These tests verify that PyMongo works with Apache and mod_wsgi. They are primarily intended to prevent regression of PYTHON-353, a connection leak when PyMongo 2.2 was used with Python 2.6 and mod_wsgi 2.8. However, the test may also catch concurrency bugs, or incompatibilities between PyMongo's C extensions and the way mod_wsgi manages Python sub interpreters. It is generally useful to test PyMongo in the unconventional environment that mod_wsgi creates.
PyMongo should be tested with several versions of mod_wsgi and a selection
of Python versions. Each combination of mod_wsgi and Python version should
be tested with a standalone and a replica set. mod_wsgi_test.wsgi
detects if the deployment is a replica set and connects to the whole set.
We need a Python interpreter built as a shared library. Download the source tarball for each Python version tested, untar it, and run:
./configure --prefix=/some/directory --enable-shared make make install
This results in an executable named "python" and a shared library named something like "libpython2.7.so.1.0". It may be necessary to add /some/directory/lib to your system's LD_LIBRARY_PATH, or to make a symlink from your system's default library directory to the shared library. For example, on Ubuntu:
ln -s /some/directory/lib/libpython2.7.so.1.0 /usr/lib64/
Compile mod_wsgi for each combination for Python and mod_wsgi version in the test matrix. For example, to compile mod_wsgi 3.4 for Python 2.6 on a RedHat-like Linux:
sudo yum install -y httpd httpd-devel wget https://modwsgi.googlecode.com/files/mod_wsgi-3.4.tar.gz tar xzf mod_wsgi-3.4.tar.gz cd mod_wsgi-3.4 ./configure --with-python=/some/directory/python make make install
To ease testing of several matrix combinations, copy the resulting
mod_wsgi.so to a safe place.
Start a standalone listening on port 27017, or a replica set with a member listening on port 27017.
Copy the appropriate version of mod_wsgi.so into Apache's modules
directory. Start Apache with the mod_wsgi_test.conf in this directory.
Run the included test_client.py script:
python test/mod_wsgi_test/test_client.py -n 2500 -t 100 parallel \
http://localhost/${WORKSPACE}
...where the "n" argument is the total number of requests to make to Apache,
and "t" specifies the number of threads. WORKSPACE is the location of
the PyMongo checkout.
Run this script again with different arguments to make serial requests:
python test/mod_wsgi_test/test_client.py -n 25000 serial \
http://localhost/${WORKSPACE}
The test_client.py script merely makes HTTP requests to Apache. Its
exit code is non-zero if any of its requests fails, for example with an
HTTP 500.
The core of the test is in the WSGI script, ``mod_wsgi_test.wsgi`. This script inserts some documents into MongoDB at startup, then queries documents for each HTTP request.
If PyMongo is leaking connections and "n" is much greater than the ulimit, the test will fail when PyMongo exhausts its file descriptors.
At MongoDB, Inc. we use a Jenkins job that tests each combination in the
matrix. The job copies the appropriate version of mod_wsgi.so into
place, sets up Apache, starts a single server or replica set,
and runs test_client.py with the proper arguments.