As per our call today, here (and elsewhere) is the main thing we'd like to change: instead of having ports hardcoded as a range in the code, let's have a separate file mapping knowledge base (KB) names to URLs. (We said ports in the call, but now that I think of it URLs is even more flexible.)
The content of the file could be something like:

$ cat kb-urls.txt
known0 http://localhost:6000/api/1/
known10 http://localhost:6001/api/1/
known20 http://localhost:6002/api/1/
# ignore this line
# ...
known100 http://localhost:6010/api/1/
swhknown http://localhost:6011/api/1/

This way we have a single place where we define all the KBs to test and the associated base URLs.

Having comment support as in the above example would enable easily switching on/off knowledge bases for the experiments.

this is going away too, in favor of a KB-mapping file

subprocess.check_call is the standard way to avoid having to test for exit code by hand.

But, even better, since Python 3.5 you have subprocess.run which supports a check=True kwarg.

See: https://docs.python.org/3/library/subprocess.html#subprocess.run

please drop this (see my other comment about repo info below)

if we have a mapping file, you can also switch to passing that as an argument to this script: gunicorn will start all (and only) the KB backends listed in the file (you just have to strip the "http://" prefix, and what you get will be a good value for gunicorn -b

this will go too, in favor of the mapping file name as input

if you want to simplify this, there's: https://docs.python.org/3/library/collections.html#collections.Counter

c = Counter()
c['api_calls'] += 1
c['queries'] += 1

To keep disk usage manageable, we have cloned all rope with --depth 1, which means you will always obtain one revision in the rev list. So this is pointless.

Also, I've stored repo info on granet already, like this:

$ cat repos-20000/99/13399.info.json 
{
  "origin": "https://github.com/mjsoltysiak/Python",
  "commit": "359a4e613084b053ee6b5ebb696ba0be7c284aa7"
}

So let's drop the usage of PythonGit all together. Just load the info available from the repo info.json file and add them to the tuples you emit.

add a trailing "else" that will fail if you get passed an unexpected value (or at least log it)

alternatively you can switch to a dictionary used as a dispatched from algorithm names to Callables

add to this the repo numerical ID, just in case

It would be nice to compare on some repos the result of len(source_tree) with the number of nodes I've independently counter for each repo.

You can obtain it like this on granet:

$ zstdcat repos-20000/99/01399.nodes.zst | wc -l
281

in theory, len(source_tree) should return the same value for the given repo.

why do you need a dedicated logger?
AFAICT the following should be enough:
logging.basicConfig(filename="benchmark.log")
and you can also customize the logging style with style=...

see: https://docs.python.org/3/library/logging.html#logging.basicConfig

or am I missing something?