add sample directory size exporter

[anarcat]

This is a possible workaround for the lack of metrics in the new
storage backend, as documented in:

prometheus/prometheus#3684

Partly inspired by this post as well:

https://www.robustperception.io/monitoring-directory-sizes-with-the-textfile-collector/

[SuperQ]

Thanks. One note, textfile scripts follow the standard of sending the data to STDOUT.

[anarcat]

i followed the example in https://www.robustperception.io/monitoring-directory-sizes-with-the-textfile-collector/, which ensures atomicity... should i change this? i don't mind :)

[SuperQ]

Yes, for the examples here, we decided to go with simple STDOUT as our interface method. This way the script operators can deal with the atomicity themselves.

It also makes it easier to test out scripts by running them manually, so they don't automatically update metrics.

[anarcat]

alright, how does 541c05e look instead then?

[anarcat]

hmm... i can't review the build failure on buildkite as i don't have an account. can anyone enlighten me?

[SuperQ]

The buildkite tests are broken, don't worry about them.

[SuperQ]

I would suggest adding s/"/\\"/g ; to the beginning of the sed expression in order to escape any " in the path.

[anarcat]

hmm... is that the only character we would need to escape? this path is provided by the admin... they would be shooting themselves in the foot by submitting such a nasty path...

[SuperQ]

Yes, double quote is the only thing you need to escape.

[brian-brazil]

Backslash also needs escaping in label values.

[anarcat]

so how about adding s/"/\\"/g;s/\\/\\\\/g? does that cover the escaping correctly?

what a mess shell is :p

[SuperQ]

@anarcat Yes, except possibly \\ before \"

[anarcat]

@SuperQ not sure that's necessary:

$ du --block-size=1 --summarize foo\"bar\"baz\\back/| sed -ne 's/\\/\\\\/;s/"/\\"/g;s/^\([0-9]\+\)\t\(.*\)$/node_directory_size_bytes{directory="\2"} \1/p'
node_directory_size_bytes{directory="foo\"bar\"baz\\back/"} 4096

my original sed was backwards, so it was escaping the backslashes from the previous ones...

i've added a commit to properly escape stuff, see if it fits for you... too bad there's no unit tests for this... another reason why this would be a good addition in the core...

[SuperQ]

LGTM