multiple pipelines using splitmuxsink causing Out Of Memory (OOM)

[Tarun Tej K]

Hi,



I want to use the following gstreamer pipeline for multiple IP cameras as
an infinite process to keep the video buffer of a fixed duration always.

Below is a sample pipeline I am using:



gst-launch-1.0  -e rtspsrc
location="rtsp://<user>:<password>@<IP_ADDR>:554/cam/realmonitor?channel=1&subtype=0"
name = d \

 !  rtph264depay ! h264parse  ! splitmuxsink name = mux
location=a_video%02d.mp4 max-size-bytes=5000000 max-size-time=2000000000
max-files=5



But when I run this pipeline for 4 IP cameras  on the actual target (iMX6UL
<https://www.nxp.com/products/processors-and-microcontrollers/arm-based-processors-and-mcus/i.mx-applications-processors/i.mx-6-processors/i.mx-6ultralite-processor-low-power-secure-arm-cortex-a7-core:i.MX6UL?tab=Design_Tools_Tab>)
I am facing an excessive memory usage where ultimately the OOM_Killer will
be started and one of the application gets killed.

I have tried the same scenario on two three other setups below are the
observations:



   1. *iMX6UL – 696 MHz** -  512MB RAM* - Custom Built using Yocto (
   imx-linux-rocko
   <https://source.codeaurora.org/external/imx/imx-manifest/tree/?h=imx-linux-rocko>)
   – gstreamer 1.14.2

Ran the 4 pipelines. RAM’s used memory kept growing rapidly, and within 15
minutes OOM_Killer task has killed few processes.

Also observed that some segment files generated by splitmuxsink are growing
upto 80MB at times. This should ideally not cross more than 5MB
(max-size-bytes = 5000000 and max-size-time=2000000000)



   1. *Raspberry Pi3(Model B) – 1.2 GHz - 1GB RAM* - Custom Built using
   Yocto (/meta-raspberrypi/tree/recipes-core/images/rpi-basic-image.bb
   <https://git.yoctoproject.org/cgit/cgit.cgi/meta-raspberrypi/tree/recipes-core/images/rpi-basic-image.bb>)
   - gstreamer 1.14.2

Ran the 4 pipelines. RAM’s used memory kept growing rapidly to 600 MB
within 30 minutes, didn’t wait till OOM_Killer to start. Manually stopped
the processes.



   1. *x86 – Intel Xeon - 6GB RAM - Ubuntu 18.04* – gstreamer 1.14.1
   (pre-built)

Ran the 4 pipelines for about 6 hours. Consistent RAM of 96MB used memory
for these pipelines. No OOM_Kill situation occurred



   1. *Raspberry Pi3(Model B) – 1.2 GHz - 1GB RAM - debian* – gstreamer
   1.10.x (pre-built)

Ran the 4 pipelines for about 1 hour. Consistent RAM of 96MB used memory.
No OOM_Kill situation occurred





>From these observations it looks like some additional settings related to
Memory management needs to be done to use the memory efficiently.  Cases 3
and 4 show that processes actually doesn’t need more than 100MB of memory
but somehow the memory is being used excessively in cases 1 and 2. Any idea
what settings will prevent this situation?

I have tried GST_DEBUG with TRACE but unable to deduce anything the flood
of debug messages. Is there a better way to debug only the buffers flowing
in and out of splimuxsink?


Thanks

Tarun
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/gstreamer-devel/attachments/20180924/30467d3d/attachment.html>