最近在做通知系统，前端用的轮询方式（后端压力不小），因为时间比较紧，开始我准备把未读通知标识存在数据库中，但是每次拿数据的时候需要查询一遍数据库，总监说你这样效率较低，说你可以根据用户id作为key放在redis中存储。说说自己在做这个的一些体会和闲下来总结的一些笔记。

phpredis connect pconnect

我最开始使用的是connect,每次请求完毕关闭连接

1 <?php
2    $redis = new Redis();
3    $nError = $redis->connect('127.0.0.1', 6379);    
4    if ($nError != 1)
5        echo -9998;
6    $redis->incr('newCount');
7    $redis->close();

而另外一种是pconnect

1 <?php
2    $redis = new Redis();
3    $nError = $redis->pconnect('127.0.0.1', 6379);  
4   if ($nError != 1)
5        echo -9998;
6    $redis->incr('newCount');

进行了一下ab压测 ab -c500 -n10000 http://192.168.23.128/redis.php

1.connect

Concurrency Level:      500
Time taken for tests:   25.484 seconds
Complete requests:      10000
Failed requests:        146
   (Connect: 0, Receive: 0, Length: 146, Exceptions: 0)
Non-2xx responses:      146
Total transferred:      1868840 bytes
HTML transferred:       78402 bytes
Requests per second:    392.40 [#/sec] (mean)
Time per request:       1274.198 [ms] (mean)
Time per request:       2.548 [ms] (mean, across all concurrent requests)
Transfer rate:          71.62 [Kbytes/sec] received

2.pconnect

Concurrency Level:      500
Time taken for tests:   13.274 seconds
Complete requests:      10000
Failed requests:        4
   (Connect: 0, Receive: 0, Length: 4, Exceptions: 0)
Non-2xx responses:      4
Total transferred:      1792160 bytes
HTML transferred:       2148 bytes
Requests per second:    753.34 [#/sec] (mean)
Time per request:       663.713 [ms] (mean)
Time per request:       1.327 [ms] (mean, across all concurrent requests)
Transfer rate:          131.85 [Kbytes/sec] received

会发现pconnect比connect会快一倍（会重用redis连接，不需要消耗建立连接时间）

phpredis api上有描述The connection will not be closed on close or end of request until the php process ends

也就是说在php脚本访问结束，也并不会关闭连接，直到php进程结束

我在做pconnect的ab压测结束后，发现有几个redis的连接一直在，直到我杀死php-fpm的进程，也可以通过设置redis.conf文件中的timeout参数

nginx+lua+redis 短连接 连接池

上篇文章提到过nginx+lua+redis，之所以没用在项目中，主要原因是时间紧，对新的东西了解的不多，这里贴一下我闲暇时对于nginx+lua+redis的ab测试

短连接

 1 local redis = require "resty.redis"
 2 local red = redis:new()
 3 local ok, err = red:connect("127.0.0.1", 6379)
 4 if not ok then
 5     ngx.say("failed to connect: ", err)
 6     return
 7 end
 8 ok,err=red:incr("newCount")
 9 if not ok then
10     ngx.say("failed to increase newCount",err)
11     return
12 end
13 
14 local ok,err = red:close()
15 if not ok then 
16     ngx.say("close redis error : ",err)
17     return
18 end

连接池

 1 local redis = require "resty.redis"
 2 local red = redis:new()
 3 local ok, err = red:connect("127.0.0.1", 6379)
 4 if not ok then
 5     ngx.say("failed to connect: ", err)
 6     return
 7 end
 8 ok,err=red:incr("newCount")
 9 if not ok then
10     ngx.say("failed to increase newCount",err)
11     return
12 end
13 
14 local pool_max_idle_time = 10000 --10s
15 local pool_size = 100 --连接池大小
16 local ok,err = red:set_keepalive(pool_max_idle_time,pool_size)
17 if not ok then 
18     ngx.say("set keepalive error : ",err)
19     return
20 end

ab测试  ab -c500 -n10000 http://192.168.23.128/redis

短连接

Concurrency Level:      500
Time taken for tests:   4.791 seconds
Complete requests:      10000
Failed requests:        72
   (Connect: 0, Receive: 0, Length: 72, Exceptions: 0)
Total transferred:      1492016 bytes
HTML transferred:       1944 bytes
Requests per second:    2087.06 [#/sec] (mean)
Time per request:       239.571 [ms] (mean)
Time per request:       0.479 [ms] (mean, across all concurrent requests)
Transfer rate:          304.09 [Kbytes/sec] received

连接池

Concurrency Level:      500
Time taken for tests:   2.986 seconds
Complete requests:      10000
Failed requests:        0
Total transferred:      1490000 bytes
HTML transferred:       0 bytes
Requests per second:    3348.62 [#/sec] (mean)
Time per request:       149.315 [ms] (mean)
Time per request:       0.299 [ms] (mean, across all concurrent requests)
Transfer rate:          487.25 [Kbytes/sec] received

从上面结果可以看出nginx+lua+redis的效率比phpredis的快5倍左右，以后有机会试试在项目中使用

参考链接

http://blog.csdn.net/qmhball/article/details/46988111

https://github.com/openresty/lua-resty-redis