Incorrect metric values for multi processes

[sallyruthstruik]

Looks like metric you send doesn't consider suprocesses.
For example, I run gunicorn with 8 workers:

gunicorn wsgi:app -w 8

and write in stderr your rss:absolute metric and sum of children values:

logging.error("{}-{}".format(
        memory_info.rss,
        sum(p.memory_info().rss for p in process.children())
    ))

And I got

20766720-1070051328

So master process take ~20Mb but with children processes it takes about 1Gb of memory.

I think better solution is to allow metrics to take an account of their child processes. But in this way it is unclear how to calculate percentage values.

[sallyruthstruik]

Fix example: 787a074