Performance problem (clj-json could be nearly 10x faster) about clj-json HOT 5 CLOSED

Jan,
Thanks for the heads up! That's a recent addition, I had no idea it was that detrimental to performance. clj-json is intended to be highly performant, I'll make fixing this a high priority.
I suspect it's the postwalk.
Thanks again!
-lance

On Monday, September 19, 2011 at 12:48 AM, Jan Rychter wrote:

Hi,

Since clojure.data.json is a big performance problem in our app (nearly 30% of request servicing time), I decided to try clj-json. Unfortunately, it isn't much of an improvement, and all because of a tiny issue. On our data, the "coerce" stage constitutes over 91% of cpu time spent in clj-json!

Since we actually do not need any coercing at all, I tested a variant of generate-string with the call to coerce removed. We get excellent performance that way.

Oh, and before you ask - this is a search engine and every millisecond counts. We really do care about performance.

Please provide a fast variant of generate-string for those of us who do not need any fancy conversions. I would submit a pull request, but it's a really simple fix, I can think of several ways this can be implemented and I think it is up to the library author to decide.

thanks,
--J.

Reply to this email directly or view it on GitHub:
#11

from clj-json.

Yes, it's the postwalk — and in our case we'd like to have a way to short-circuit it completely. When generating JSON replies we need to be as fast as possible and we only use a subset of JSON anyway.

Thanks and I'll be waiting for a clojars release so that we can use it!

from clj-json.

Hi Jan,
I just released version 0.4.1 to clojars, removing the post walk. Let me know how it performs for you.
-lance

On Monday, September 19, 2011 at 12:20 PM, Jan Rychter wrote:

Yes, it's the postwalk — and in our case we'd like to have a way to short-circuit it completely. When generating JSON replies we need to be as fast as possible and we only use a subset of JSON anyway.

Thanks and I'll be waiting for a clojars release so that we can use it!

Reply to this email directly or view it on GitHub:
#11 (comment)

from clj-json.

Hi,

0.4.1 works great. I haven't tried any coercions, though — so I can't comment on how those perform. But in our case we only need the fast path and that works great now.

Thanks for the quick response!

--J.

from clj-json.

Great to hear! I also pushed 0.4.2 a little later with performant support for anything that implements the Iterator interface. That probably eliminates most of the need for coercions.

Thanks again for the notice,
lance

On Tuesday, September 20, 2011 at 12:37 AM, Jan Rychter wrote:

Hi,

0.4.1 works great. I haven't tried any coercions, though — so I can't comment on how those perform. But in our case we only need the fast path and that works great now.

Thanks for the quick response!

--J.

Reply to this email directly or view it on GitHub:
#11 (comment)

from clj-json.