https://github.com/lucee/Lucee/commit/b19112959000f532687af764c9e2c602d7647ee0

I like being able to get stats, but I think we should be careful about adding any extra overhead

i.e. being able to query to connection pool, rather than doing it per request

Also, we may wanna upgrade to HttpClient 5.0 before tuning against an older release?

http://hc.apache.org/httpcomponents-client-5.0.x/migration-guide/index.html

but let’s not overload this issue, create new ones

Back on the “root” ticket, i’m trying to find out a few things

So basically, I’m searching if we can get informations / statistics about the connection reuse to “attach” as meta in the response, and even if it’s possible to get some “pool wait time”.

What i’d like to do, is move the HttpClient responsibility entirely inside HTTP4ClientImpl, this would allow to avoid building a client on each request and win some more performances. i’m not sure we’ll be able to keep the exact same signatures though, so not sure if it might impact existing extensions. if they get the builder instead of use the client directly.

One thing positive here would be to be able to reuse clients as the creation of client seems to be a quite extensive job, what i need to find, is how to efficiently allow multiple clients to exists (based on the builder configuration) and yet avoid leaking memory

So, IMHO the next step would be :

• update the Http.java to get the httpClient directly instead of the builder.

• Find if we can expose connection reuse informations attached to the response.

• Prepare structure to allow avoiding creating httpClient on eachRequest. (i suspect this should address the osgi overhead quite a bit)

script would be the same

For the ssl impact it's better as we reuse the connection and maybe the session keys in some case, reducing the number of roundtrip quite aggressively, moving to https was the reason i fixed that at lucee level