OpenResty lua split benchmarkを自分でも試してみた

README

## オリジナル
https://github.com/toritori0318/Dockerfiles/tree/b82cb44834e33f0777b42d5f9b799b230c3e3237/openresty-split-bench
http://toritori0318.hatenadiary.jp/entry/20191201/1575141320

## 改変
/lua/lua_split_f は私が追加しました。固定数でないと使えないので split と同列ではないのですが、参考として。

## Run Server

```
docker build -t local/openresty-split-bench .
docker run -it -p 89:80 local/openresty-split-bench
```

## Benchmark

```
# curl
curl 'http://localhost:89/lua/lua_split_a?c=10'
```

Dockerfile

FROM openresty/openresty:1.15.8.1-3-alpine-fat

ADD nginx.conf /usr/local/openresty/nginx/conf/nginx.conf

nginx.conf

#daemon off;
worker_processes  1;
error_log /dev/stdout warn;
events {
    worker_connections 1024;
}

pcre_jit on;

http {
  init_by_lua_block {
    function clock_time(title, block)
      local st = os.clock()
      block()
      local ed = os.clock()

      ngx.say(title .. ": " .. ed-st.. " sec")
    end
  }


  server {
    listen 80;

    location / {
      index index.html;
    }
    location /lua {
      default_type text/html;
      content_by_lua_block {
        ngx.say("hello, lua")
      }
    }

    location = /lua/ngx_re_split {
      default_type text/html;
      content_by_lua_block {
        local ngx_re = require "ngx.re"
        -- ngx_re.opt("jit_stack_size", 128 * 1024)

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("ngx_re_split", function()
          for i = 1, c do
            local t, err = ngx_re.split(d, ",")
            if err then
              ngx.log(ngx.ERR, string.format("err: %s", err))
            else
              -- ngx.log(ngx.WARN, string.format("t: %d", #t))
            end
          end
        end)
      }
    }

    location = /lua/lua_split_a {
      default_type text/html;
      content_by_lua_block {
        local function csplit(str,sep)
           local ret={}
           local n=1
           for w in str:gmatch("([^"..sep.."]*)") do
              ret[n] = ret[n] or w -- only set once (so the blank after a string is ignored)
              if w=="" then
                 n = n + 1
              end -- step forwards on a blank but not a string
           end
           return ret
        end

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_a", function()
          for i = 1, c do
            local t = csplit(d, ",")
            -- ngx.log(ngx.WARN, string.format("t: %d", #t))
          end
        end)
      }
    }

    location = /lua/lua_split_b {
      default_type text/html;
      content_by_lua_block {
        local function gsplit(s,sep)
           local i, done, g = 1, false, s:gmatch('(.-)'..sep..'()')
           local function pass(...)
              if ... == nil then
                 done = true
                 return s:sub(i)
              end
              i = select(select('#',...),...)
              return ...
           end
           return function()
              if done then
                 return
              end
              if s == '' or sep == '' then
                 done = true
                 return s
              end
              return pass(g())
           end
        end

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_b", function()
          for i = 1, c do
            local t={} for c in gsplit(d, ",") do table.insert(t,c) end
            -- ngx.log(ngx.WARN, string.format("t: %d", #t))
          end
        end)
      }
    }

    location = /lua/lua_split_c {
      default_type text/html;
      content_by_lua_block {
        function string.gsplit(s, sep, plain)
           local start = 1
           local done = false
           local function pass(i, j, ...)
              if i then
                 local seg = s:sub(start, i - 1)
                 start = j + 1
                 return seg, ...
              else
                 done = true
                 return s:sub(start)
              end
           end
           return function()
              if done then
                 return
               end
              if sep == '' then
                 done = true
                 return s
              end
              return pass(s:find(sep, start, plain))
           end
        end

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_c", function()
          for i = 1, c do
            local t={} for c in d:gsplit(",") do table.insert(t,c) end
            -- ngx.log(ngx.WARN, string.format("t: %d", #t))
          end
        end)
      }
    }

    location = /lua/lua_split_d {
      default_type text/html;
      content_by_lua_block {
        local strfind = string.find
        local strsub = string.sub
        local tinsert = table.insert
        local function strsplit(delimiter, text, plain)
           plain = plain or false
           local list = {}
           local pos = 1
           if strfind("", delimiter, 1, plain) then -- this would result in endless loops
              error("delimiter matches empty string!")
           end
           while 1 do
              local first, last = strfind(text, delimiter, pos, plain)
              if first then -- found?
                 tinsert(list, strsub(text, pos, first-1))
                 pos = last+1
              else
                 tinsert(list, strsub(text, pos))
                 break
              end
           end
           return list
        end

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_d", function()
          for i = 1, c do
            local t = strsplit(",", d)
            -- ngx.log(ngx.WARN, string.format("t: %d", #t))
          end
        end)
      }
    }

    location = /lua/lua_split_e {
      default_type text/html;
      content_by_lua_block {
        local function split(str, pat)
           local t = {}  -- NOTE: use {n = 0} in Lua-5.0
           local fpat = "(.-)" .. pat
           local last_end = 1
           local s, e, cap = str:find(fpat, 1)
           while s do
              if s ~= 1 or cap ~= "" then
                 table.insert(t,cap)
              end
              last_end = e+1
              s, e, cap = str:find(fpat, last_end)
           end
           if last_end <= #str then
              cap = str:sub(last_end)
              table.insert(t, cap)
           end
           return t
        end

        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_e", function()
          for i = 1, c do
            local t = split(d, ",")
            -- ngx.log(ngx.WARN, string.format("t: %d", #t))
          end
        end)
      }
    }

    location = /lua/lua_split_f {
      default_type text/html;
      content_by_lua_block {
        local params = ngx.req.get_uri_args()
        local c = params["c"] or 1
        local d = "aaaaaaa,bbbbbbbbbbb,ccccccccccccc,ddddddddddddddd,eeeeeeeeeeeeee,f,ggggg,hhh,iii,jjjjjjj,k"

        clock_time("lua_split_e", function()
          for i = 1, c do
            local t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11 = string.match(d, '([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*),([^,]*)')
            -- ngx.log(ngx.WARN, string.format("t11: %s", t11))
          end
        end)
      }
    }

  }
}

result.txt

$ curl 'http://localhost:89/lua/lua_split_f?c=10'
lua_split_e: 0.000276 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10'
lua_split_e: 7.2e-05 sec
$ curl 'http://localhost:89/lua/ngx_re_split?c=10'
ngx_re_split: 0.000731 sec
$ curl 'http://localhost:89/lua/lua_split_a?c=10'
lua_split_a: 0.000443 sec
$ curl 'http://localhost:89/lua/lua_split_b?c=10'
lua_split_b: 0.000394 sec
$ curl 'http://localhost:89/lua/lua_split_c?c=10'
lua_split_c: 0.000434 sec
$ curl 'http://localhost:89/lua/lua_split_d?c=10'
lua_split_d: 0.00029 sec
$ curl 'http://localhost:89/lua/lua_split_e?c=10'
lua_split_e: 0.000162 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10'
lua_split_e: 5.4e-05 sec
$ curl 'http://localhost:89/lua/ngx_re_split?c=10'
ngx_re_split: 0.000418 sec
$ curl 'http://localhost:89/lua/lua_split_a?c=10'
lua_split_a: 0.000256 sec
$ curl 'http://localhost:89/lua/lua_split_b?c=10'
lua_split_b: 0.000174 sec
$ curl 'http://localhost:89/lua/lua_split_c?c=10'
lua_split_c: 0.000317 sec
$ curl 'http://localhost:89/lua/lua_split_e?c=10'
lua_split_e: 0.000196 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10'
lua_split_e: 4.3999999999999e-05 sec
$ curl 'http://localhost:89/lua/ngx_re_split?c=10000000'
ngx_re_split: 28.056177 sec
$ curl 'http://localhost:89/lua/lua_split_a?c=10000000'
lua_split_a: 41.272921 sec
$ curl 'http://localhost:89/lua/lua_split_b?c=10000000'
lua_split_b: 61.272173 sec
$ curl 'http://localhost:89/lua/lua_split_c?c=10000000'
lua_split_c: 19.150326 sec
$ curl 'http://localhost:89/lua/lua_split_d?c=10000000'
lua_split_d: 11.329389 sec
$ curl 'http://localhost:89/lua/lua_split_e?c=10000000'
lua_split_e: 40.524021 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10000000'
lua_split_e: 13.989293 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10000000'
lua_split_e: 14.0411 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10'
lua_split_e: 5.7999999995673e-05 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=100'
lua_split_e: 0.00018199999999524 sec
$ curl 'http://localhost:89/lua/ngx_re_split?c=100'
ngx_re_split: 0.00036599999998543 sec
$ curl 'http://localhost:89/lua/lua_split_d?c=100'
lua_split_d: 0.00029900000001248 sec
$ curl 'http://localhost:89/lua/lua_split_a?c=100'
lua_split_a: 0.0012230000000102 sec
$ curl 'http://localhost:89/lua/lua_split_b?c=100'
lua_split_b: 0.00065299999999979 sec
$ curl 'http://localhost:89/lua/lua_split_c?c=100'
lua_split_c: 0.00089599999998313 sec
$ curl 'http://localhost:89/lua/lua_split_d?c=100'
lua_split_d: 0.00037800000001198 sec
$ curl 'http://localhost:89/lua/lua_split_e?c=100'
lua_split_e: 0.00061199999998962 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=100'
lua_split_e: 0.00015500000000657 sec
$ curl 'http://localhost:89/lua/ngx_re_split?c=10000'
ngx_re_split: 0.035384999999991 sec
$ curl 'http://localhost:89/lua/lua_split_a?c=10000'
lua_split_a: 0.048181999999997 sec
$ curl 'http://localhost:89/lua/lua_split_b?c=10000'
lua_split_b: 0.069175999999999 sec
$ curl 'http://localhost:89/lua/lua_split_c?c=10000'
lua_split_c: 0.02699100000001 sec
$ curl 'http://localhost:89/lua/lua_split_d?c=10000'
lua_split_d: 0.020365999999996 sec
$ curl 'http://localhost:89/lua/lua_split_e?c=10000'
lua_split_e: 0.04945699999999 sec
$ curl 'http://localhost:89/lua/lua_split_f?c=10000'
lua_split_e: 0.022514999999999 sec

READMEにも書いてますがオリジナルは
http://toritori0318.hatenadiary.jp/entry/20191201/1575141320
です。
/lua/lua_split_f は私が追加しました。固定数でないと使えないので split と同列ではないのですが、参考として。

試してみてわかったのですが回数によって順位が変動するんですね。実行した順に関係するかもしれないので並び変えずそのままにしています。
私は Core i5 に換装した Express5800 で試しました。
string.match (/lua/lua_split_f) 方式は10回だと最速でしたが、10000000 回だと他のLua実装のほうが速いことがわかりました。