nvstreammux batch-size bug

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi,

we have found the “nvstreammux” has bug when set property “batch-size”. we reproduce in the attachment. we write a rtsp-server to offer rtsp source. and we use write a demo to reproduce this issue.

In the demo, we start 11 rtsp sources and we set nvstreammux batch-size is 30. This means the bath-size is more than number of souces.

After many hours, maybe 5 hours later or longer, some sources frames can’t go throuh nvstreammux any more, we set a probe in nvstreammux src pad to count the frames.

Thanks.
test.zip (10.9 KB)

2

Can you export GST_DEBUG=nvstreammux:7 and run your application to enable logging?

In the logs you can see from which sources input data is coming.

The log looks like below:

“nvstreammux gstnvstreammux.c:320:gst_nvstreammux_chain:<src_bin_muxer> Got buffer 0x7f4940018ab0 from
source 0”

It means you have received input from source 0.

3

Hi,

I am sorry. I have written a demo for you, but I am very puzzled that I have to test for you, I am not NVIDIA employee. And we have changed our project and not used deepstream framework any more, because It has many bugs for our project.

Thanks.

4

Our QA tried to repro this on Jetson Xavier and Tesla T4 with the demo app given in #1. We were not able to reproduce it after running it for ~ 48 hours.
We also tried the same app with “export GST_DEBUG=nvstreammux:7” as given in #2. We could not see any frames getting dropped.
Deepstream version is 4.0.2

5

Hi,

I am sorry to hear that you can’t reproduce the issue,

and I will show you the phenomenon in our machine later.

Thanks.

6

Hi,

This is our log, from start time 8:40 to 12:57 is normal, and after 12:57, there is no frame come with some cameras. you can see log in attachment.

we use glog instead of gprinter to print timestamp.

LOG(INFO) << "index: " << i << ", frame count: " << test_live[i];

on 1080Ti and deepstream_sdk_v4.0_x86_64

Thanks.
log.zip (61.5 KB)

7

Our QA test version is deepstream_sdk_v4.0.2_x86_64
Test platform: Jetson Xavier and Tesla T4

Do you mind try the latest version v4.0.2 ?

8

Hi,

we have tried the latest version v4.0.2 but also happen the error. We used ‘export GST_DEBUG=nvstreammux:7’, and the log is in attachment. At the beginning, it is normal, but many hours later, the terminal was blocking, stop printing log.

The log is not complete, only the later part is provided．

Thanks.

9

Hi,

And we found the memory would go upper from 500MB to 2G before the program stop.

Thanks.

10

Thanks for sharing. We will check it.

11

Hi,

can you reproduce this issue?

Thanks.

12

From our QA, no repro was found for ~48 hours, tried on both T4 & AGX.