boost asio Io_server problem

[Former user]

I get a segmentation fault when I use trac_ik at:

std::size_t io_service::run() { boost::system::error_code ec; std::size_t s = impl_.run(ec); boost::asio::detail::throw_error(ec); return s; }

"Works" if I comment out in trac_ik.cpp: initialize() ...

// threads.create_thread(boost::bind(&boost::asio::io_service::run, // &io_service)); // threads.create_thread(boost::bind(&boost::asio::io_service::run, // &io_service));

Seen this before?

Observations: * I noticed if I single stepped real slow in Netbeans it didn't bomb. * I tried to catch exception, but couldn't * I don't use boost asio in this ROS package. I have had problems with boost asio io_server in other ROS packages so I know its prickly.

TIA, John

[Patrick Beeson]

I've never heard of anyone else having this issue.

1) whats the OS/ROS version you are using? 2) can you install the tracik deb files and reproduce the segfault?

[John Michaloski]

ROS indigo on ubuntu 14.04

I am not using moveit so trac_ik is not a plugin. I downloaded code and made into local package.

I am used to boost asio issues with io_server.

Thanks for you help.

[Patrick Beeson]

This was originally built for 14.04 indigo. No one has seeen these issues in many thousands of hours of testing/usage.

Are you using the standard indigo Boost libraries that compile via apt?

Could you help me diagnose by removing trac-ik from your catkin workspace, installing ros-indigo-trac-ik via apt, and link your application against that?

[John Michaloski]

Ok, I'll apt-get ros-indigo-trac-ik and remove locally.

[John Michaloski]

Removed locally. Same issue with io_server. Already had latest version of trac-ik in ROS.

michalos@woodsy:nistfanuc_ws> sudo apt-get install ros-indigo-trac-ik Reading package lists... Done Building dependency tree
Reading state information... Done ros-indigo-trac-ik is already the newest version.

I am not CMake guru, so commented this line out in CMakeLists.txt: set(CMAKE_CXX_FLAGS "-DBOOST_LOG_DYN_LINK ${CMAKE_CXX_FLAGS}")

boost version (from include file) installed from ROS (I think) - old version for sure: #define BOOST_LIB_VERSION "1_54"

Anyway, like I mentioned I have had issues with io_server.run before. But it was only with one asio thread, and ran ok if io_server.run in main thread.

[Patrick Beeson]

Is there a way you can help me reproduce this deterministically on my machine?

i.e. send me your test case?

Can you run the trac-ik example without this issue?

[John Michaloski]

We are using trac_ik as one of a couple motoman FK/IK solvers.

The thing that faults is the boost asio run threads in trak_ic.cpp

The code is not quite spaghetti but there are meatballs so that I will develop a small trac_ik test case.

[John Michaloski]

When I pare down the ROS workspace to a minimum, the trac_ik spawning of boost asio io_server threads works fine.

[Patrick Beeson]

We are using this as part of a larger Motoman-based provision system and have seen no problems. This uses trac-ik as a library in a planner inside a plugin. No issues. Perhaps something about your larger system is clobbering memory?

[Patrick Beeson]

Was this ever resolved? Again, no one else has reported these issues.

[Patrick Beeson]

On Hold until further info is received.

[Diogo Almeida]

Hi! I just tried to use trac_ik on my machine (ROS Indigo + Ubuntu 14.04) and I have exactly the same issue with a segfault when the threads the OP mentioned are created.

I have no idea on how to help you debug this, but I am willing to try. Right now my solution was to disable the threads and call the solvers sequentially.

[Patrick Beeson]

Does this happen when you install via apt-get install ros-indigo-trac-ik ?

[Diogo Almeida]

I get the segfault when I install via apt-get, yes.

[Patrick Beeson]

Hmm. So the previous discussion indicates that when they paired down their program to a simpler form this went away. We haven't seen this in many hours of testing, so I'm wondering again is if there isn't something in your larger program that might be stepping on memory.

Or perhaps there is an issue in the boost versions in 14.04?

In any case, it people are willing to use c++-11 or higher, it might be worthwhile just switching to C++ futures instead of the boost asio stuff.

[Diogo Almeida]

I will try to just run the test examples that come in the repository and check if I still get this issue.

[Diogo Almeida]

The tests run without an issue. I'll try to figure out what are the differences in my code that lead to this segfault and I'll post back once I have an answer.

[John Michaloski]

I never resolved multithreading issue. Used ikfast instead in larger apps (not with moveit). Trac ik worked fine in test programs.

[Patrick Beeson]

Oh yes, KDL is certainly not thread safe so you can’t make concurrent call to trac-IK solver from multiple threads.

[Patrick Beeson]

I guess I could add mutexes inside the public methods and make the class a singleton so that any multi-threaded application would work without changes.