TimesTen 应用层数据库缓存学习：8. 配置Sliding Window(滑动窗口)

Sliding Window（滑动窗口）是Cache Group常用的一种场景。
本文介绍滑动窗口的基本概念以及如何配置滑动窗口。

滑动窗口的描述如下：

In a sliding window configuration, new rows are inserted into and old rows are deleted from the cache tables on a regular schedule so that the tables contain only the data that satisfies a specific time interval.

如何实现：

You can configure a sliding window for a cache group by using incremental autorefresh mode and defining a time-based aging policy. The autorefresh operation checks the timestamp of the rows in the cached Oracle Database tables to determine whether new data should be refreshed into the TimesTen cache tables. The system time and the time zone must be identical on the Oracle Database and TimesTen systems.

有两点必须注意，一个是时区和时间必须一致（在我们的实验环境中没有问题，TimesTen和Oracle都在一台主机上），一个是使用基于时间的Aging和增量autorefresh。

滑动窗口最常用的配置是Explicitly load Read-only Cache Group

Explicitly load Read only Sliding Window

只读滑动窗口是最常用的，可以使用incremental auto refresh。

例如，下面的缓存组缓存近一个月的订单，然后每天将新的订单加入缓存组：

CREATE READONLY CACHE GROUP recent_shipped_orders AUTOREFRESH MODE INCREMENTAL INTERVAL 1440 MINUTES STATE ON FROM oratt.orders (ord_num NUMBER(10) NOT NULL, cust_num NUMBER(6) NOT NULL, when_placed DATE NOT NULL, when_shipped DATE NOT NULL, PRIMARY KEY(ord_num)) AGING USE when_shipped LIFETIME 30 DAYS CYCLE 24 HOURS ON;

我们可以模拟一下这个场景，但时间可以短些。

# 先在Oracle中建立订单表
$ sqlplus tthr/oracle@ttorcl
create table orders(ord_num int primary key, ship_time timestamp not null);
insert into orders values(1, sysdate);
insert into orders values(2, sysdate);
commit;

SQL> set linesize 100
SQL> select * from orders;

   ORD_NUM SHIP_TIME
---------- ---------------------------------------------------------------------------
         1 14-APR-16 07.10.03.000000 AM
         2 14-APR-16 07.10.04.000000 AM



SQL> grant select on orders to cacheadm;

Grant succeeded.

ttisql -connstr "dsn=cachedb1_1122;uid=cacheadm;pwd=timesten;oraclepwd=oracle" -e "set prompt 'cacheadm>'"
CREATE READONLY CACHE GROUP "RO_SW" 
 AUTOREFRESH MODE INCREMENTAL INTERVAL 10 SECONDS
 STATE ON
 FROM
  "TTHR"."ORDERS" (
    "ORD_NUM"   NUMBER(38)   NOT NULL,
    "SHIP_TIME" TIMESTAMP(6) NOT NULL,
    PRIMARY KEY("ORD_NUM")
  )
  AGING USE SHIP_TIME LIFETIME 300 seconds CYCLE 10 seconds ON

cacheadm>cachegroups;

Cache Group CACHEADM.RO_SW:

  Cache Group Type: Read Only
  Autorefresh: Yes
  Autorefresh Mode: Incremental
  Autorefresh State: On
  Autorefresh Interval: 10 Seconds
  Autorefresh Status: ok
  Aging: Timestamp based uses column SHIP_TIME lifetime 300 seconds cycle 10 seconds on

  Root Table: TTHR.ORDERS
  Table Type: Read Only

1 cache group found.

tthr>select * from orders;
< 1, 2016-04-14 07:10:03.000000 >
< 2, 2016-04-14 07:10:04.000000 >
2 rows found.

# Oracle数据库中
# 用户sys执行 grant execute on dbms_lock to public;
# 用户tthr执行以下过程，产生订单：
declare ord_num number;      
begin
for i in 1..1000 loop
select max(ord_num) into ord_num from orders;
insert into orders values(ord_num+1, sysdate);
commit;
dbms_lock.sleep( 10 );
end loop;
end;
/

$ ttisql  -connstr "dsn=cachedb1_1122;uid=tthr;pwd=timesten;oraclepwd=oracle" -e "set prompt 'tthr>'"

tthr>select ord_num, TIMESTAMPDIFF(SQL_TSI_SECOND,ship_time, sysdate) from orders;
< 12, 299 >
< 13, 289 >
< 14, 279 >
< 15, 269 >
< 16, 259 >
< 17, 249 >
< 18, 239 >
< 19, 229 >
< 20, 219 >
< 21, 209 >
< 22, 199 >
< 23, 189 >
< 24, 179 >
< 25, 169 >
< 26, 159 >
< 27, 149 >
< 28, 139 >
< 29, 129 >
< 30, 119 >
< 31, 109 >
< 32, 99 >
< 33, 89 >
< 34, 79 >
< 35, 69 >
< 36, 59 >
< 37, 49 >
< 38, 39 >
< 39, 29 >
< 40, 19 >
< 41, 9 >
30 rows found.
tthr>
过一小会
tthr>select ord_num, TIMESTAMPDIFF(SQL_TSI_SECOND,ship_time, sysdate) from orders;
< 18, 291 >
< 19, 281 >
< 20, 271 >
< 21, 261 >
< 22, 251 >
< 23, 241 >
< 24, 231 >
< 25, 221 >
< 26, 211 >
< 27, 201 >
< 28, 191 >
< 29, 181 >
< 30, 171 >
< 31, 161 >
< 32, 151 >
< 33, 141 >
< 34, 131 >
< 35, 121 >
< 36, 111 >
< 37, 101 >
< 38, 91 >
< 39, 81 >
< 40, 71 >
< 41, 61 >
< 42, 51 >
< 43, 41 >
< 44, 31 >
< 45, 21 >
< 46, 11 >
29 rows found.
tthr>

# 可以看到，时间差总是在300秒以内，订单从41滑动到46，老的订单12-17被淘汰

对于dynamic read-only以及AWT是否可以配置滑动窗口呢，答案是肯定的，只不过不是很常用。

If the cache group does not use incremental autorefresh mode, you can configure a sliding window by using a LOAD CACHE GROUP, REFRESH CACHE GROUP, or INSERT statement, or a dynamic load operation to bring new data into the cache tables.

Dynamic Load Read only Sliding Window

CREATE DYNAMIC READONLY CACHE GROUP "DRO_SW" 
 AUTOREFRESH MODE INCREMENTAL INTERVAL 10 SECONDS
 STATE PAUSED
 FROM
  "TTHR"."ORDERS" (
    "ORD_NUM"   NUMBER(38)   NOT NULL,
    "SHIP_TIME" TIMESTAMP(6) NOT NULL,
    PRIMARY KEY("ORD_NUM")
  )
  AGING USE SHIP_TIME LIFETIME 300 seconds CYCLE 10 seconds ON
cacheadm>cachegroups;

Cache Group CACHEADM.DRO_SW:

  Cache Group Type: Read Only (Dynamic)
  Autorefresh: Yes
  Autorefresh Mode: Incremental
  Autorefresh State: Paused
  Autorefresh Interval: 10 Seconds
  Autorefresh Status: ok
  Aging: Timestamp based uses column SHIP_TIME lifetime 300 seconds cycle 10 seconds on

  Root Table: TTHR.ORDERS
  Table Type: Read Only

1 cache group found.

tthr>select count(*) from orders;
< 0 >
1 row found.
# 虽然可以通过dynamic load导入数据，不过批量导入还是得靠load。
# Load的作用是将新的数据导入缓存组，而autofresh同时也在工作，将已在缓存组的数据进行更新。
# 例如如果在TimesTen中的订单79，在Oracle端删除了，那么autorefresh也会将其删除

# 用户tthr执行以下过程，产生订单：
declare ord_num number;      
begin
for i in 1..1000 loop
select max(ord_num) into ord_num from orders;
insert into orders values(ord_num+1, sysdate);
commit;
dbms_lock.sleep( 10 );
end loop;
end;
/
cacheadm>load cache group dro_sw commit every 256 rows;
3 cache instances affected.

tthr>select count(*) from orders;
< 3 >
1 row found.

cacheadm>load cache group dro_sw commit every 256 rows;
2 cache instances affected.

tthr>select count(*) from orders;
< 5 >
1 row found.

tthr>select ord_num, TIMESTAMPDIFF(SQL_TSI_SECOND,ship_time, sysdate) from orders;
< 78, 299 >
< 79, 289 >
< 81, 269 >
< 82, 259 >
< 83, 249 >
< 84, 239 >
< 85, 229 >
< 86, 219 >
< 87, 209 >
< 88, 199 >
< 89, 189 >
< 90, 179 >
< 91, 169 >
< 92, 159 >
< 93, 149 >
< 94, 139 >
< 95, 129 >
< 96, 119 >
< 97, 109 >
< 98, 99 >
< 99, 89 >
< 100, 79 >
< 101, 69 >
< 102, 59 >
< 103, 49 >
< 104, 39 >
< 105, 29 >
< 106, 19 >
28 rows found.
时间差总在300秒内

Explicitly Load AWT Sliding Window

前面两例为只读缓存，数据来源是Oracle。本例为AWT，数据来源在TimesTen。
使用场景为TimesTen作为数据收集的集中点，然后保存时间窗口的数据作为实时分析。
下面三个提示符分别是用以下三个命令登录的
* cacheadm - ttisql -connstr "dsn=cachedb1_1122;uid=cacheadm;pwd=timesten;oraclepwd=oracle" -e "set prompt 'cacheadm>'"

* tthr(schema user) - ttisql -connstr "dsn=cachedb1_1122;uid=tthr;pwd=timesten;oraclepwd=oracle" -e "set prompt 'tthr>'"

* Oracle DB - sqlplus tthr/oracle@ttorcl

SQL> grant select, insert, update, delete on orders to cacheadm;

CREATE ASYNCHRONOUS WRITETHROUGH CACHE GROUP "AWT_SW" 
 FROM
  "TTHR"."ORDERS" (
    "ORD_NUM"   NUMBER(38)   NOT NULL,
    "SHIP_TIME" TIMESTAMP(6) NOT NULL,
    PRIMARY KEY("ORD_NUM")
  )
  AGING USE SHIP_TIME LIFETIME 300 seconds CYCLE 10 seconds ON

cacheadm>call ttrepstart;
tthr>select * from orders;
0 rows found.
cacheadm>cachegroups;

Cache Group CACHEADM.AWT_SW:

  Cache Group Type: Asynchronous Writethrough
  Autorefresh: No
  Aging: Timestamp based uses column SHIP_TIME lifetime 300 seconds cycle 10 seconds on

  Root Table: TTHR.ORDERS
  Table Type: Propagate

1 cache group found.
cacheadm>repschemes;

Replication Scheme TTREP._AWTREPSCHEME:

  Element: _1798096                       
  Type: Table TTHR.ORDERS
  Master Store: CACHEDB1_1122 on TIMESTEN-HOL Transmit Durable
  Subscriber Store: _ORACLE from TIMESTEN-HOL 

  Store: CACHEDB1_1122 on TIMESTEN-HOL
    Port: (auto)
    Log Fail Threshold: (none)
    Retry Timeout: 120 seconds
    Compress Traffic: Disabled

  Store: _ORACLE from TIMESTEN-HOL
    Port: (auto)
    Log Fail Threshold: (none)
    Retry Timeout: 120 seconds
    Compress Traffic: Disabled

1 replication scheme found.
SQL> select count(*) from orders;

  COUNT(*)
----------
       134
# Oracle中有数据，但是一条也没load进来，原因是这些数据过老，不满足时间窗口的要求
cacheadm>load cache group awt_sw commit every 256 rows;
0 cache instances affected.
# 为了演示方便，我们删除掉Oracle中的数据
SQL> truncate table orders;

Table truncated.
# 然后在Oracle中初始化一条数据
SQL> insert into orders values(1, sysdate);

1 row created.

SQL> commit;

Commit complete.
cacheadm>load cache group awt_sw commit every 256 rows;
1 cache instance affected.
tthr>select * from orders;                                                                                                                                              < 1, 2016-04-14 22:28:35.000000 >
1 row found.

# 然后在TimesTen端插入数据
tthr>ALTER SESSION SET PLSQL_TIMEOUT = 0; <- 缺省是30秒

Session altered.
declare ord_num number;      
begin
for i in 1..1000 loop
select max(ord_num) into ord_num from orders;
insert into orders values(ord_num+1, sysdate);
commit;
dbms_lock.sleep( 10 );
end loop;
end;
/


# 另起一个tthr会话
tthr>select * from orders;
< 2, 2016-04-14 22:29:59.000000 >
< 3, 2016-04-14 22:30:09.000000 >
< 4, 2016-04-14 22:30:19.000000 >
< 5, 2016-04-14 22:33:35.000000 >
< 6, 2016-04-14 22:33:45.000000 >
< 7, 2016-04-14 22:33:55.000000 >
6 rows found.
# 我们可以看到，TimesTen中的数据不断增加
# 而时间差在300秒以内
tthr>select ord_num, TIMESTAMPDIFF(SQL_TSI_SECOND,ship_time, sysdate) from orders;
< 16, 300 >
< 17, 290 >
< 18, 280 >
< 19, 270 >
< 20, 260 >
< 21, 250 >
< 22, 240 >
< 23, 230 >
< 24, 220 >
< 25, 210 >
< 26, 200 >
< 27, 190 >
< 28, 180 >
< 29, 170 >
< 30, 160 >
< 31, 150 >
< 32, 140 >
< 33, 130 >
< 34, 120 >
< 35, 110 >
< 36, 100 >
< 37, 90 >
< 38, 80 >
< 39, 70 >
< 40, 60 >
< 41, 50 >
< 42, 40 >
< 43, 30 >
< 44, 20 >
< 45, 10 >
30 rows found.

# 而在Oracle中保留有所有的历史数据
SQL> select count(*) from orders;

  COUNT(*)
----------
        55

Dynamic Load AWT Sliding Window

就不讨论了，和上例差不多