Skip to content

Latest commit

 

History

History
194 lines (122 loc) · 9.96 KB

README.md

File metadata and controls

194 lines (122 loc) · 9.96 KB
Raw

gstreamer-playground

Balena-based gstreamer debug and profiling setup

gstreamer is incredibly powerful but notoriously hard to learn, use and debug.

This project aim is to help debug and profile gstreamer pipeline.

Note that it will install gstreamer from the ubuntu repository.

TL;DR

balena deploy button

  1. Deploy to your balena device
  2. Set the ENV vars for your pipeline
  3. Point your browser to your balena device url or ip to get the result
Var Type Default Description
PIPELINE string videotestsrc ! fakesink Pipeline to execute
STOP_PIPELINE_AFTER_SEC int (seconds) 10 Stop the pipline after duration
GST_TRACERS string (tracers separated by ;) cpuusage;framerate Gst-Shark tracers to use

(cf all env, at the end of this readme)

Architecture

This currently should runs on AMD64 (tested) and Raspberry pi (ARMv7 and aarch64, untested).

It could run on a jetson (aarch64) but without hardware accelleration (todo)

What's inside ?

  • ubuntu
  • gstreamer
  • gstreamer-plugins-good
  • gstreamer-plugins-bad
  • gstreamer-plugins-ugly
  • gst-shark
  • graphviz
  • v4l-utils
  • mkvtoolnix
  • alsa
  • xserver (balenablocks/xserver)
  • files (edwin3/files)

Running a pipe using gst-launch

Connect using ssh (either thru balena cli or cloud) into the service.

Use gst-launch-1.0 -ve YOUR ! TEST ! PIPE. Let it run as long as you need, press CTRL+C to interupt it.

Using ENV

If you set PIPELINE env to a pipe, it will be run at startup.

ie : PIPELINE=videotestsrc ! fakesink

You can limit the running duration of the pipeline with STOP_PIPELINE_AFTER_SEC (in sec).

Log level

Log level is set to 7 by default. You can change it to any level using the GST_DEBUG environment variable (in docker-compose.yml, balena cloud, or in the running container).

Valid Levels are 0 to 9, check the gstreamer doc for more infos.

Visualizing a pipeline

Thanks to GST_DEBUG_DUMP_DOT_DIR env var set in docker-compose.yml, gstreamer will output a dot file at each change of pipeline status. (PREROLL, RUNNING, ...).

Dots files will be automatically turned into PDF and put into the /files/ folder.

You can retrieve them using the web interface (cf Acessing files below)

Note that files will be erase at startup, unless you use the PERSIST_FILES ENV.

If you prefer png over pdf, set DOT_PROCESSING_PNG to true.

Manually turn the dots into images

Turn off the automatic process by adding NO_DOT_PROCESSING to the env.

The dot files are stored in /files/dots which are served on port 80 by edwin3/files block.

You can turn those to png or pdf with the following command :

dot -Tpng /files/dots/DOT_FILE_OF_INTEREST.dot > /files/pipeline.png

or

dot -Tpdf /files/dots/DOT_FILE_OF_INTEREST.dot > /files/pipeline.pdf

Accessing files

You can access the files (dot, png, pdf, ...) using the files webserver.

Everything you put inside /files will be available at http://ip_of_your_device (or http://short_hash_of_your_device.local)

You can also activate the public url in balenacloud and access those thru there.

Using GSTShark

(this is currently broken)

Gstshark is a profiling tool for your gstreamer pipeline. It's made by great folks at https://ridgerun.com.

The doc of the tool is on their wiki, be sure to read it.

At the time of writing, GSTShark includes 9 tracers :

  • InterLatency : Measures the latency time at different points in the pipeline.
  • ProcTime : Measures the time an element takes to produce an output given the corresponding input.
  • Framerate : Measures the amount of frames that go through a src pad every second.
  • ScheduleTime : Measures the amount of time between two consecutive buffers in a sink pad.
  • CPUUsage : Measures the CPU usage every second. In multiprocessor systems this measurements are presented per core.
  • Graphic : Records a graphical representation of the current pipeline.
  • Bitrate : Measures the current stream bitrate in bits per second.
  • Queue Level : Measures the amount of data queued in every queue element in the pipeline.
  • Buffer : Prints information of every buffer that passes through every sink pad in the pipeline. This information contains PTS and DTS, duration, size, flags and even refcount.

To activate GSTShark, you need to set a few vars in the environment (again using docker-compose, balenacloud or directly on the device) :

  • GST_DEBUG="GST_TRACER:7"
  • GST_TRACERS="cpuusage"
  • GST_SHARK_LOCATION=/files/traces (set by default)

You can choose multiple tracer at the same time if you like, just separate them with ';' (ie. "queue;framerate")

There's other options you can set if needed, refer to the GSTShark doc for more infos.

Tracers accepts parameters using this synthax : GST_TRACERS="tracer(parameter=value)" The list of paramters for each tracer is in the official doc

Parsing traces to generate graphs

To make sense of the traces files generated by GSTShark, the easiest way it to plot them on a graph.

To do so GSTShark comes with a dedicated tool called gstshark-plot.

The traces will be auto-converted to pdf a few seconds after the end of the pipeline and put in the /files folder (cf Accessing Files).

Manually processing traces

You can turn off the auto-processing of traces by setting the NO_TRACE_PROCESSING env.

It's available in /gst-shark/scripts/graphics/gstshark-plot.

Usage is : cd /gst-shark/scripts/graphics/ && ./gstshark-plot /files/traces -s pdf -l outside

You can change the output format to png by replacing pdf, and choose another place for the legends (inside, outside or extern).

XServer

An X11 Server is launched thanks to balenablocks/xserver.

The start script will wait for the server to be ready before processing something else. You can turn off the wait by setting NOWAIT_X to true in the ENV.

Other tools

  • Alsa : for audio test, you can use arecord and aplay to test audio devices
  • v4l-utils : for changing camera parameters or listing devices with v4l2-ctl
  • mkvtoolnix : if you want to play around with matroska files

Useful commands :

  • arecord -l : list audio devices availble for recording
  • aplay -l : list audio devices availble for playing
  • v4l2-ctl --list-devices : list video devices

ENV

Var Type Default Description
PIPELINE string videotestsrc ! fakesink Pipeline to execute
STOP_PIPELINE_AFTER_SEC int (seconds) 10 Stop the pipline after duration
GST_TRACERS string (tracers separated by ;) cpuusage;framerate Gst-Shark tracers to use
GST_DEBUG string GST_TRACER:7 Log Level (7 to get the traces) with filters to only get GST_TRACER from Gst-Shark
GST_DEBUG_DUMP_DOT_DIR string /files/dots/ Directory to ouput the dot files
PERSIST_FILES boolean undefined If set, will not delete previous dots, traces and graphs on rerun; it will probably override some of them tho
DOT_PROCESSING_PNG boolean undefined Output the graphical representation of the pipeline as a PNG instead of PDF
NO_DOT_PROCESSING boolean undefined Prevent the production of a graphical representation of the pipeline as PDF or PNG
GST_SHARK_LOCATION string /files/traces/ Directory to ouput the traces files from gst-shark
NO_TRACE_PROCESSING boolean undefined Prevent the production of a graphical representation of the traces as PDF
NOWAIT_X boolean undefined Don't wait for a X Server to be ready before running the pipeline

TODO and next steps for this project

  • make a jetson version
  • make a webpage with options to run an arbitrary pipe and get the resulting visualizations and graphs
  • build a living library of useful pipelines
  • push everything to balena hub