Backport of wrk 4.0.0 changes to wrk2

[rrva]

Is below of interest for wrk2? If so, I can spend time cherry-picking individual commits from upstream, rather than this large commit if that is preferred.

Backporting non-statistics related changes which occurred in wrk since wrk2 forked. Primary need was the lua scripting changes making it possible to set each threads ip address from lua.

Upstream scripting changes

• eb165ce prepare wrk 4.0.0 release
• 57f3b33 simplify script state setup
• 9b84d3e add script setup() and thread methods
• 6f0aa32 move address resolution into lua
• 93348e2 add CHANGES file and simplify script init

HTTP parser

• 6c15549 update http parser to 2.4.2
• 522ec60 pass EOF through to parser

[rrva]

One use-case is following redirects, using a lua script like this:

url = require "url"
orig_address = wrk.lookup(wrk.host, wrk.port or "http")[1]
wrk.headers["Connection"] = "close"

response = function(status, headers, body)
   if status == 302 or status == 301 then
        if redirect_address == nil then
            location = url.parse(headers["Location"])
            redirect_address = wrk.lookup(location.host, location.port or "http")[1]
            print(("Will redirect to: %s"):format(redirect_address))
        end
       wrk.thread.addr = redirect_address
   else
       wrk.thread.addr = orig_address
   end
end

[ahothan]

Are you still working on this? I was about to consider doing the same (backport to wrk 4.x) until I see you did the work already.

[rrva]

I'm not working on it anymore. Feel free to use these changes and continue if you want. I needed the lua scripting changes and the constant rate load generation option.

[ahothan]

allright - thx

[rrva]

Out of curiosity, I saw your go-wrk project. Why would you prefer this one over that? Because of better stats reporting? Higher max load? More exact request rate?

[ahothan]

We're actually using and evaluating giltene/wrk2 right now. If it works sufficiently for us we will stick with it. The go/wrk was just for experimenting and was done before I found wrk2. What is amazing is the number of small forks and each one has its pros/cons (pick your poison). A pity nobody could really unifies every feature back (like backporting wrk2 work to wrk). go-wrk looks nice (smaller code and maybe faster) but not too many go developers right now and does not have the -R feature which I need. For now all we need is a generator that can sustain 10K connections at about 1 rps per connection (on a relatively fast Intel Xeon CPU). I have no idea if that can work but we will see soon. We are lining up a large number of wrk2 generators to load a large number of HTTP servers for an OpenStack scaling app. We can't use wrk unthrottled traffic because that does not represent live http traffic and that overwhelms any form of scale testing very quickly.

[giltene]

I initially didn't want to move the base much, since wrk2 started as a demonstration (of applying correct measurement to a load generator). but it seems to be used by a bunch of people, so I figured updating it makes sense, and the changes here don't seem to interfere with the measurement logic or the throughput throttling. So thanks for the back port work...