-object is not destroyed during a process exit if using ClassLoader mentioned at https://github.com/ros/class_loader/pull/201#issue-1384752935. To make the sanitizer happy, I have created a new PR below for it.