pipe.stop() after set_real_time(False) freezes the program when capturing from a rosbag file

[mcasl]

---

Required Info
Camera Model	rosbag source from D435
Firmware Version	(Open RealSense Viewer --> Click info)
Operating System & Version	{Win 10
Kernel Version (Linux Only)
Platform	PC
SDK Version	2.17.1.457
Language	python
Segment	{Robot/Smartphone/VR/AR/others }

Issue Description

After opening a Rosbag file with real_time = False, stopping the pipe using pipe.stop() freezes the program.
Same code with real_time = True works flawlessly and stops the pipe.

This code freezes the program:

import pyrealsense2 as rs

filename = 'D:\\video.bag'
real_time = False

pipe = rs.pipeline()
cfg  = rs.config()
cfg.enable_device_from_file(filename)
profile  = pipe.start(cfg)
device   = profile.get_device()
playback = rs.playback(device)
playback.set_real_time(real_time)

pipe.stop()

This code works well, the only difference is the value of real_time:

filename = 'D:\\video.bag'
real_time = True

pipe = rs.pipeline()
cfg  = rs.config()
cfg.enable_device_from_file(filename)
profile  = pipe.start(cfg)
device   = profile.get_device()
playback = rs.playback(device)
playback.set_real_time(real_time)

pipe.stop()

This code also freezes the program, the only difference is setting real_time a second time:

filename = 'D:\\video.bag'
real_time = False

pipe = rs.pipeline()
cfg  = rs.config()
cfg.enable_device_from_file(filename)
profile  = pipe.start(cfg)
device   = profile.get_device()
playback = rs.playback(device)
playback.set_real_time(real_time)

playback.set_real_time(True)

pipe.stop()

[lramati]

Is this behavior consistent for you? I ran the second snippet in a loop and couldn't get the program to freeze

[mcasl]

@lramati It's actually the first snippet the one that freezes, not the second. In the second I just change the real time option to True to point out that it works then.

[mcasl]

Possibly related to: #2472

[dorodnic]

@mcasl - I can confirm this is a bug.
For now I work-around #2472, since we need to think how to handle this properly.

[dmandrioli]

The problem occurs erratically in 2.18.1, it's really annoying for production code, I need to kill the process from outside and setting real_time to True is not a workaround since you can loose frames to process. @dorodnic Do you have an actual workaround for this? Thanks.

[dmandrioli]

Well, a working approach is to postpone the processing after caching all frames in memory in real_time mode. Not so annoying :)

[soarwing52]

using the lastest 2.21 version and still getting the freeze situation.

[krejov100]

I am also facing this issue in 2.21

[upxela]

facing this in 2.23 with both python and matlab.

[kevindehecker]

Me too 2.24. (C++ in my case)

[kevindehecker]

I just updated everything, now running kernel 5.0.0-29-generic with realsense api v 2.29, and this problems seems to be fixed!