I am writing a program that uses several instances of the same code in production. Instances then communicate over network, but that is irrelevant here.
During development, I fork the initial process multiple times to get a complete setup of instances, which is very convenient.
However, trying to write integration tests on this gives me problems: The test framework (unittest/nosetests/py.test) is of course also forked, so that I end up with multiple copies test framework, providing garbled results as output of each child's unittest overlaps.
A simplified example:
To achieve this, I'd need to "unload" or stop the test frameworks from the child processes. How can I do this?
I could fix this by invoking the test suite only after forking off the children, but I need to re-start all children for every test.
Forking a process in this context is a big hassle, as the child processes inherit the full stack of the parent, including the running instance of the test framework.
There are essentially two options to solve this:
Number one is to fork before initializing the test framework, and then access the child processes from the tests. This works well if child processes are long-lived, but proves to be a hassle if child processes are to be started per test, maybe even with different test fixtures/scenarios.
Number two is to forgo forking all together by simply starting the processes by using subprocess / Popen. I found a pytest plugin called xprocess which helps you with this task. It didn't provide exactly what I needed, but it pointed me in the right direction.