Materialized View
Harry Yuan
目錄
- View 是啥？
- View 痛點
- Materialized View
- 效能優化
- Demo
View 是啥？
又稱檢視表、虛擬資料表。
由一份指令定義的可查詢物件
View 是啥？
由一份指令定義的可查詢物件
又稱檢視表、虛擬資料表
-
一種關連式資料庫的物件
-
主要用來儲存一組查詢指令
-
它代表你的指令，方便撈取資料
View 是啥？
select count(id)
from users
where created_at > now() - interval '1 day'
老闆：今天台灣DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date and
u.country_code = 'TW'
老闆：今天韓國DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date and
u.country_code = 'KR'
老闆：今天立本DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date and
u.country_code = 'JP'
保留經常重用的code
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date and
u.country_code = 'JP'
拔掉會變動的部分
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date and
u.country_code = 'JP'
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
老闆：今天__DAU多少？by device？
create view dau_report as
老闆：今天__DAU多少？by device？
create view dau_report as
老闆：今天__DAU多少？by device？
create view dau_report as
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at::date = now()::date
dau_report
老闆：今天__DAU多少？by device？
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at > now() - interval '1 day'
select * from dau_report
老闆：今天__DAU多少？by device？
select * from dau_report
老闆：今天台灣DAU多少？by device？
select * from dau_report
where country_code='TW'
老闆：今天韓國DAU多少？by device？
select * from dau_report
where country_code='KR'
老闆：今天立本DAU多少？by device？
select * from dau_report
where country_code='JP'
View 優點
select * from dau_report
方便使用
- 大幅減少指令長度，可快速查詢
- 重用性高（可 join 其他資料表、sub query）
- 權限控管
- 在資料庫中幾乎不佔空間
View 缺點？
- 並沒有儲存任何結果，僅儲存「指令」和「規格」
- 每次都會執行查詢，使用硬體資源
僅儲存「指令」和「規格」
每次都會真的執行查詢
select count(u.id) as total_dau,
sum(case when d.os='ios' then 1 else 0 end) as ios_dau,
sum(case when d.os='aos' then 1 else 0 end) as aos_dau
from users u
inner join device_info d on u.id=d.user_id
where u.last_sign_in_at > now() - interval '1 day'
select * from dau_report
每次都會真的執行查詢
explain analyze select * from dau_report
每次都會真的執行查詢
explain analyze select * from dau_report
Aggregate (cost=31.82..31.83 rows=1 width=24)
-> Hash Join (cost=13.15..31.24 rows=47 width=28)
...
-> Seq Scan on users u (cost=0.00..17.70 rows=147 width=4) (actual time=0.012..0.020 rows=19 loops=1)
Filter: (last_sign_in_at > (now() - '1 day'::interval))
...
-> Seq Scan on device_info d (cost=0.00..11.40 rows=140 width=28) ...
建議
- 盡量單獨使用，用作純搜尋捷徑
- 權限控管：不想讓其他人可以知道 Table 其他細節時可以包成 View
Materialized View
又稱實體化檢視表。
- 由一份指令定義的可查詢物件
- 具有儲存結果的功能
- 使用時只會從自身結果搜尋，不會執行指令
- 特定時候才會執行指令
MView - 儲存功能
在建立時會執行一次指令，並且將結果儲存起來。
下 REFRESH 時，會再次執行指令並刷新資料。
MView - 儲存功能
在建立時會執行一次指令，並且將結果儲存起來。
下 REFRESH 時，會再次執行指令並刷新資料。
MView - 儲存功能
在建立時會執行一次指令，並且將結果儲存起來。
下 REFRESH 時，會再次執行指令並刷新資料。
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
線上看 API: 影片觀看時數
線上看 API: 影片觀看時數
如果資料表非常大？
select count(id) from video_watching_histories
select count(id) from video_watching_histories
count
----------------------------
9029840
如果資料表非常大？
如果資料表非常大？
Server
如果資料表非常大？
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
如果資料表非常大？
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
如果資料表非常大？
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
如果資料表非常大？
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
如果資料表非常大？
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
如果資料表非常大？
- 做 caching
不適合 Caching 的時機
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
cache: user_1
cache: user_4
cache: user_2
cache: user_9
不適合 Caching 的時機
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
cache: user_1
expired: user_4
expired: user_2
expired: user_9
不適合 Caching 的時機
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
cache: user_1
expired: user_4
expired: user_2
expired: user_9
如果資料表非常大？
1. 做 caching
2. 幫大家統一先算好一次
使用 Materialized View
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
使用 Materialized View
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
使用 Materialized View
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
使用 Materialized View
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
使用 Materialized View
create materialized view user_video_report as
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
使用 Materialized View
create materialized view user_video_report as
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
使用 Materialized View
create materialized view user_video_report as
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
建立時會執行一次指令，並儲存結果。
使用 Materialized View
select * from user_video_report
使用 Materialized View
select * from user_video_report
where user_id=1
使用 Materialized View
select * from user_video_report
where user_id=1
1. 真的從 user_video_report 去 select
使用 Materialized View
select * from user_video_report
where user_id=1
1. 真的從 user_video_report 去 select
2. select 不會執行裡面的指令
只搜尋自己的資料
explain analyze select * from user_video_report
where user_id=1
只搜尋自己的資料
explain analyze select * from user_video_report
where user_id=1
Seq Scan on user_video_report (cost=0.00..1.88 rows=5 width=48)
Filter: (user_id = 1)
Rows Removed by Filter: 65
Planning Time: 15.753 ms
Execution Time: 13.564 ms
刷新 Materialized View 資料
1. 真的從 user_video_report 去 select
2. select 不會執行裡面的指令
3. 只有 REFRESH 時，才執行指令並更新資料
refresh user_video_report
select * from user_video_report
where user_id=1
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 2
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 4
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 9
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
Server
使用 Materialized View
select vwh.user_id,
vwh.video_id,
v.name as video_name,
sum(vwh.end_time - vwh.start_time) as time_used
from video_watching_histories vwh
inner join videos v on v.id = vwh.video_id
where vwh.user_id = 1
group by vwh.user_id, v.id, v.name
order by sum(vwh.end_time - vwh.start_time) desc;
select *
from user_video_report
where vwh.user_id = 1
select *
from user_video_report
where vwh.user_id = 2
select *
from user_video_report
where vwh.user_id = 4
select *
from user_video_report
where vwh.user_id = 9
Server
使用 Materialized View
select *
from user_video_report
where vwh.user_id = 1
select *
from user_video_report
where vwh.user_id = 2
select *
from user_video_report
where vwh.user_id = 4
select *
from user_video_report
where vwh.user_id = 9
Server
使用 Materialized View
Response Time 大幅減少
RDS CPU % 減少
select *
from user_video_report
where vwh.user_id = 1
select *
from user_video_report
where vwh.user_id = 2
select *
from user_video_report
where vwh.user_id = 4
select *
from user_video_report
where vwh.user_id = 9
Server
使用 Materialized View
Worker
每15分鐘
refresh user_video_report
select *
from user_video_report
where vwh.user_id = 1
select *
from user_video_report
where vwh.user_id = 2
select *
from user_video_report
where vwh.user_id = 4
select *
from user_video_report
where vwh.user_id = 9
Server
使用 Materialized View
Worker
每15分鐘
refresh user_video_report
自行決定何時執行 CPU Intense 的指令
Recap
| 種類 | 執行指令 | 儲存功能 | 主要用途？ |
|---|---|---|---|
| View | 每次 | 🚫 | 簡化方便查詢 資訊穩定 內建安全性 |
| Materialized View | 建立時 Refresh時 |
✅ | 暫存大量運算結果 本身為table，查詢快速 |
Question
Updatable View?
- 若可以從 view 對應回原資料表欄位就可以。
- 一般都不太會直接從 View 去 Update 資料，也不建議這樣做。
Question
影片觀看記錄其他解法？
- 建立新 table user_video_summary
- 在原資料表 video_watching_records 上建立 trigger
- 當產生一筆 video_watching_records 時觸發 trigger，去更新 user_video_summary
- API 只對 user_video_summary 做搜尋
Demo
Materialized View
By Harry Yuan
