Andrew Halberstadt [:ahal] Assignee
	Description • 9 years ago

People have profiled ujson [1] being 400-600% faster than json. I did an unscientific test, and without ujson, we spend ~5 seconds in json.dumps during a build. With ujson, we only spend ~0.4s.

Seems worth using if we can. Also seems like it might be worth checking into the tree.

[1] https://github.com/esnme/ultrajson

	Andrew Halberstadt [:ahal] Assignee
	Comment 1 • 9 years ago

Note: ujson gets some of it's performance boosts by doing no validation on the validity of data being serialized/deserialized, see [1].

I think this is ok for us, since we already do that in logtypes.py. The stdlib json module is probably doing unnecessary validation for us atm.

[1] http://mod16.org/hurfdurf/?p=266

	Armen [:armenzg]
	Comment 2 • 9 years ago

Also, for pulse_actions we switched away from json for memory issues to ijson:
http://explique.me/Ijson

	James Graham [:jgraham]
	Comment 3 • 9 years ago

I'm OK with using ujson since we pre-validate the data anyway. It might be nice to know if that would also work for the pulse usecase so that we can avoid using 3 different json libraries in different parts of the tree.

	Andrew Halberstadt [:ahal] Assignee
	Comment 4 • 9 years ago

Bug 1193522 - [mozlog] Use ultrajson instead of stdlib json if available for performance, r=jgraham

	James Graham [:jgraham]
	Comment 5 • 9 years ago

Comment on attachment 8647617 [details]
MozReview Request: Bug 1193522 - [mozlog] Use ultrajson instead of stdlib json if available for performance, r=jgraham

https://reviewboard.mozilla.org/r/16023/#review14295

Ship It!

	Pulsebot
	Comment 6 • 9 years ago

https://hg.mozilla.org/integration/mozilla-inbound/rev/aa9cc17290fc

	Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)
	Comment 7 • 9 years ago

https://hg.mozilla.org/mozilla-central/rev/aa9cc17290fc