Detect Hot Items in High‑Concurrency Sales Using OpenResty, Kafka & Elasticsearch

In high‑concurrency scenarios such as JD or Taobao flash‑sale events, thousands of users simultaneously try to purchase items, but 90% of traffic concentrates on a few hot products, which can crash servers.

This article shows how to recognize hot product data in real time using an OpenResty + Lua + Kafka + Elasticsearch solution.

1. Implementation Flow

(1) User accesses a URL (e.g., http://www.longxia.com/web/item/index.html) which is routed to OpenResty.

(2) Nginx location intercepts the request, strips a specific part of the path (e.g., “/web”), and forwards the request to the real service.

(3) Lua assembles product information and sends a message to Kafka; Kafka forwards the data to Logstash, which stores it in Elasticsearch.

2. Core Implementation Details

(1) Configure lua‑resty‑kafka (https://github.com/doujiang24/lua-resty-kafka) as an Nginx module.

(2) Write Lua script to collect request data and produce Kafka messages.

-- dependencies
local cjson = require "cjson"
local client = require "resty.kafka.client"
local producer = require "resty.kafka.producer"

-- Kafka broker list
local broker_list = {
  { host = "192.168.203.237", port = 9092 }
}

-- create producer
local bp = producer:new(broker_list, { producer_type = "async" })

-- get request headers and IP
local headers = ngx.req.get_headers()
local ip = headers["X-REAL_IP"] or headers["X_FORWARDED_FCR"] or ngx.var.remote_addr or "0.0.0.0"

-- build message
local message = {}
message["uri"] = ngx.var.uri
message["ip"] = ip
message["token"] = "987654"
message["actime"] = os.date("%y-%M-%d %H:%m:%S")

-- send message
local ok, err = bp:send("collection_hot_item", nil, cjson.encode(message))

-- strip "/web" from URI and forward
local uri = ngx.var.uri
uri = string.gsub(uri, "/web", "")
ngx.exec(uri)

(3) Nginx configuration to intercept page requests.

# intercept /web/item requests
location /web/item/ {
    content_by_lua_file /usr/local/openresty/nginx/lua/item-access.lua;
}

location /item/ {
    root /usr/local/web;
}

(4) Logstash configuration to consume Kafka messages and push them to Elasticsearch.

input {
  kafka {
    bootstrap_servers => "192.168.203.237:9092"
    topics => ["collection_hot_item"]
    group_id => "logstash_group"
    consumer_threads => 3
    codec => json
  }
}

filter {
}

output {
  elasticsearch {
    hosts => ["http://192.168.203.238:9200"]
    index => "hot_item"
  }
}

Summary: Operators can use Kibana to query hot‑item data stored in Elasticsearch and take targeted mitigation actions; alternatively, Druid can be used for richer analytics.