Comment 4 for bug 1737989

when we originally switched to it in gnocchi[1], a former colleague and i were benchmarking its performance but i seemed to have removed my benchmarks when i updated my computer. i would say http://artem.krylysov.com/blog/2015/09/29/benchmark-python-json-libraries/ is an accurate representation. for our use case, we got more than than the 2x-10x improvement.

i would argue we probably don't need ujson in aodh (we merged it recently so i imagine we could just revert that if required).

for ceilometer, we use it for gnocchi publisher and http publisher.
- the gnocchi publisher we arguably don't need it as it's only used in the event workflow (which is less frequent).
- the http publisher, well it's used for everything. so basically it's a matter of do we care to slow down http publisher that in theory isn't broken. i don't use the http publisher so i'm indifferent and i don't know how much of the workflow is serialisation time.

tl;dr if there is a patch to remove ujson from aodh, i'll +A. for ceilometer, i'm indifferent and will +A depending on feedback.

[1] https://github.com/gnocchixyz/gnocchi/commit/e7d8fafa77b0cef20b653020425c15be76e389b5