You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As discussed in the March 11th monthly meetup, we have some occasional unit test failures. These failures have seemingly been narrowed down to rclrs_tests and appear to be related to DDS usage. See #364 for more details.
With that in mind, it is imperative that we take multiple approaches to thoroughly investigate and analyze the root causes of these failures.
For this issue, we agreed that it would be beneficial to attach GDB to some release builds that contain debug symbols, in a fresh ros2 workspace. As I understand it, the hope was we could get a nicer stack trace for our intermittent failures than what we saw in #364.
The text was updated successfully, but these errors were encountered:
I actually wasn't able to get gdb to work correctly on any test cases because the rapid creation and destruction of threads was leaving gdb without a stack to latch onto.
But it's reassuring that you're no longer reproducing it, and it's a good indication that this was a drop order issue on our end all along.
I'm skeptical that we're totally out of the woods yet. There may be potential drop orders that just don't happen to show up in our tests. So I'll still plan on auditing the rcl handlers more closely.
As discussed in the March 11th monthly meetup, we have some occasional unit test failures. These failures have seemingly been narrowed down to
rclrs_tests
and appear to be related to DDS usage. See #364 for more details.With that in mind, it is imperative that we take multiple approaches to thoroughly investigate and analyze the root causes of these failures.
For this issue, we agreed that it would be beneficial to attach GDB to some release builds that contain debug symbols, in a fresh ros2 workspace. As I understand it, the hope was we could get a nicer stack trace for our intermittent failures than what we saw in #364.
The text was updated successfully, but these errors were encountered: