Performance issue

[mwamp]

Hi,

I just want to report that using ros2_web_bridge saturates one CPU thread on my machine.
The likely issue is that processing power is never yielded as it should and some routines are constantly nagging to check for some event.

Here is a perf report, I produced this with very few topics; actions and services running.

31.44% node libpthread-2.23.so [.] pthread_mutex_lock
27.91% node rclnodejs.node [.] _ZN9rclnodejs8Executor3RunEPv
20.04% node libpthread-2.23.so [.] pthread_mutex_unlock
4.46% node node [.] uv_mutex_unlock
4.33% node node [.] uv_mutex_lock
3.72% node rclnodejs.node [.] uv_mutex_unlock@plt
3.30% node rclnodejs.node [.] uv_mutex_lock@plt
2.31% node node [.] 0x00000000004b5b90
2.24% node node [.] 0x00000000004b5be0

One question, is rclnodejs a wrapper on rclcpp or was is built from scratch. Wouldn't using swig or the like provide a suitable enough wrapper and easy to maintain?

[minggangw]

Thanks for submitting the issue, I'd like to answer them:

The likely issue is that processing power is never yielded as it should and some routines are >constantly nagging to check for some event.

For the unreasonable CPU usage, it should be caused by frequent calling of uv_mutex_lock/uv_mutex_unlock. The reason why it happens is that we set a fixed timeout for rcl_wait currently. This parameter should be customized.

One question, is rclnodejs a wrapper on rclcpp or was is built from scratch.

No, rclnodejs is not based on rclcpp, actually all clients of ROS2, including C++, Python and Node.js are based on rcl which is a C library, not platform related. You could consider rclnodejs is kind of wrapper, meanwhile, implement some platform specific features, like thread.

Wouldn't using swig or the like provide a suitable enough wrapper and easy to maintain?

I'm not familiar with swig, not sure it can meet the requirement to implement the functions.

Finally, you are welcome to fix any problem you found, thanks again!