Comment on page
rosbag
In short, the problem is that the current data collection speed is greater than the maximum memory access bandwidth, such that more and more data are queued in the buffer and eventually results in queue overflow. In other words,
rosbag is collecting too much data in a short time, such that the operating system cannot write them all in the memory/disk.There is no actual solution to this issue, as you may always want to record more data at the same time. Some walkarounds are as follows.
- On the recording side, you can 1) reduce the image resolution, 2) reduce the frame rate, or 3) reduce the number of topics to be recorded at the same time. The main objective is to reduce the amount of data written to the memory/disk.
- On the memory side, if the system is working on a removable disk (e.g., SD card), you can try to replace it with a faster one. If it is on your native laptop, as you know, it is hard to replace the disk. (If you can, replacing the traditional disk with SSD would help.)
- On the buffer side, you may use command
rosbag record -b 2048 <topic name>to extend the buffer size. This can postpone the queue overflow (where the warning happens), and give you a bit more time to collect data. - If using the above command, please note that you have to wait for a while after you terminate the
rosbagcommand, and make sure thexxx.bag.activefile has been transformed intoxxx.bagfile. (Some threads in the backend are still working on the rest of data in the buffer.)
In the above example, one may ask "What would happen if this warning prompted?" or "Can I just use this bag file given that some messages are dropped?".
To check if a bag file is healthy or not, there is a useful command
rosbag info xxx.bag. In the following, let's see two examples from EuRoC MAV Dataset. An example of healthy ROS bag looks like the following.
an example of healthy ROS bag
How to tell if it is healthy or not? You can just check if the number of messages is reasonable or not. The total record time is 113 sec, and
/fcu/imu topic has 1140 messages. It means that this topic is running at roughly 10Hz. Similarly, we can infer that the camera frame rate is 20Hz, the raw imu is running at 200Hz, and vicon is running at 100Hz. An example of unhealthy ROS bag could be the following.
an example of unhealthy ROS bag
In this case, the total record time is 116 sec. Therefore, given the frame rate of 20Hz, we expect that the
image_raw topic should have about 2320 messages. However, the first camera cam0 only has 1922 messages. It lost 300 messages during the data collection process, perhaps due to the similar buffer issue. This may affect the performance of the algorithms running on the dataset in the future. Last modified 3yr ago