Add CLI to plot memory monitors

[knelli2]

Proposed changes

Examples of output:

no_color.pdf

with_color.pdf

Upgrade instructions

Code review checklist

• The code is documented and the documentation renders correctly. Run
  make doc to generate the documentation locally into BUILD_DIR/docs/html.
  Then open index.html.
• The code follows the stylistic and code quality guidelines listed in the
  code review guide.
• The PR lists upgrade instructions and is labeled bugfix or
  new feature if appropriate.

Further comments

[knelli2]

For some reason, the command itself takes a very long time to run. I wasn't able to determine why, even with spectre --profile. Any suggestions of where I'm going wrong are welcome.

[nilsdeppe]

I'll let @nilsvu review the actual code, but I have a feature request: would it be easy to add a flag like --top-n 5 to show only the 5 most significant users? On the plots you showed it was tricky to figure out from the legend which were the 3 big ones shown.

[nilsvu]

Good idea, and/or how about a stacked area plot like this: https://python-graph-gallery.com/stacked-area-plot/
Then only show the legend entry for the big ones (above 5% or something like that)

[nilsdeppe]

I'd be fine with adding that as an option. I do find it easier in the current style to read off how many GB something is using. I think the stacked plot makes relative comparisons nice, though. So maybe a CLI flag to enable the stacking? E.g. --stacked-plot?

[knelli2]

I'm personally not a big fan of the stacked plot. It's a bit ambiguous to me whether to interpret a stack as relative to zero, or relative to the stack below it. I do like the --top-n option though. I'll add that.

[nilsvu]

True. Maybe just combine the entries with only a couple percent into a single entry, then you don't need an extra CLI option and you can't see those small entries in the plot anyway. Also sort the entries.

[knelli2]

Examples of output now with fixup

[nilsdeppe]

LGTM, once @nilsvu is happy with the python code, please go ahead and merge. No need for my approval.

[nilsvu]

I'd remove this. Either make it always GB or switch to MB if the total memory is below 1 GB. This is just another option the person using this has to learn and set.

[knelli2]

I'd prefer to keep this one. I did make the default GB though since that's more likely

[nilsvu]

Ok

[nilsvu]

Some simplifications with pandas. You can squash everything.

[nilsvu]

Is the line width needed because the time sampling is so dense? Generally better to leave style up to the stylesheet.

[knelli2]

I did this so the total line would stand out in comparison to the other lines

[nilsvu]

Ah ok, maybe the color is enough? Up to you.