Consider employing `ros2 bag`

[KaoruNishikawa]

Motivation

• Actively maintained in ROS official organization; https://github.com/ros2/rosbag2
• ROS communication will fully be recorded
• Output is quite popular and well-established SQLite format
• Easy copying/slicing of database is realized
• No new development of data recorder is required, except for parameter file saving

Concerns

• Reliability of recorder ros2 bag (database itself must be reliable, as its popularity proves)
• Knowledge on SQL syntax is required to develop the decoder
• SQLite supports minimal data types, e.g. both float and double will consume 8-byte
  • This can cause spectral data record size to be doubled

Notes

• Cannot put arbitrary parameter file inside the database
  • Will be resolved by developing custom database of following structure

    database/
    ┣━ parameter.file
    ┣━ ...
    ┗━ rosbag2_*.db3

• Easy adoption
  • Pre-installed in Docker image ros:*-ros-base
  • To use with docker-launch, add ros2 bag record -a to launch file

[r-yamada1998]

Looks interesting. One issue to be concerned about is writing speed and additional network load. Is writing speed enough for high-frequency publishers such as PID computation? Does additional network load exist? If so, does it cause any problems?

[KaoruNishikawa]

@r-yamada1998 rosbag2 should undergo detailed inspection on recording rate before we employ it, possibly using tools like below. https://index.ros.org/p/rosbag2_performance_benchmarking/ I am somewhat optimistic on this aspect, since ROS 2 design article says >100k small messages nicely recorded within 1s, and SQLite can record 3M data in 1s. https://github.com/ros2/design/blob/ros2bags/articles/rosbags.md#sqlite https://sqlite.org/forum/info/f832398c19d30a4a

On network load, there would be no difference from current recorder with NECSTDB, since both recorders use same approach on capturing messages, i.e. subscribing all the topics.