Analysis header values with percentiles

header_value_analysis.lua

wrk.method = "POST"
local f = io.open("batch_4.json", "r")
wrk.body   = f:read("*all")
wrk.headers["Content-Type"] = "application/json"
wrk.headers["Authorization"] = "Bearer <token>"

local threads = {}

function percentile(data, p)
    local sortedData = { }
    for k, v in pairs(data) do
            sortedData[k] = v
    end
    table.sort(sortedData)
    return sortedData[math.ceil(#data * p / 100)]
end

function setup(thread)
    thread:set("id", counter)
    table.insert(threads, thread)
end

function init(args)
    count_worker1 = 0
    count_worker2 = 0
    count_idk = 0

    duration_worker1 = {}
    duration_worker2 = {}
    duration_envoy1 ={}
    duration_envoy2 = {}
end

function response(status, headers, body)
    local srv_id = headers["x-ms-worker-id"]
    if status == 200 then
        local duration = tonumber(headers["x-ms-model-execution-duration"]) * 1000
        local duration_envoy = tonumber(headers["x-envoy-upstream-service-time"])

        if srv_id == "31311" then
            count_worker1 = count_worker1 + 1
            duration_worker1[#duration_worker1 + 1] = duration
            duration_envoy1[#duration_envoy1 + 1] = duration_envoy - duration
        elseif srv_id == "31312" then
            count_worker2 = count_worker2 + 1
            duration_worker2[#duration_worker2 + 1] = duration
            duration_envoy2[#duration_envoy2 + 1] = duration_envoy - duration
        else
            count_idk = count_idk + 1
        end
    end
end


function done(summary, latency, requests)
    total_worker1 = 0
    total_worker2 = 0
    total_workeridk = 0
    total_duration1 = {}
    total_duration2 = {}
    total_envoy_duration1 = {}
    total_envoy_duration2 = {}

    total_envoy_duration = {}

    for index, thread in ipairs(threads) do
      total_worker1 = total_worker1 + thread:get("count_worker1")
      total_worker2 = total_worker2 + thread:get("count_worker2")
      total_workeridk = total_workeridk + thread:get("count_idk")
      for _, v in pairs(thread:get("duration_worker1")) do
         table.insert(total_duration1, v)
      end
      for _, v in pairs(thread:get("duration_worker2")) do
         table.insert(total_duration2, v)
      end
      for _, v in pairs(thread:get("duration_envoy1")) do
         table.insert(total_envoy_duration1, v)
         table.insert(total_envoy_duration, v)
      end
      for _, v in pairs(thread:get("duration_envoy2")) do
         table.insert(total_envoy_duration2, v)
         table.insert(total_envoy_duration, v)
      end
    end


   print("------------------------------\n")
   print("Response from Worker GPU: ", total_worker1)
   print("Response from Worker CPU: ", total_worker2)
   print("Response from Worker idk: ", total_workeridk)
   print(" ")
   print("P50 model execution on GPU (ms):", percentile(total_duration1, 50))
   print("P90 model execution on GPU (ms):", percentile(total_duration1, 90))
   print("P99 model execution on GPU (ms):", percentile(total_duration1, 99))
   print(" ")
   print("P50 model execution on CPU (ms):", percentile(total_duration2, 50))
   print("P90 model execution on CPU (ms):", percentile(total_duration2, 90))
   print("P99 model execution on CPU (ms):", percentile(total_duration2, 99))
   print(" ")
   print("P50 envoy upstream - model execution on GPU (ms):", percentile(total_envoy_duration1, 50))
   print("P90 envoy upstream - model execution on GPU (ms):", percentile(total_envoy_duration1, 90))
   print("P99 envoy upstream - model execution on GPU (ms):", percentile(total_envoy_duration1, 99))
   print(" ")
   print("P50 envoy upstream - model execution on CPU (ms):", percentile(total_envoy_duration2, 50))
   print("P90 envoy upstream - model execution on CPU (ms):", percentile(total_envoy_duration2, 90))
   print("P99 envoy upstream - model execution on CPU (ms):", percentile(total_envoy_duration2, 99))
   print(" ")
   print("P50 envoy upstream - model execution (ms):", percentile(total_envoy_duration, 50))
   print("P90 envoy upstream - model execution (ms):", percentile(total_envoy_duration, 90))
   print("P99 envoy upstream - model execution (ms):", percentile(total_envoy_duration, 99))
   print("------------------------------\n")
end