MySQL分区表、HBase 融合秘术

共计 8101 个字符，预计需要花费 21 分钟才能阅读完成。

导读	时常都会有人问 MySQL 分区表要如何使用，MySQL 分区表的和 Oracle 的差远了，该不该用 MySQL 分区表。

啰哩八嗦

其实该不该用，我也不能给予很好的建议。还是那句话，觉得适合自己才是最好的。觉得自己可以搞定分区表那就用。

我多虑了

好很好的使用分区表就需要做好对开发人员培训的准备，让他们知道要怎么样才能很好的使用分区表。
最好需要有一个自动化的计划，定时的自动处理分区的问题。
DELETE 删除数据不会释放磁盘空间，DROP PARITION 会释放磁盘空间，这样节省了空间的同时，也不会因为 delete 标记数据过多带来性能问题。

开干了

创建 MySQL 分区数据

DROP TABLE ord_order;

-- 创建订单分区表
CREATE TABLE ord_order(
order_id BIGINT NOT NULL AUTO_INCREMENT COMMENT '订单 ID',
user_id INT NOT NULL COMMENT '用户 ID',
goods_id INT NOT NULL COMMENT '商品 ID',
order_price INT NOT NULL DEFAULT 0 COMMENT '订单实际价格 (分)',
create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
PRIMARY KEY(order_id, create_time)
)
PARTITION BY LIST (YEAR(create_time)*100 + MONTH(create_time))
(PARTITION p201601 VALUES IN (201601),
PARTITION p201602 VALUES IN (201602),
PARTITION p201603 VALUES IN (201603),
PARTITION p201604 VALUES IN (201604),
PARTITION p201605 VALUES IN (201605),
PARTITION p201606 VALUES IN (201606),
PARTITION p201607 VALUES IN (201607),
PARTITION p201608 VALUES IN (201608),
PARTITION p201609 VALUES IN (201609),
PARTITION p201610 VALUES IN (201610),
PARTITION p201611 VALUES IN (201611),
PARTITION p201612 VALUES IN (201612)
);

-- 插入相关数据
INSERT INTO ord_order VALUES
(NULL, 10000001, 11111111, 1000, '2016-01-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-01-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-01-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-01-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-01-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-02-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-02-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-02-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-02-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-02-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-03-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-03-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-03-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-03-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-03-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-04-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-04-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-04-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-04-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-04-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-05-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-05-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-05-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-05-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-05-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-06-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-06-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-06-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-06-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-06-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-07-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-07-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-07-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-07-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-07-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-08-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-08-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-08-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-08-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-08-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-09-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-09-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-09-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-09-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-09-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-10-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-10-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-10-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-10-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-10-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-11-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-11-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-11-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-11-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-11-13 05:00:50'),
(NULL, 10000001, 11111111, 1000, '2016-12-13 01:00:10'),
(NULL, 10000001, 11111112, 2000, '2016-12-13 02:00:20'),
(NULL, 10000001, 11111113, 3000, '2016-12-13 03:00:30'),
(NULL, 10000001, 11111114, 4000, '2016-12-13 04:00:40'),
(NULL, 10000001, 11111115, 5000, '2016-12-13 05:00:50');

-- 查看分区 p201601 数据
SELECT * FROM ord_order PARTITION(p201601);

-- 组合成的 row key
SELECT CONCAT(user_id, 10000000000-UNIX_TIMESTAMP(create_time), goods_id)
FROM ord_order PARTITION(p201601);

结合 HBase 咯

创建 HBase 表 ord_order

由于版本兼容的问题，这边我需要先创建好 HBase 对应的表。不然会报不能自动创建 column family 的错误。

使用 hbase shell 创建 ord_order 表

hbase(main):033:0> create 'ord_order', {NAME => 'cf1'}

使用 Sqoop 将 MySQL 的 ord_order 表的 p201601 分区的数据导入 HBase 表。

/usr/local/sqoop/bin/sqoop import \
--connect jdbc:mysql://192.168.137.11:3306/test \
--username HH \
--password oracle \
--query 'SELECT CONCAT(user_id, 10000000000-UNIX_TIMESTAMP(create_time), goods_id) AS order_id, order_price, create_time FROM ord_order PARTITION(p201601) WHERE $CONDITIONS' \
--hbase-table ord_order \
--hbase-create-table \
--hbase-row-key order_id \
--split-by order_id \
--column-family cf1 \
-m 1

导入成功后就可以在 MySQL 上面将相关分区删除，并且创建之后需要的分区

ALTER TABLE ord_order
ADD PARTITION (PARTITION p201701 VALUES IN (201701));

ALTER TABLE ord_order DROP PARTITION p201601;

查看 Hbase 中导入的数据

hbase(main):001:0> scan 'ord_order'
ROW COLUMN+CELL
10000001854736755011111115 column=cf1:create_time, timestamp=1479224942888, value=2016-01-13 05:00:50.0
10000001854736755011111115 column=cf1:order_price, timestamp=1479224942888, value=5000
10000001854737116011111114 column=cf1:create_time, timestamp=1479224942888, value=2016-01-13 04:00:40.0
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000
10000001854737477011111113 column=cf1:create_time, timestamp=1479224942888, value=2016-01-13 03:00:30.0
10000001854737477011111113 column=cf1:order_price, timestamp=1479224942888, value=3000
10000001854737838011111112 column=cf1:create_time, timestamp=1479224942888, value=2016-01-13 02:00:20.0
10000001854737838011111112 column=cf1:order_price, timestamp=1479224942888, value=2000
10000001854738199011111111 column=cf1:create_time, timestamp=1479224942888, value=2016-01-13 01:00:10.0
10000001854738199011111111 column=cf1:order_price, timestamp=1479224942888, value=1000
5 row(s) in 0.5390 seconds

ROW KEY 设计详解

HBase 中的 row key 为 user_id, 10000000000-UNIX_TIMESTAMP(create_time), goods_id 3 个字段组成。

这边值得注意的是 10000000000-UNIX_TIMESTAMP(create_time), 这样设计的原因是为了让订单能按时间的倒序排列, 这样就符合越新的数据越先显示

如: 现在需要对用户 10000001 的订单进行分页, 每页两条数据, 并且按时间的倒序排序 (最新订单最先显示)

hbase(main):003:0> scan 'ord_order', {COLUMNS=>['cf1:order_price'], ROWPREFIXFILTER=>'10000001', LIMIT=>2}
ROW COLUMN+CELL
10000001854736755011111115 column=cf1:order_price, timestamp=1479224942888, value=5000
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000

点击下一页的数据:

hbase(main):004:0> scan 'ord_order', {COLUMNS=>['cf1:order_price'], LIMIT=>3, STARTROW=>'10000001854737116011111114'}
ROW COLUMN+CELL
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000
10000001854737477011111113 column=cf1:order_price, timestamp=1479224942888, value=3000
10000001854737838011111112 column=cf1:order_price, timestamp=1479224942888, value=2000
3 row(s) in 0.0260 seconds
上面获得了三行数据，在实际展现的时候去除第一行就好了，实际展示如下:
10000001854737477011111113 column=cf1:order_price, timestamp=1479224942888, value=3000
10000001854737838011111112 column=cf1:order_price, timestamp=1479224942888, value=2000

点击上一页

hbase(main):008:0> scan 'ord_order', {COLUMNS=>['cf1:order_price'], LIMIT=>3, STARTROW=>'10000001854737477011111113', REVERSED=>true}
ROW COLUMN+CELL
10000001854737477011111113 column=cf1:order_price, timestamp=1479224942888, value=3000
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000
10000001854736755011111115 column=cf1:order_price, timestamp=1479224942888, value=5000
3 row(s) in 0.0640 seconds
上面同样获得了三条数据，我们需要去除第一行，让后按数据集合倒序显示
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000
10000001854736755011111115 column=cf1:order_price, timestamp=1479224942888, value=5000
↓↓↓↓↓ 上面两行是集合数据 下面两行数倒序遍历集合的数据 (也是最终显示的数据)
10000001854736755011111115 column=cf1:order_price, timestamp=1479224942888, value=5000
10000001854737116011111114 column=cf1:order_price, timestamp=1479224942888, value=4000

总结

对 HBase 的设计和使用是门学问，在实际使用中，一般需要和开发产品一起讨论如何设计 row key 比较好。当然，一般多多阅读过来人的经验往往也能够解决很多问题。因为你遇到的问题，别人可能也会遇到。