- 1 Links
- 2 GStreamer debugging approaches
- 2.1 Use standard GStreamer debug output with filter
- 2.2 Create DOT file to see exact pipeline and capabilities used
- 2.3 Use gst-tracelib library to log key pipeline behavior
- 2.4 Watch system interrupts
- 2.5 Capturing a core dump
- 2.6 Building application and GStreamer libraries with debug enabled
- 2.7 Use gdb to attach to GStreamer application and examine all the threads
- 2.8 Use gdbserver in multi-process mode to attach to GStreamer application and examine all the threads
- 2.9 Exit locked GStreamer application and see what works stand alone
- 2.10 Examine a history of what transpired just prior to lockup
- 2.11 Activate verbose from application
- 2.12 Enable DMAI Davinci Multimedia Application Interface debug output
- 2.13 Enable CE Codec Engine debug output
- 2.14 Use NFS mount to make it easy to transfer files
- 2.15 Look for memory leaks
- 2.16 Use remote syslog
- 2.17 Generated files debug
- 3 GStreamer Performance Tuning
- 4 GStreamer Application Development
- GStreamer debugging tools - http://docs.gstreamer.com/display/GstSDK/Basic+tutorial+11%3A+Debugging+tools
- Good collection of useful hints (including core dumps, using gdb) for GStreamer applications -http://wiki.buzztrax.org/index.php/Debugging
- GStreamer application debugging - http://www.gstreamer.net/data/doc/gstreamer/head/gstreamer/html/gst-running.html
- GStreamer debugging (helpful syntax) - http://www.gstreamer.net/data/doc/gstreamer/head/manual/html/section-checklist-debug.html
- Flumotion approach - http://www.flumotion.net/doc/flumotion/manual/en/trunk/html/chapter-debug.html
GStreamer debugging approaches
Some of these approaches are only useful when you are running a pipeline and audio and/or video stops at an unexpected place in the data stream.
Use standard GStreamer debug output with filter
To see what debug can be enabled, add --gst-debug-help to your GStreamer application arguments, such as:
Then you can see which debug you are interested, such as reference counting and buffer movement, and enable all debug output (level output 5).
gst-launch videotestsrc num-buffers=3 ! fakesink --gst-debug=GST_REFCOUNTING:5,GST_BUFFER:5 --gst-debug-no-color=1 2>&1 | grep "\->0" > log.txt
Gets useful data, but typically slows pipeline performance to the point of being not usable.
Another way you can generate the same output is set the GST_DEBUG shell variable:
GST_DEBUG=GST_REFCOUNTING:5 gst-launch videotestsrc num-buffers=3 ! fakesink
If you get an error like:
WARNING: erroneous pipeline: could not link capsfilter0 to ffenc_mjpeg0
that means the two elements couldn't exchange GStreamer buffers because they couldn't find a common format they both support. To get an insight on what formats are supported, you can add
Create DOT file to see exact pipeline and capabilities used
In order to generate a DOT file, set the GST_DEBUG_DUMP_DOT_DIR environment variable to point to the folder where you want the .dot files to be stored.
1) Install "dot" tool on your host machine...to do so simply install graphviz:
sudo apt-get install graphviz
2) On your target board, set the environment dot dump variable:
export GST_DEBUG_DUMP_DOT_DIR=/root mkdir -p $GST_DEBUG_DUMP_DOT_DIR
3) Run your pipeline, a .dot file will be generated according to each of the state changes that occur showing how the caps are negotiated during the process.
4) Convert the dot files to PNG image files or SVG graphics files
dot -Tpng input.dot > output.png dot -Tsvg input.dot > output.svg
Using SVG allows you to infinitely zoom to be able to read all the text. You can use inkscape to view and edit SVG files.
Use gst-tracelib library to log key pipeline behavior
The gst-tracelib library hooks into gstreamer key functions and logs the function usage behavior. When the application exits it displays some general statistics. Further analysis can be done based on the data written to the log file. Detailed usage description in the source code README file.
- dataflow, messages, queries and events
- caps set and get activity
- pipeline topology changes
- resource usage
export GSTTL_HIDE="caps;chk;topo" export GSTTL_LOG_SIZE=1048576 AV_FILE=/SD/content/MoMen-dm365.mov LD_PRELOAD=/usr/lib/gst-tracelib/libgsttracelib.so gst-launch filesrc location = $AV_FILE ! qtdemux name=demux ! queue ! dmaidec_h264 numOutputBufs=12 ! \ priority nice=-10 ! queue ! priority nice=-10 ! dmaiperf ! TIDmaiVideoSink accelFrameCopy=true \ videoOutput=DVI videoStd=720P_60 demux.audio_00 ! queue ! priority nice=-5 ! dmaidec_aac ! alsasink
By default, the log file is named /tmp/gsttl.log. Copy the log file somewhere on your host PC where the scripts gsttl_splitlog.py and gsttl_plt.sh are available (or include the path where the scripts are on $PATH). You can use the following command to create a png file with the information gathered by gst-tracelib:
rm -rf gsttl gsttl.png; gsttl_splitlog.py; gsttl_plot.sh | gnuplot; evince gsttl.png
Watch system interrupts
Run the telnet daemon on the target - likely:
/etc/init.d/inetd start # or inetd
telnet into the target hardware and watch the interrupt count
while sleep 1 ; do cat /proc/interrupts ; done
If the pipeline is suppose to be running, the changes in the interrupt count may provide clues as to what is going on.
Capturing a core dump
If your GStreamer application is crashing with a seg fault or similar condition, enable saving a core dump before running the application.
ulimit -c 10000 mkdir -m 777 /root/dumps echo "/root/dumps/%e.core" > /proc/sys/kernel/core_pattern
Once you have a /root/dumps/*.core file, copy it to your workstation and and inspect it with
ddd -debugger arm-linux-gnueabi-gdb $GSTREAMER_APPLIATION
Then in gdb,
target core <core file> bt
and see what function caused the core dump.
Building application and GStreamer libraries with debug enabled
Before using gdb, you need to build your application and the GStreamer libraries with debug enabled.
- Make sure you optimization level is not set to -O3
- Add -ggdb to your CFLAGS
- Remove -fomit-frame-pointer flag from CFLAGS
For TI DMAI plugin and other gstreamer components in the RidgeRun SDK, the can be done via:
cd $DEVDIR/proprietary/gst-dmai-plugins make clean make APPS_CFLAGS=-ggdb build install
Use gdb to attach to GStreamer application and examine all the threads
Built your application with symbols (-g) and no optimization (-O0). Use GStreamer libraries that are built with symbols.
Attach to your running GStreamer application using gdbserver
ps PID=4512 # set the right value based on your application's PID gdbserver :2345 --attach $PID
Then start the cross compile version of the GNU debugger, like
arm-linux-gnueabi-gdb -tui $GSTREAMER_APP
and get the gdb debugger connected to gdb server on the target
set solib-absolute-prefix <path to devdir>/fs/fs file <path to file> target remote <ip address>:2345
List the threads and do a back trace on each one
info threads bt thread 2 bt
Use gdbserver in multi-process mode to attach to GStreamer application and examine all the threads
If the info threads command only shows one thread running, you can see if using gstserver in multi-process mode will work better.
On the target (assuming your program name is gstd), first list the PIDs for all the threads in your application:
PROG=gstd ls /proc/`pidof $PROG`/task
Then start gdbserver in multi-process mode:
gdbserver --multi :2345
Then on the host start the cross compile version of the GNU debugger, like
arm-linux-gnueabi-gdb -tui $GSTREAMER_APP
and get the gdb debugger connected to gdb server on the target
set solib-absolute-prefix <path to devdir>/fs/fs file <path to file> target extended-remote <ip address>:2345
Now you can attach to a specific PID, do a backtrack, snoop around, and detach from that PID. Once detach, you can do the same thing again using a different PID from the output of the ls command above.
attach <pid> bt detach
Once you detach, the PID will run normally.
Exit locked GStreamer application and see what works stand alone
If you have a GStreamer application that locks up and doesn't run correctly even after you exit the program (possibly with cntl-C) and restart, then it is possible some kernel provided resource is the culprit. For example, if you are using a defective ALSA audio out driver, you might find the GStreamer pipeline locks up in the middle. If you exit the GStreamer application and try a simple audio application, like aplay, you might be able to identify the source of your problem.
Examine a history of what transpired just prior to lockup
If you have a GStreamer application that locks up and can change the pipeline to include a (at this point mythical) recent activity history logger. Such a logger element could be put anywhere in the pipeline. The logger would have circular buffers to keep track of all potentially interesting recent history, such as pad activity, bus activity, and any other relevant information. The circular buffer entries would all be timestamped. When some event occurs (a file exists, a message/signal is received, etc), the element would dump the history, and continue capturing new data.
This idea is after the pipeline locks up, you could cause the history logger to dump it data, and then get an idea of what is suppose to be happening that isn't not occurring.
Activate verbose from application
Some elements (like fakesink and identity) print out information when you enable the verbose output (running the pipeline through gst-launch with the option -v), in order to activate this output manually from a C application you can use:
g_signal_connect( pipeline, "deep-notify", G_CALLBACK( gst_object_default_deep_notify ), NULL );
Enable DMAI Davinci Multimedia Application Interface debug output
If you are using the TI DMAI GStreamer plug-in for the TI Davinci and OMAP SoCs, then you can enable DMAI debug to get more information if a hardware accelerated encoder or decoder is throwing an error.
DMAI_DEBUG=2 gst-launch -e alsasrc num-buffers=50 ! 'audio/x-raw-int,rate=(int)44100,channels=(int)2' ! queue ! dmaienc_aac bitrate=128000 ! qtmux ! filesink location=gstaudio_aac_2ch_128k.mp4
Enable CE Codec Engine debug output
If you are using the TI DMAI GStreamer plug-in for the TI Davinci and OMAP SoCs, then you can enable CE debug to get more information if a hardware accelerated encoder or decoder is throwing an error. Note that the stack is GStreamer DMAI plug-in which uses DMAI which uses CE which uses codecs that take advantage of the hardware accelerators.
CE_DEBUG=3 gst-launch -e alsasrc num-buffers=50 ! 'audio/x-raw-int,rate=(int)44100,channels=(int)2' ! queue ! dmaienc_aac bitrate=128000 ! qtmux ! filesink location=gstaudio_aac_2ch_128k.mp4
And you can combine DMAI and CE debug
CE_DEBUG=3 DMAI_DEBUG=2 gst-launch -e alsasrc num-buffers=50 ! 'audio/x-raw-int,rate=(int)44100,channels=(int)2' ! queue ! dmaienc_aac bitrate=128000 ! qtmux ! filesink location=gstaudio_aac_2ch_128k.mp4
To write the debug to a file on tmpfs, which should reduce the performance impact of enabling debug, add
2>&1 | cat - > /tmp/gst-debug.txt
to the end of the gst-launch command.
Use NFS mount to make it easy to transfer files
If you are creating files on a network connected embedded device, you can use NFS mount (which is different than NFS root mount) so that once you generate a file, it is easy to get it on your desktop machine. For any real-time debugging, don't use NFS root mount as the network activity will effect the timing of your pipelines. See Setting Up A NFS Service for instructions on configuring your desktop machine.
HOST_IP=10.0.1.1 HOST_NFS_SHARE_DIR=/local/home/tfischer/work/sdk/fs/fs mkdir /tmp/nfs mount -t nfs -n -o nolock,rsize=1024,wsize=1024 $HOST_IP:$HOST_NFS_SHARE_DIR /tmp/nfs
Now you can copy files between host and target over the network. If you have a file like /tmp/debug-out.txt on the target, you can copy it to the $HOST_NFS_SHARE_DIR on the host using
cp /tmp/debug-out.txt /tmp/nfs
Look for memory leaks
Looking for memory leaks in programs is a very challenging problem. This is even more true for embedded systems, since some of the best memory analysis tools aren't always available, or are difficult to use on this environment.
An easy way to debug the memory leak of a GStreamer program that is running is by looking at the reference counters. Memory leaks on gstreamer happened because reference counters on one or more objects never go back to zero.
GStreamer GST_REFCOUNTING Debug Variable
One of the debug variables is GST_REFCOUNTING. With that, you can get a log of all the refcount changes that happened inside a gst application.
Add the following to your application's command line parameters:
Another option is to set the GStreamer debugging shell variable, then run your program:
Filtering debug output
Here are several examples on how to filter the debug output.
You can see which objects disappear, meaning their reference count reaches zero:
gst-launch --gst-debug=GST_REFCOUNTING:5 videotestsrc num-buffers=3 ! fakesink 2>&1 | grep "\->0" | cut -d' ' -f 19
On useful applications there are always some objects that aren't freed until the program exists. With a simple gst-hello example all the objects are always freed. If you set the pipe to run, lets say 13 times, and in the log you see one object that wasn't freed and the count reached 13 or more, then you have a suspicious object or if you see that the number of non-freed objects increase as you increase the number of runs, then you have an indication of a problem.
Use remote syslog
If your application logs warnings and errors, use remote syslog to make the information more easily available.
If you application generates output, you can redirect it to syslog as follows:
MY_APP=gst-render $MY_APP | logger -t "$MY_APP:"
Generated files debug
There are some tools that can be used for getting information from generated files by gstreamer:
With MP4Box you can use the -info in order to get more info from a generated mp4 file:
MP4Box -info test.mp4
GStreamer Performance Tuning
This is worthy of it own Embedded GStreamer Performance Tuning topic.
GStreamer Application Development
If you can use GStreamer Daemon as your streaming media server, then avoid writing a custom application in the first place!
Developing on host computer
If you have a pipeline where source pads and sink pads are getting connected and disconnected, then you will likely need to write a custom streaming media server. Most people find it easier to build and debug by running the application on the host before trying it on the target hardware.