ros2 param list does not work with multiple rmw implementaion

[fujitatomoya]

Bug report

• Operating System:
  • Ubuntu 20.04
• Installation type:
  • source code
• Version or commit hash:
  • ros2/ros2@57b618f
• DDS implementation:
  • multiple rmw implementation Fast-DDS and Cyclonedds
• Client library (if applicable):
  • rclcpp and rclpy

Steps to reproduce issue

• talker with cyclonedds and ros2 param list with fastdds

it seems that talker does not even receive the service request from client.

[Terminal - 1]
root@5c7a784cba7f:~/ros2_ws/colcon_ws# ros2 run demo_nodes_cpp talker
[INFO] [1611218728.615967478] [talker]: Publishing: 'Hello World: 1'
[INFO] [1611218729.615458541] [talker]: Publishing: 'Hello World: 2'
...<snip>

[Terminal - 2]
root@5c7a784cba7f:~/ros2_ws/colcon_ws# RMW_IMPLEMENTATION=rmw_fastrtps_cpp ros2 param list
Exception while calling service of node '/talker': None

• talker with fastdds and ros2 param list with cyclonedds

this could be security issue since we can crash the node via parameter request.

[Terminal - 1]
root@5c7a784cba7f:~/ros2_ws/colcon_ws# RMW_IMPLEMENTATION=rmw_fastrtps_cpp ros2 run demo_nodes_cpp talker
[INFO] [1611218853.091887892] [talker]: Publishing: 'Hello World: 1'
[INFO] [1611218854.091851082] [talker]: Publishing: 'Hello World: 2'
[INFO] [1611218855.091853219] [talker]: Publishing: 'Hello World: 3'
[INFO] [1611218856.091876166] [talker]: Publishing: 'Hello World: 4'
[INFO] [1611218857.091860947] [talker]: Publishing: 'Hello World: 5'
[INFO] [1611218858.091872255] [talker]: Publishing: 'Hello World: 6'
[INFO] [1611218859.091883693] [talker]: Publishing: 'Hello World: 7'
[INFO] [1611218860.091857624] [talker]: Publishing: 'Hello World: 8'
[INFO] [1611218861.091900999] [talker]: Publishing: 'Hello World: 9'
[INFO] [1611218862.091897370] [talker]: Publishing: 'Hello World: 10'
[INFO] [1611218863.091898792] [talker]: Publishing: 'Hello World: 11'
[INFO] [1611218864.091924685] [talker]: Publishing: 'Hello World: 12'
[INFO] [1611218865.091936687] [talker]: Publishing: 'Hello World: 13'
[INFO] [1611218866.091938711] [talker]: Publishing: 'Hello World: 14'
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc

[Terminal - 2]
root@5c7a784cba7f:~/ros2_ws/colcon_ws# ros2 param list
root@5c7a784cba7f:~/ros2_ws/colcon_ws# ros2 param list
root@5c7a784cba7f:~/ros2_ws/colcon_ws# ros2 param list
Exception while calling service of node '/talker': None

Expected behavior

able to get parameter list for both cases

root@5c7a784cba7f:~/ros2_ws/colcon_ws# ros2 param list
/talker:
  use_sim_time

Actual behavior

either unable to parameter list or application core crashhttps://index.ros.org/doc/ros2/Tutorials/Working-with-multiple-RMW-implementations/#multiple-rmw-implementations

Additional information

according to https://index.ros.org/doc/ros2/Tutorials/Working-with-multiple-RMW-implementations/#multiple-rmw-implementations, there is no constraints / limitation between cyclonedds and fastdds.

[clalancette]

Services do not work cross-vendor at the moment; since parameters are built on services, they won't work either. This is a result of there being no standard way to do services in DDS.

That being said, the crash in the Fast-DDS talker when called from a Cyclone service is worrying. That should not happen. @EduPonz @MiguelCompany could you take a look?

[MiguelCompany]

Seems like a deserialization exception we are not capturing. I will take a look.

[fujitatomoya]

Services do not work cross-vendor at the moment; since parameters are built on services, they won't work either. This is a result of there being no standard way to do services in DDS.

okay, i thought that should work. probably it would be nice to have description about the limitation.

[MiguelCompany]

I have checked that the std::bad_alloc is thrown from Fast CDR code here.

There are two problems here:

1. Fast CDR should check the length before resizing the vector.
2. The rmw should capture Fast CDR exceptions and discard the received message

The first one is being solved by eProsima/Fast-CDR#98.

I will prepare a fix for the second one next week.

[MiguelCompany]

@fujitatomoya @clalancette Please see ros2/rmw_fastrtps#505 for the fix on rmw_fastrtps

[clalancette]

okay, i thought that should work. probably it would be nice to have description about the limitation.

Agreed. I thought we had something in the documentation, but I can't find it. @fujitatomoya please feel free to add something to https://github.com/ros2/ros2_documentation explaining this limitation.

1. Fast CDR should check the length before resizing the vector.

2. The rmw should capture Fast CDR exceptions and discard the received message

Great! Thanks for the fast turnaround.

[fujitatomoya]

@MiguelCompany

wow, that was quick 💨 i will verify both of them to check if that works just in case.

@clalancette

please feel free to add something to https://github.com/ros2/ros2_documentation explaining this limitation.

i will consider this, and will come back to you with PR.

[fujitatomoya]

@MiguelCompany with both fixes, i cannot reproduce the core dump. thanks for the fixes.

i will close this issue after,

• Check sequence lengths when deserializing [10336] eProsima/Fast-CDR#98 merged
• Capture cdr exceptions rmw_fastrtps#505 merged
• documentation is updated

[fujitatomoya]

since eProsima/Fast-CDR#98 and ros2/rmw_fastrtps#505 are merged, the rest is to update documentation which tells that services do not work cross-vendor rmw implementation. i will assign to this on me.