数据库连接池性能比对(hikari druid c3p0 dbcp jdbc)

小鱼儿 2022-05-25 05:57 148阅读 0赞

#     #

# 背景 #

对现有的[数据库][Link 1]连接池做调研对比，综合性能，可靠性，稳定性，扩展性等因素选出推荐出最优的数据库连接池 。

NOTE: 本文所有测试均是[MySQL][Link 1]库

# 测试结论 #

1：性能方面 hikariCP>druid>tomcat-jdbc>dbcp>c3p0 。hikariCP的高性能得益于最大限度的避免锁竞争。

2：druid功能最为全面，sql拦截等功能，统计数据较为全面，具有良好的扩展性。

3：综合性能，扩展性等方面，可考虑使用druid或者hikariCP连接池。

4：可开启prepareStatement缓存，对性能会有大概20%的提升。

# 功能对比 #

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">功能</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">dbcp</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">druid</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">c3p0</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">tomcat-jdbc</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">HikariCP</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">是否支持PSCache</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">是</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">是</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">是</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">否</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">否</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">监控</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">jmx</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">jmx/log/http</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">jmx,log</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">jmx</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">jmx</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">扩展性</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">弱</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">好</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">弱</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">弱</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">弱</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">sql拦截及解析</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">无</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">支持</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">无</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">无</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">无</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">代码</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">简单</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">中等</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">复杂</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">简单</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">简单</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">更新时间</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2015.8.6</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2015.10.10</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;2015.12.09</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2015.12.3</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">特点</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">依赖于common-pool</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">阿里开源，功能全面</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">历史久远，代码逻辑复杂，且不易维护</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">优化力度大，功能简单，起源于boneCP</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">连接池管理</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">LinkedBlockingDeque</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">数组</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">FairBlockingQueue</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">threadlocal+CopyOnWriteArrayList</td> 
  </tr> 
 </tbody> 
</table>

*   由于boneCP被hikariCP替代，并且已经不再更新，boneCP没有进行调研。
 *  proxool网上有评测说在并发较高的情况下会出错，proxool便没有进行调研。
 *   druid的功能比较全面，且扩展性较好，比较方便对jdbc接口进行监控跟踪等。
 *  c3p0历史悠久，代码及其复杂，不利于维护。并且存在deadlock的潜在风险。

# 性能测试 #

环境配置：

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">CPU</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">Intel(R) Xeon(R) CPU E5-2430 v2 @ 2.50GHz，24core</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">msyql version</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5.5.46</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">tomcat-jdbc version</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">8.0.28</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">HikariCP version</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2.4.3</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">c3p0 Version</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">0.9.5-pre8</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">dbcpVersion</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2.0.1</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">druidVersion</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1.0.5</td> 
  </tr> 
 </tbody> 
</table>

1：获取关闭连接性能测试

测试说明：

*  初始连接和最小连接均为5，最大连接为20。在borrow和return均不心跳检测
 *  其中打开关闭次数为: 100w次
 *  测试用例和mysql在同一台机器上面，尽量避免io的影响
 *  使用mock和连接mysql在不同线程并发下的响应时间

图形：

![20160106104925916][]

![20160106104944246][]

mock性能数据 (单位:ms)

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">20</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">50</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">100</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">tomcat-jdbc</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">442</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">447</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,013</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,264</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">c3p0</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">4,480</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5,527</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">7,449</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">10,725</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">dbcp</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">676</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">689</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">867</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,292</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">hikari</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">38</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">33</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">38</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">30</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">druid</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">291</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">293</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">562</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">985</td> 
  </tr> 
 </tbody> 
</table>

mysql性能数据 (单位:ms)

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">20</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">50</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">100</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">tomcat-jdbc</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">436</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">453</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,033</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,291</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">c3p0</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">4,378</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5,726</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">7,975</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">10,948</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">dbcp</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">671</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">679</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">897</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,380</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">hikari</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">96</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">82</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">87</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">78</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">druid</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">304</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">424</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">690</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,130</td> 
  </tr> 
 </tbody> 
</table>

测试结果：

*  mock和mysql连接性能表现差不多，主要是由于初始化的时候建立了连接后期不再建立连接，和使用mock连接逻辑一致。 
 *  性能表现：hikariCP>druid>tomcat-jdbc>dbcp>c3p0。
 *   hikariCP 的性能及其优异。hikariCP号称java平台最快的数据库连接池。
 *   hikariCP在并发较高的情况下，性能基本上没有下降。
 *   c3p0连接池的性能很差，不建议使用该数据库连接池。

hikariCP性能分析：

*  hikariCP通过优化(concurrentBag，fastStatementList )集合来提高并发的读写效率。
 *  hikariCP使用threadlocal缓存连接及大量使用CAS的机制，最大限度的避免lock。单可能带来cpu使用率的上升。
 *  从字节码的维度优化代码。 (default inline threshold for a JVM running the server Hotspot compiler is 35 bytecodes ）让方法尽量在35个字节码一下，来提升jvm的处理效率。

2：查询一条语句性能测试

测试说明：

*  初始连接和最小连接均为8，最大连接为8。在borrow和return均不心跳检测
 *  查询的次数为10w次，查询的语句为 1：打开连接 2：执行 ：select 1 3：关闭连接
 *  测试用例和mysql在同一台机器上面，尽量避免io的影响

图形：

![20160106105001512][]

测试数据：

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">&nbsp;</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">8</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">20</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">50</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">100</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">tomcat-jdbc</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,178</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,495</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,769</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,818</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,858</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">c3p0</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">3,237</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">3,451</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">4,488</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">5,994</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">7,906</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">dbcp</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,816</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,935</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,097</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,243</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,280</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">hikari</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,299</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,546</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,682</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,751</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,772</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">druid</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,297</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,551</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,800</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,977</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,032</td> 
  </tr> 
 </tbody> 
</table>

测试结果：

*    在并发比较少的情况下，每个连接池的响应时间差不多。是由于并发少，基本上没有资源竞争。
 *    在并发较高的情况下，随着并发的升高，hikariCP响应时间基本上没有变动。
 *    c3p0随着并发的提高，性能急剧下降。

3：pscache性能对比

测试说明：

*  通过druid进行设置pscache和不设置pscache的性能对比
 *  初始连接和最小连接均为8，最大连接为8。在borrow和return均不心跳检测。并且执行的并发数为8.
 *  查询10w次。查询流程为：1：建立连接，2：循环查询preparestatement语句 3：close连接
 *  测试用例和mysql在同一台机器上面，尽量避免io的影响

测试数据：

<table style="border-spacing:0px;width:852px;margin:0px;"> 
 <tbody> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">cache</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">1,927</td> 
  </tr> 
  <tr> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">not cache</td> 
   <td style="padding:7px 10px;margin:0px;vertical-align:top;">2,134</td> 
  </tr> 
 </tbody> 
</table>

测试结果：

*  开启psCache缓存,性能大概有20%幅度的提升。可考虑开启pscache.

测试说明：

*  psCache是connection私有的，所以不存在线程竞争的问题，开启pscache不会存在竞争的性能损耗。
 *  psCache的key为prepare执行的sql和catalog等，value对应的为prepareStatement对象。开启缓存主要是减少了解析sql的开销。

[Link 1]: http://lib.csdn.net/base/14
[20160106104925916]: /images/20220525/490f5b98dcc84a27afa8a5ce3fc936e8.png
[20160106104944246]: /images/20220525/90ace48d9cd54d869c3554cb94da00f0.png
[20160106105001512]: /images/20220525/01c53ed8f0cf4de2bc98351232325d93.png