第 5 章 HBase 优化

5.1 RowKey 设计

一条数据的唯一标识就是 rowkey,那么这条数据存储于哪个分区,取决于 rowkey 处于
哪个一个预分区的区间内,设计 rowkey的主要目的 ,就是让数据均匀的分布于所有的 region
中,在一定程度上防止数据倾斜。接下来我们就谈一谈 rowkey 常用的设计方案。

1)生成随机数、hash、散列值
2)时间戳反转
3)字符串拼接

**需求:**使用 hbase 存储下列数据,要求能够通过 hbase 的 API 读取数据完成两个统计需求。

第 5 章 HBase 优化_第1张图片

5.1.1 实现需求 1

为了能够统计张三在 2021 年 12 月份消费的总金额,我们需要用 scan 命令能够得到张三在这个月消费的所有记录,之后在进行累加即可。Scan 需要填写 startRow 和 stopRow:
在这里插入图片描述

scan : startRow -> ^A^Azhangsan2021-12 
 endRow -> ^A^Azhangsan2021-12.

注意点:
(1)避免扫描数据混乱,解决字段长度不一致的问题,可以使用相同阿斯卡码值的符
号进行填充,框架底层填充使用的是阿斯卡码值为 1 的^A。

(2)最后的日期结尾处需要使用阿斯卡码略大于’-’的值
第 5 章 HBase 优化_第2张图片
最终得到 rowKey 的设计为:

//注意 rowkey 相同的数据会视为相同数据覆盖掉之前的版本
rowKey: userdate(yyyy-MM-dd HH:mm:SS)

5.1.2 实现需求 2

问题提出:按照需要 1 的 rowKey 设计,会发现对于需求 2,完全没有办法写 rowKey 的
扫描范围。此处能够看出 hbase 设计 rowKey 使用的特点为:

适用性强 泛用性差 能够完美实现一个需求 但是不能同时完美实现多个需要。

如果想要同时完成两个需求,需要对 rowKey 出现字段的顺序进行调整。

调整的原则为:可枚举的放在前面。其中时间是可以枚举的,用户名称无法枚举,所以
必须把时间放在前面。

最终满足 2 个需求的设计
可以穷举的写在前面即可
rowKey 设计格式 => date(yyyy-MM)^A^Auserdate(-dd hh:mm:ss ms)1)统计张三在 202112 月份消费的总金额
scan: startRow => 2021-12^A^Azhangsan
	  stopRow => 2021-12^A^Azhangsan.2)统计所有人在 202112 月份消费的总金额
scan: startRow => 2021-12
	  stopRow => 2021-12.

5.1.3 添加预分区优化

预分区的分区号同样需要遵守 rowKey 的 scan 原则。所有必须添加在 rowKey 的最前面,前缀为最简单的数字。同时使用 hash 算法将用户名和月份拼接决定分区号。(单独使用用户名会造成单一用户所有数据存储在一个分区)。

添加预分区优化
startKey stopKey
001
001 002
002 003
...
119 120

分区号=> hash(user+date(MM)) % 120

分区号填充 如果得到 1 => 001

rowKey 设计格式 => 分区号 date(yyyy-MM)^A^Auserdate(-dd hh:mm:ss ms)

缺点:实现需求 2 的时候,由于每个分区都有 12 月份的数据,需要扫描 120 个分区。

解决方法:提前将分区号和月份进行对应。

提前将月份和分区号对应一下:
000 到 009 分区 存储的都是 1 月份数据
010 到 019 分区 存储的都是 2 月份数据

110 到 119 分区 存储的都是 12 月份数据

是 9 月份的数据

  • 分区号=> hash(user+date(MM)) % 10 + 80
  • 分区号填充 如果得到 85 => 085

得到 12 月份所有人的数据

  • 扫描 10 次
scan: startRow => 1102021-12
	  stopRow => 1102021-12.
...
	startRow => 1122021-12
	stopRow => 1122021-12.
..
	startRow => 1192021-12
	stopRow => 1192021-12.

5.2 参数优化

1)Zookeeper 会话超时时间

hbase-site.xml

属性:zookeeper.session.timeout

解释:默认值为 90000 毫秒(90s)。当某个 RegionServer 挂掉,90s 之后 Master 才能察觉到。
可适当减小此值,尽可能快地检测 regionserver 故障,可调整至 20-30s。

看你能有都能忍耐超时,同时可以调整重试时间和重试次数
hbase.client.pause(默认值 100ms)
hbase.client.retries.number(默认 15 次)

2)设置 RPC 监听数量

hbase-site.xml

属性:hbase.regionserver.handler.count
解释:默认值为 30,用于指定 RPC 监听的数量,可以根据客户端的请求数进行调整,读写请求较多时,增加此值。

3)手动控制 Major Compaction

hbase-site.xml

属性:hbase.hregion.majorcompaction
解释:默认值:604800000 秒(7 天), Major Compaction 的周期,若关闭自动 Major
Compaction,可将其设为 0。

如果关闭一定记得自己手动合并,因为大合并非常有意义。

4)优化 HStore 文件大小

hbase-site.xml

属性:hbase.hregion.max.filesize

解释:默认值 1073741824010GB),如果需要运行 HBase 的 MR 任务,可以减小此值,
因为一个 region 对应一个 map 任务,如果单个 region 过大,会导致 map 任务执行时间过长。
该值的意思就是,如果 HFile 的大小达到这个数值,则这个 region 会被切分为两个 Hfile。

5)优化 HBase 客户端缓存

hbase-site.xml

属性:hbase.client.write.buffer

解释:默认值 2097152bytes(2M)用于指定 HBase 客户端缓存,
增大该值可以减少 RPC调用次数,但是会消耗更多内存,反之则反之。
一般我们需要设定一定的缓存大小,以达到减少 RPC 次数的目的。

6)指定 scan.next 扫描 HBase 所获取的行数

hbase-site.xml

属性:hbase.client.scanner.caching

解释:用于指定 scan.next 方法获取的默认行数,值越大,消耗内存越大。

7)BlockCache 占用 RegionServer 堆内存的比例

hbase-site.xml

属性:hfile.block.cache.size
解释:默认 0.4,读请求比较多的情况下,可适当调大

8)MemStore 占用 RegionServer 堆内存的比例

hbase-site.xml

属性:hbase.regionserver.global.memstore.size
解释:默认 0.4,写请求较多的情况下,可适当调大

Lars Hofhansl(拉斯·霍夫汉斯)大神推荐 Region 设置 20G,刷写大小设置 128M,其它默认。

hbase-site.xml文件:

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<!--
/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
-->
<configuration>
  <!--
    The following properties are set for running HBase as a single process on a
    developer workstation. With this configuration, HBase is running in
    "stand-alone" mode and without a distributed file system. In this mode, and
    without further configuration, HBase and ZooKeeper data are stored on the
    local filesystem, in a path under the value configured for `hbase.tmp.dir`.
    This value is overridden from its default value of `/tmp` because many
    systems clean `/tmp` on a regular basis. Instead, it points to a path within
    this HBase installation directory.

    Running against the `LocalFileSystem`, as opposed to a distributed
    filesystem, runs the risk of data integrity issues and data loss. Normally
    HBase will refuse to run in such an environment. Setting
    `hbase.unsafe.stream.capability.enforce` to `false` overrides this behavior,
    permitting operation. This configuration is for the developer workstation
    only and __should not be used in production!__

    See also https://hbase.apache.org/book.html#standalone_dist
  -->
  <property>
    <name>hbase.cluster.distributed</name>
    <value>false</value>
  </property>
  <property>
    <name>hbase.tmp.dir</name>
    <value>./tmp</value>
  </property>
  <property>
    <name>hbase.unsafe.stream.capability.enforce</name>
    <value>false</value>
  </property>
<property>
 <name>hbase.zookeeper.quorum</name>
 <value>hadoop102,hadoop103,hadoop104</value>
 <description>The directory shared by RegionServers.
 </description>
 </property>
<property>
 <name>hbase.zookeeper.quorum</name>
 <value>hadoop102,hadoop103,hadoop104</value>
 </property>
<!-**加粗样式**- <property>-->
<!-- <name>hbase.zookeeper.property.dataDir</name>-->
<!-- <value>/export/zookeeper</value>-->
<!-- <description> 记得修改 ZK 的配置文件 -->
<!-- ZK 的信息不能保存到临时文件夹-->
<!-- </description>-->
<!-- </property>-->
 <property>
 <name>hbase.rootdir</name>
 <value>hdfs://hadoop102:8020/hbase
 <description>The directory shared by RegionServers.
 </description>
 </property>
 <property>
 <name>hbase.cluster.distributed</name>
 <value>true</value>
 </property>
</configuration>

5.3 JVM 调优

JVM 调优的思路有两部分:一是内存设置,二是垃圾回收器设置

垃圾回收的修改是使用并发垃圾回收,默认 PO+PS 是并行垃圾回收,会有大量的暂停。
理由是 HBsae 大量使用内存用于存储数据,容易遭遇数据洪峰造成 OOM,同时写缓存的数
据是不能垃圾回收的,主要回收的就是读缓存,而读缓存垃圾回收不影响性能,所以最终设
置的效果可以总结为:防患于未然,早洗早轻松。

1)设置使用 CMS 收集器:

-XX:+UseConcMarkSweepGC

2)保持新生代尽量小,同时尽早开启 GC,例如:

//在内存占用到 70%的时候开启 GC
-XX:CMSInitiatingOccupancyFraction=70

//指定使用 70%,不让 JVM 动态调整
-XX:+UseCMSInitiatingOccupancyOnly

//新生代内存设置为 512m
-Xmn512m

//并行执行新生代垃圾回收
-XX:+UseParNewGC

// 设 置 scanner 扫 描 结 果 占 用 内 存 大 小 , 在 hbase-site.xml 中,设置
hbase.client.scanner.max.result.size(默认值为 2M)为 eden 空间的 1/8
(大概在 64M)

// 设置多个与 max.result.size * handler.count 相乘的结果小于 Survivor 
Space(新生代经过垃圾回收之后存活的对象)

5.4 HBase 使用经验法则

官方给出了权威的使用法则:
第 5 章 HBase 优化_第3张图片

你可能感兴趣的:(hbase,大数据,hadoop)