ClickHouse 实战整理 - 统计 pv uv

ClickHouse 官方文档

文中举例均为简单示例，根据业务自行扩展。

实际业务场景中很多会需要统计 pv uv 值，比如页面的 pv uv 值，商品、店铺的 pv uv 值……

这边举一个简单的例子，比如说统计商品访问的 pv uv：

简单的源数据表创建如下：

CREATE TABLE IF NOT EXISTS test.product_operation
(
    ts_date_time  DateTime COMMENT '事件触发时间',
    gen_date_time DateTime COMMENT '入库时间',
    product_id    UInt64 COMMENT '商品ID',
    user_id       UInt64 COMMENT '用户ID',
    operation     UInt16 COMMENT '商品操作类型（1：访问，2：购买，3：收藏）'
) ENGINE = MergeTree()
      ORDER BY (toDate(ts_date_time), operation);

像这种数据量可能很大，又需要预先聚合计算以减小获取耗时的，我们添加物化视图来帮助我们做这个事。

创建按天聚合数据的物化视图：

CREATE MATERIALIZED VIEW IF NOT EXISTS test.product_operation_day_pv_uv_mv
            ENGINE = AggregatingMergeTree()
                ORDER BY (ts_date, product_id, operation)
AS
SELECT toDate(ts_date_time)        as ts_date,
       product_id,
       operation,
       sumState(1)                 as pv,
       uniqState(user_id, ts_date) as uv
FROM test.product_operation
GROUP BY ts_date, product_id, operation;

创建总的聚合数据的物化视图：

CREATE MATERIALIZED VIEW IF NOT EXISTS test.product_operation_total_pv_uv_mv
            ENGINE = AggregatingMergeTree()
                ORDER BY (product_id, operation)
AS
SELECT product_id,
       operation,
       sumState(1)        as pv,
       uniqState(user_id) as uv
FROM test.product_operation
GROUP BY product_id, operation;

现在我们往源数据表中插入一条数据：

1 2	INSERT INTO test.product_operation(ts_date_time, gen_date_time, product_id, user_id, operation) VALUES (toDateTime('2022-01-26 10:10:10'), now(), 1, 1, 1)

再查询：

忽略时区问题导致的时间显示

1	SELECT * FROM test.product_operation ORDER BY gen_date_time DESC;

ts_date_time	gen_date_time	product_id	user_id	operation
2022-01-26 02:10:10	2022-01-26 03:54:36	1	1	1

查询视图（day）：

1 2	SELECT ts_date, product_id, operation, sumMerge(pv) AS pv, uniqMerge(uv) AS uv FROM test.product_operation_day_pv_uv_mv GROUP BY ts_date, product_id, operation;

ts_date	product_id	operation	pv	uv
2022-01-26	1	1	1	1

查询视图（total）：

1 2	SELECT product_id, operation, sumMerge(pv) AS pv, uniqMerge(uv) AS uv FROM test.product_operation_total_pv_uv_mv GROUP BY product_id, operation;

product_id	operation	pv	uv
1	1	1	1

再往源数据插入数据：

1 2	INSERT INTO test.product_operation(ts_date_time, gen_date_time, product_id, user_id, operation) VALUES (toDateTime('2022-01-26 23:10:10'), now(), 1, 1, 1);

查询视图：

ts_date	product_id	operation	pv	uv
2022-01-26	1	1	2	1

product_id	operation	pv	uv
1	1	2	1

换个 user_id 插入数据：

1 2	INSERT INTO test.product_operation(ts_date_time, gen_date_time, product_id, user_id, operation) VALUES (toDateTime('2022-01-26 23:20:10'), now(), 1, 2, 1);

查询视图：

ts_date	product_id	operation	pv	uv
2022-01-25	1	1	3	2

product_id	operation	pv	uv
1	1	3	2

再换个 ts_date_time 插入数据：

1
2
3

INSERT INTO test.product_operation(ts_date_time, gen_date_time, product_id, user_id, operation)
VALUES (toDateTime('2022-01-27 10:20:10'), now(), 1, 1, 1),
       (toDateTime('2022-01-27 11:20:10'), now(), 1, 2, 1);

查询视图：

ts_date	product_id	operation	pv	uv
2022-01-26	1	1	3	2
2022-01-27	1	1	2	2

product_id	operation	pv	uv
1	1	5	2

综上，实现不难，clickhouse 丰富的聚合函数可以节省很多操作。

还可以用 Kafka引擎做数据的增量同步，结合视图。