Task Performance Analysis¶
The Python standard library provides the cProfile module to assist developers in collecting statistics that describes how often and for how long which parts of the program executed.
Enabling cProfiling may cause a reduction in performance. It is not recommended for production unless enabled temporarily.
There is built in support for profiling individual tasks. This can be enabled by editing the server.conf file:
[profiling] enabled: true
This will, by default, put task profile output into /var/lib/pulp/c_profiles named per task ID. You can modify the location these profiles are stored:
[profiling] enabled: true directory: /var/www/html/pub
apache user must be able to write to the path specified by
Custom Runtime Performance Analysis¶
Custom code locations be analyzed as well. To capture information about all function calls, add the following lines of code before the code you are interested in analyzing:
import cProfile pr = cProfile.Profile() pr.enable()
After the code that is of interest, add the following two lines:
The file should be written to somewhere the application has write access. /tmp does not seem to work for this use case. /var/lib/pulp/ is guaranteed to provide Pulp write access.
Once the captured statistics are written to a file, the file can be examined in two ways. Using pstats module in the Python standard library or cprofilev. The following command will open the file using pstats:
$ python -m pstats /var/lib/pulp/profile_stats_dump
Once in pstats, view the list of available commands by running help. The stats command outputs the list of all calls made while profiler was enabled. The sort command is used to sort the data by different metrics.
cprofilev provides a web UI for interacting with the data. You can examine the set of statistics by installing cprofilev from PyPi and running it:
$ sudo pip install cprofilev $ cprofilev -f ~/devel/profile_stats_dump
At this point, browse to http://localhost:4000/ to view profiling information.
In addition, gprof2dot can convert the profiling output into a dot graph. Make sure graphviz is installed by yum or dnf before using this tool. Then you can get statistics graph by installing gprof2dot from PyPI and running it:
$ sudo pip install gprof2dot $ gprof2dot -f pstats ~/devel/profile_stats_dump | dot -Tpng -o output.png