Debugging, profiling and tests¶
Sympathy offers a few tools that will help you fix problems in your nodes, notably interactive debugging and profiling.
When a node isn’t working as expected a very handy tool to use is the node debugger. Run a workflow up and to the node that you want to debug. Right click the node and choose “Debug” from the context menu.
This will bring up Spyder with the node with correct data on the input ports,
ready to be debugged simply by setting a breakpoint and pressing “play”. After
running the code at least once you will also have access to the node’s
node_context in the interactive python prompt under the name
for debug node context). See Node context reference for information on how to
Please refer to the Spyder manual for more info on it’s debugging features.
Profiling nodes and workflows¶
If your node or workflow is running too slow you can run the profiler on it to see what parts are taking the most time. If you have configured Graphviz, see Configuring Graphviz, you will also get a call graph.
To profile a single node simply right click on a node that can be executed and choose Profile. This will execute the node and any nodes before it that need to be executed, but only the node for which you chose Profile will be included in the profiling. To profile an entire workflow go to the Controls menu and choose Profile flow. This will execute all nodes in the workflow just as via the Execute flow command. After either execution a report of the profiling is presented in the Error view. Profiling of single subflows is similar to profiling of single nodes but include all the executable nodes in the subflow.
The profile report consists of a part called Profile report files and a part called Profile report summary.
Profile report files¶
The Profile report files part of the profile report consists of two or three file paths. There is always a path to a txt file and a stats file, and also a pdf file if Graphviz is configured, see Configuring Graphviz. The txt file is a more verbose version of the summary but with full path names and without any limit on the number of rows. The pdf file contains a visualization of the information in the summary, also called a call graph.
The call graph contains a node for each function that has been called in the profiled code. The color of the node gives you a hint about how much of the total time was spent inside a specific function. Blue nodes represent functions with low total running time and red nodes represent functions with high total running time. The nodes have the following layout:
First row is the name of the function. Second row is the percentage of the running time spent in this function and all its children. Third row (in parentheses) is the percentage of the running time spent in this function alone. The forth row is the total number of times this function was called (including recursive calls).
The edges of the graph represent the calls between functions and the label at an edge tells you the percentage of the running time transferred from the children to this parent (if available). The second row of the label tells you the number of calls the parent function called the children.
Please note that the total running time of a function has to exceed a certain cut-off to be added to the call graph. So some very fast workflows can produce almost empty call graphs.
A third file will also always be provided with the file ending ”.stats”. This file contains all the statistics that was used to create the summary and the call graph. To begin digging through this file open a python interpreter and write:
>>> import pstats >>> s = pstats.Stats('/path/to/file.stats') >>> s.print_stats()
For more information look at the documentation for the Stats class.
Profile report summary¶
The summary contains a row for each function that has been called in the profiled code. Several calls to the same function are gathered into a single row. The first column tells you the number of times a function has been called. The next four columns measure the time that it took to run a specific function. In the last column you can see what function the row is about. See https://docs.python.org/2/library/profile.html for details on how to interpret this table.
The summary also includes up to 10 node names from nodes included in the profiling and an indication of the number of nodes that were ommited to save space.
Writing tests for your nodes¶
As with any other code, writing tests for your nodes is a good way of assuring that the nodes work and continue to work as you expect.
Let’s start by running the following command from a terminal or command prompt:
python launch.py tests
This will run an extensive test suite on the sympathy platform and on all configured libraries. It tests that the documentation for all nodes can be generated without any errors or warnings and that the configuration guis for all nodes can be created. But it doesn’t run the node.
The easiest way to test the execution of your nodes is to add them to a workflow and put that workflow in <library path>/Test/Workflow/. All workflows in that folder and subfolders are automatically run when running the above command.
Look in <sympathy folder>/Library/Test/Workflow/ for examples of such test workflows.
It is also a good idea to write unit tests to ensure the quality of your
modules. Put unit test scripts in <library path>/Test/Unit/. If the tests are
named correctly they will automatically be found by the python module
Which is run as a part of
launch.py tests. See
https://nose.readthedocs.org/en/latest/finding_tests.html for more details
about how to name your unit tests.
For example a unit test script that tests the two functions
bar() in the module
boblib.bobutils could be called
test_bobutils.py and look something like this:
import numpy as np from nose.tools import assert_raises import boblib.bobutils def test_foo(): """Test bobutils.foo.""" assert boblib.bobutils.foo(1) == 2 assert boblib.bobutils.foo(0) == None with assert_raises(ValueError): boblib.bobutils.foo(-1) def test_bar(): """Test bobutils.bar.""" input = np.array([True, False, True]) expected = np.array([False, False, True]) output = boblib.bobutils.bar(input) assert all(output == expected)
For more examples of real unit tests take a look at the scripts in <sympathy
folder>/Library/Test/Unit/ or have a look at the documentation for the
nose module at https://nose.readthedocs.org/.
You can run only the unit tests of your own library by running the following command from a terminal or command prompt:
python launch.py tests <library path>/Test/Unit