Collect ideas

[akoerner1]

Mission

The goal of this project is to create a C++ library that can observe ROS2.

Some observations that are interesting for the mission:

• Interrogate ROS for a list of nodes, data types, data structures(messages and message types). - DONE
• Every ROS node/program has a run function. When and for how long did this function execute.
• Data throughput ie when is a message received, how much data etc
• Fingerprinting data and associating it to a specific run execution

Requirements

The following basic information would be interesting to observe:

• Minimal modification of user programs/nodes. A user should not have to make significant changes to their program/node to leverage the observer
• minimal external requirements/dependencies
• Library must output json to file and stdio to support piping
• Include the simulation time and wall clock time in output data
• Run function metadata
  • What host and cpu did the run or node function execute on, function name
• minimal performance overhead
• Include data structure of all messages
• include overall structure of entire ros2 running environment nodes and edges(publishers and subscribers)
• What kind of user interface do we want?

Notes

• When a given node/program "run" function starts and returns with a high precision timer.
• When a subscribed message is received
• Data rate of a given topic.
• hash/fingerprint of the raw data.
• Can we integrate this or leverage lttng https://lttng.org/?

Timeline

First working example by January 31, 2025

Tasks

• Write readme describing project and highlighting inchron and dlr collaboration @akoerner1
• Add ros command line data aggregator tool to repo.
• Provide a sample data structure @vpollex

Discovery

ROS2 has a built in tracing tool: https://github.com/ros2/ros2_tracing

Goals for tracing tool

• experiment with tracer
• Build into ADORe cli
• create parser for tracing data output to json and other formats
• ????? integrate into Inchon tool and dlr big net observer

ROS2 node failure modes

• Node does not receive data in time.
• Data is junk i.e., 1 of every 5 data packets is junk
• Node does not finish critical section/calculation in time.

Ideas

• callbacks per second