Built-in Functions in Teradata

http://readvitamin.com/2007/08/09/how-to-use-built-in-functions-in-teradata/


Written by admin in Thursday, August 9th 2007 under IT  

First of all what are Built-in functions? These are functions which do not need any sort of arguments but still returns information about the system; user, date, time, session etc. Built-in functions are sometimes referred to as special registers. Mostly used with the SELECT statement. Usage;

SELECT

• ACCOUNT: - If you are wondering about your Teradata Account information just use this function to see your current account string.

• CURRENT_DATE:- Returns the current system date and it is same as the function DATE.

• CURRENT_TIME:- This function returns the current system time and current session ‘Time Zone’ displacement. This function is similar to the TIME function but with extra feature of Time Zone displacement.

• CURRENT_TIMESTAMP:- Returns the current system timestamp (including year, month and day) and current session Time Zone displacement.

• DATABASE: - If you are wondering how to find your current database name, use this function which returns the name of the default database for the current user.

• DATE: - As mentioned earlier, it does return the current date.

• SESSION: - Returns a number for the session the current user is in. Two sessions does not have the same number. If you want to identify your session, this is the function to use.

 • TIME: - As mentioned, this function provides the current time based on a 24-hour day; mean to say for 4:00 pm, you would see 16:00:00.

 • USER: - If you have forgotten your username after you have logged in, this command would come to your rescue. This one gives the user name of the current user.

Tweet

Import/Export Data by SQL Assistant

Export
SQL Assistant -> Exprort Data ->输出至文件
Import
SQL Assistant -> Import Data
insert into dwhtst.TMSL_CUST_MAJ  values (?,?,?,'Y',NULL,NULL,NULL)
执行并选择输入源文件

不过好慢

Tweet

SSRS动态设置URL有无

原来URL的有无也是可以设定的


=IIF(Fields!qty_act.Value Is Nothing or Fields!customer.Value = "-" or Fields!customer.Value = "" or UCASE(Fields!customer.Value) = "OTHERS",
Nothing,
"MBI_TVSIRPTRS017")

Tweet

TD性能调优和物理实现培训总结 2

原帖地址
http://blog.chinaunix.net/u3/103331/showart_2050758.html

总结 2
1、Parcel
   Client to Server(Request Parcel(1)、Data Parcel(0 or N)、Response Parcel)

   Parcel步骤：REQUEST PARCEL -> SYNTAXER -> RESOLVER -> SECURITY -> OPTIMIZER -> GENERATOR -> GNCApply -> AMP STEPS
   
   使用Data Parcel的好处，是可以让PE缓存执行计划，重复执行，提高效率。
   使用意见：对于那种重复执行的动作，比如重复插入一堆的数据的时候，最好将数据弄成文件的形式，采用Using 的方式在BTEQ中提交，能提高很多效率。
   如何查看执行的sql是否被cache呢？
   Answer:通过DBQLogTbl表，里面的字段CacheFlag可以查看该语句在执行的时候是否有被cache（PE的解析Cache是采用LRU算法，4个小时最长生命周期，既4个小时候会把该sql语句给unmarked）

   Parcel的RESOLVER过程，会使用系统字典表来检查sql语句是否正确，那么是否会锁系统表呢？
   Answer:Parcel的RESOLVER也是使用缓存来存系统字典表信息，不会锁系统表。Data Dictionary Cache

2、Join Strategies
   Teradata的设计思想是：Share Nothing，也就是说各个AMP之间是独立，每个AMP都不知道其他AMP到底存储有什么样的数据，所以对于Join、比较等，都需要在本AMP进行了，也就有了所有的Redistribution和Duplicate to all amp。

   Product Join：Duplicate smaller table to all amp,(not sort)
   Merge Join：Duplicate smaller table or redistribute table which don't join with pi. Sort on join column row hash.
   merge join 由于Sort column row hash，所以可以减少比较的次数（不用像Product Join那样每个都需要比较）。
   
   Join如果非正常的慢可能有以下几个情况：
   which is redistributed by hash code to all AMPs ： Skew表被Redistribute了
   which is duplicated on all AMPs ： 大表被Duplicate了。  
   这就涉及到一个统计信息的问题。如果收集的统计信息没有及时的更新，比如说建完一个表，收集了统计信息，然后统计信息里面这个表是小表（因为是空表）。而当这个表数据量变大时，没有更新统计信息，则优化器还是认为他是小表，可能就会duplicate这张表，那就是个大工作量了。
   
   Where条件中，关于OR操作的优化，可以采用UNION的方式，放到2个SQL语句中执行，而不做product join。
    
SELECT  Department_Name, Last_Name, First_Name 
FROM     Employee INNER JOIN Department 
ON           Employee.Department_Number = Department.Department_Number 
OR           Employee.Employee_Number = Department.Manager_Employee_Number; 
变为
SELECT  Department_Name, Last_Name, First_Name
FROM     Employee INNER JOIN Department
ON          Employee.Department_Number = Department.Department_Number
UNION UNION
SELECT  Department_Name, Last_Name, First_Name
FROM     Employee INNER JOIN Department
ON          Employee.Employee_Number = Department.Manager_Employee_Number;

   Nested Join(最简单的join)：join的一个表而且有确定的where条件（UPI或者USI） ，只能选择出一条记录

3、PPI
  PPI的好处：特殊查询能力的增强，比如说按日期查询（设置按日期的PPI），避免全表扫描
  PPI可能出现的问题：多了2个字节的空间浪费（partition号），对PI的查询可能会比没有定义PPI的慢(先确定了partition在访问里面的row hash)。
  如果partition 不是PI的一部分的话，PI不能定义为UPI，而为了保证唯一性，可以定义USI在字段上。
  思考？为什么partition不是PI的一部分的话，PI不能定义为UPI
  Answer：UPI能确保唯一性，而如果partition不是PI的一部分的话，如果使用PI来Access的话，会扫描每个partition在定位，已经失去了作为UPI的意义了。
  
  DBC.IndexConstraints中ConstraintsType标志位Q的表示PPI。

  PPI的适用情形：
  1、大表
  2、partition字段经常被Access
  3、partition的数量是有限制的（65535）

4、Aggregation
  ARSA算法：
  Aggregate - Local on the AMPs
  Redistribute - The Local Aggregations
  Sort - The Redistribution Aggregations
  Aggregate - Globally for the final aggregation

  Group by VS Distinct
  1、适用情况：
     Group by 先在本AMP上分组，然后在在Globally上分组
     Distinct 是先把数据分布到All AMP上在去重。
     所以，在重复值比较多的情况下，Group by 能提高点性能，其他情况下，Distinct比较快点。（不是绝对的）
  2、关于Eliminate
     group by 没有进行eliminate操作，distinct有进行这样的操作。
     （in、not in、unique、distinct操作会去重）

5、Date
  TD中，日期以数字来存储。如2009-09-09在数据库中被转换为20090909，在数据库中存储过数字10090909（20090909-19000000）
  如果时间小于19000000则在数据库中存储为负值。
  
        SELECT EXTRACT(YEAR FROM DATE '2004-03-31') ||
        EXTRACT(MONTH FROM DATE '2004-03-31')
        result:       2004          3
        整数转换为11位字符在||

 

Tweet

TD性能调优和物理实现培训总结 1

原帖地址

http://blog.chinaunix.net/u3/103331/showart_2045801.html

1、几种索引的判定：
Join Index 预先进行关联的表，类似于“物化视图”，需要占用表空间，Join Index可以自动更新。
Global Index 在选择出来的字段中有rowid
Sparse Index 在Index中使用了Where条件
Aggregate Join Index 预先在index中进行Aggregate操作(同summary table的区别：summary table需要手动更新，Aggregate Join index可以自动更新)

2、Store Procedure和Macro的区别：
   Macro和Stored Produre同样能使用变量，调用sql语句。其中Macro一般比Stored Procedure 简单，同时Stored Procedure支持SPL操作，既IF..ELSE...THEN等。

3、TDWM：Teradata Dynamic Workload Manager，混合负载管理，动态的管理资源分布和优先级。（目前正在做试验）

4、关于视图
  加视图的好处：
  1、SQL语句---保存复杂的SQL语句来直接运行
  2、保密---可以在sql语句中使用一定的方式来加密字段
  3、权限和索机制---最主要的，通过视图来限制对基表的访问，同时对基本的访问时候可以加Access锁

  基于视图的视图访问是极其没有效率的：
  可以试下语句create table tmp as (select * from t03_agreement)，可以发现tmp的主键已经不是t03_agreement的主键了，嵌套视图同样的，当访问嵌套视图时候，第二个视图的索引已经没用了，会造成很大的SKEW，影响数据的选取。（根据这段时间的工作，发现在生产上还是有一些使用到了嵌套视图的。）

5、事务和锁
   事务处理系统的典型特点是具备ACID特征。ACID指的是Atomic(原子的)、Consistent(一致的)、Isolated(隔离的)以及Durable(持续的)，它们代表着事务处理应该具备的四个特征。
   TD的锁这种级别：Access、Read、Write、Exclude锁。其中，DDL语句一般是Exclude锁，Insert、update一般是Write锁，Select一般是read锁。同级的锁可以互相叠加，Exclude锁除外。Access锁可以获得除Exclude锁以外的全部锁，但是可能会造成脏读的现象，视图就是以这种方式为考虑的。（问题分析，DW允许脏读的产生，因为DW多以大数据量的查询为主，比较少进行update的操作，除非是在数据批处理的时候，不像OLTP系统，需要对数据的一致性有很高的要求。）
   TD基于MPP的架构，每个AMP之间是独立的，所以对于同样的一个请求，可能会有不同的执行顺序，没有一个同步的策略，这个时候TD采用的是一个Pesudo表，这个表是每个AMP都公用的，以此来保证同步，每次需要加锁某个对象的时候，就先加Pessudo表，而不是对每个AMP进行加锁。（可以通过Explain来查看得到）
   BT ET跟Muti-statement加锁的不一样：
   1、BT ET是一个语句一个语句加锁，锁进行累积，直到遇到ET的时候才释放锁
   2、muti-statement 是一次性全部加锁
TD对于死锁机制的处理：
   在TD数据库中，每4分钟会检查一次死锁，如果出现死锁的话，就roll back 最后的那个Session
   
Presudo的思考---是不是每次访问都需要使用到Presudo表呢？
   answer:
      2个语句explain select * from table where pi字段 = ? 和explain select * from table where notpi字段= ?
      你会发现，使用pi字段的时候就不会所presudo表，这是因为TD的Presudo表是针对MPP架构设计的，如果是单个AMP操作的时候，其实就没有所谓的同步的现象，所以就不用使用Presodu。
      如上，也就是说单AMP操作不需要presodu表。

对于提交Multi-Statement Requests，将“；”号放在sql语句的前面，多个sql语句同时提交，如下
                  Update Dept 
                        set Salary_Change_Date = Date 
                 ; Update Manager 
                        set Salary_Amt = Salary_Amt *1.08 
                 ; Update Employee 
                        set Salary_Amt = Salary_Amt * 1.04;
    上述语句被放到同一个Session中提交，但是在数据库处理的时候并不是按照顺序一个个下来的，而是随机执行的，同时优化器可能根据判断，来并行的执行语句，所以不能妄图使用Multi-Statement来达到顺序执行的目的。

6、关于DBQLogtbl表
   DBQLogTbl表，是DBC的系统表，FallBack的，双倍存储空间。一般来说如果在数据库中开启了日志记录的话，这个表的空间会增长很快。
   DBQLogTbl里面有很多非常有用的信息。
   比如字段CacheFlag可以得出本sql是否使用了cache，比如说里面有CPU时间，语句执行时间等，可以由此来判断多长时间是等待，多长时间是真正执行，同时也可以日志下对于数据库的sql语句，可以由此来进行调优。
   DBQLogTbl表可以使用的策略如下：
     1、开启数据库的日志记录
     2、每天清理DBQLogTbl表，把数据库传输到一个没有fallback的表中（可以适当的删除些不需要的字段）
     3、分析导出的数据

Tweet

Hash Collision 和 Skew

1.计算每个hash值的记录个数

SELECT                 HASHROW (CHARACTER_COLUMN) 
                        AS "Hash Value"
                        ,COUNT(*)
FROM                DB.TABLE
GROUP BY        1
ORDER BY        2 DESC;

2.计算每个AMP上的记录个数

SELECT                 HASHAMP (HASHBUCKET
                        (HASHROW (CHARACTER_COLUMN))) 
                        AS "AMP #"
                        ,COUNT(*)
FROM                DB.TABLE
GROUP BY        1
ORDER BY        2 DESC; 
3.查找出skew偏移严重的表

SELECT 
 DATABASE,TABLENAME,SUM(CURRENTPERM) AS TableSize,
 MAX(CURRENTPERM) * NUM_OF_AMP  AS VirtualSize,
 100 - AVG(CURRENTPERM)/MAX(CURRENTPERM) * 100 AS TableSkew,
 MAX(CURRENTPERM) * NUM_OF_AMP - SUM(CURRENTPERM) AS SkewSavings
 FROM DBC.TABLESIZE
 WHERE CURRENTPERM > 0
 GROUP BY 1,2,3
 HAVING(SkewSavings > 1000000000 and TableSkew > 10)
 OR TableSkew > 50;

http://blog.chinaunix.net/u3/103331/showart.php?id=2028068

Tweet

TUNING TIPS

BETWEEN比IN有效
IN比EXIST有效
LIKE条件的开头如果不是通配符则优化器使用索引的可能。否则进行全表扫描
DISTINCT很耗费资源（但是和GROUP BY相比如何呢？）
如果可以的话使用UNION ALL而不是UNION
使用多个SQL语句计算不同范围的合计值会导致同一个表的多次扫描，如果将
这些SQL合并，并用CASE区分范围，可以减少扫描次数。这会极大的提高性能
DISTINCT会创建临时表，可以更改SQL的结构避免DISTINCT的使用。
使用派生表而非临时表。减少磁盘IO
避免使用混合类型的表达式。例如字符串和数值比较，TERADATA会将字符串
转换成数值。这会使索引失效并导致全表扫描。结论：混合类型比较时做显示转换
SI
JI
GLOBAL JI
PPI

Tweet

TBL01.A LIKE TBL02.B

SELECT DISTINCT
id_int_org_pty,
nm_ext_org_grp_pty
FROM
VMSL_CUST_MAJ
) TBL01
ON
TBL02.id_int_org_pty = TBL01.id_int_org_pty
WHERE
INDEX(TBL02.nm_ext_org_eng,TBL01.nm_ext_org_grp_pty) > 0

或者


SELECT
AB.A
FROM
AB
LEFT OUTER JOIN
BB
ON
AB.A LIKE '%' || B||'%'

Tweet

EXTRACT

EXTRACT
(
YEAR
MONTH
DAY
HOUR
MINUTE
SECOND
TIMEZONE_HOUR
TIMEZONE_MINUTE
FROM value)

ANSI SQL-2003

Tweet

SSRS一处BUG

设置单元格背景颜色，颜色由行中某单元格内容决定。在本地执行报表，反应正常。部署到服务器上不正常工作。

http://www.sqldev.org/sql-server-reporting-services/background-color-differences-between-visual-studio-amp-reporting-services-39081.shtml

Tweet

SSRS条件语句

switch

=Switch(Fields!Orders.Value < 100,"Red", Fields!Orders.Value > 100, "Green")

iif
=IIF(Fields!LineTotal.Value > 100, True, False)

Expression Examples (Reporting Services)
http://msdn.microsoft.com/en-us/library/ms157328.aspx

Tweet

Teradata Net URL

Portal
https://teradatanet.teradata.com
Web Mail
https://isa.teradata.com/exchange/
ADP
https://portal.globalview.adp.com/teradata

Tweet

JOIN INDEX

A JOIN B
A,B表PI不同的情况下，需要重新分布A或者B。可以考虑在A上建立于B同PI的SINGLE TABLE JOIN INDEX,避免重新分布大量数据。
或者使用于B同PI的中间表，或者VOLATILE表。
总之尽量避免重新分布数据就好

Tweet

Types of Tables in Teradata

1 SET
2 MULTI SET
3 GLOBAL TEMPORARY TABLES
4 TEMPORARY TABLES
5 VOLATILE TEMPORARY TABLES

SET表中如有UPI，则由UPI保证非重复记录

定义字段
ename CHAR(10) CHARACTER SET LATIN NOT CASESPECIFIC
select * from scott.emp where ename(CASESPECIFIC) = 'FAN'
select * from scott.emp_case where ename(not CASESPECIFIC) = 'FAN';
select * from scott.emp_case where upper(ename) = 'FAN';

CREATE GLOBAL TEMPORARY MULTISET TABLE scott.gt_deptsal
(deptno SMALLINT
,avgsal DEC(9,2)
,maxsal DEC(9,2)
,minsal DEC(9,2)
,sumsal DEC(9,2)
,empcnt SMALLINT)
PRIMARY INDEX (deptno)
ON COMMIT PRESERVE ROWS;

查数据字典：select * from dbc.tables where tablename = 'gt_deptsal'
物化表：insert into scott.gt_deptsal values (22,33,33,22,22,22)；
查询物化表：物化后DBC.TempTables 表会存在一条表记录，物化表drop后记录删除：
select * from DBC.TempTables;

本文来自CSDN博客，转载请标明出处：http://blog.csdn.net/fanxiaoliang83/archive/2009/07/30/4395457.aspx

Tweet

IS-NUMERIC FUNCTIONS IN TERADATA

SELECT * FROM TABLE WHERE TRIM(FIELD) NOT BETWEEN '0' AND '9999999999999999'
OR
(UPPER(TRIM(cd_itm)) (CASESPECIFIC) = LOWER(TRIM(cd_itm)) (CASESPECIFIC))

http://forums.teradata.com/forum/analytics/is-numeric-functions

Tweet

AVG OVER

SELECT
nm_bu,
id_cal_mth ,
AVG(qty) OVER (PARTITION BY nm_bu ORDER BY id_cal_mth ROWS 2 PRECEDING)
FROM
(
SELECT
nm_bu,
id_cal_mth,
SUM(qty_itm) AS qty
FROM
vugd_sal_act_mth
GROUP BY
nm_bu,id_cal_mth
WHERE
nm_bu = 'TV'
) AS TBL
ORDER BY
id_cal_mth

Tweet

CONNECT FIELD BY RECURSIVE TABLE

CREATE MULTISET VOLATILE TABLE VTBL_BASE,
NO FALLBACK ,
CHECKSUM = DEFAULT,
NO LOG
AS(
SELECT
id,
name,
ROW_NUMBER() OVER(PARTITION BY id ORDER BY name) AS row_no
FROM
DWHDEV.TBL_TEST_RECURSIVE
) WITH DATA
PRIMARY INDEX
(id)
ON COMMIT PRESERVE ROWS
;

WITH RECURSIVE temp_table (id, name, row_no, depth) AS
(
SELECT
id,
name,
row_no,
depth
FROM
(SELECT DISTINCT
id,
CAST('' AS VARCHAR(1000)) AS name,
0 AS row_no,
0 AS depth
FROM
VTBL_BASE
)tbl

UNION ALL

SELECT
temp_table.id,
temp_table.name || ' ' || base_table.name,
temp_table.row_no + 1 AS row_no,
temp_table.depth + 1 AS newdepth
FROM
temp_table
INNER JOIN
VTBL_BASE AS base_table
ON
temp_table.id = base_table.id
AND
temp_table.row_no + 1 = base_table.row_no
WHERE
newdepth < 10
)

SELECT
id,
name
FROM
temp_table
WHERE
(id, row_no)
IN
(SELECT
id,
MAX(row_no)
FROM
temp_table
GROUP BY
id
)
ORDER BY
id
;

DROP TABLE VTBL_BASE
;

Tweet

RECURSIVE TABLE

WITH RECURSIVE temp_table (id, name, row_no, depth) AS
(
SELECT
id,
name,
row_no,
depth
FROM
(
SELECT DISTINCT
id,
CAST('' AS VARCHAR(2000)) AS name,
0 AS row_no,
0 AS depth
FROM
TBL_FOR_RECURSIVE_TEST
)tbl
UNION ALL
SELECT
temp_table.id,
temp_table.name || ', ' || base_table.name,
temp_table.row_no + 1 AS row_no,
temp_table.depth + 1 AS newdepth
FROM
temp_table
INNER JOIN
TBL_FOR_RECURSIVE_TEST AS base_table
ON
temp_table.id = base_table.id
AND
temp_table.row_no + 1 = base_table.row_no
WHERE
newdepth < 10
)

SELECT
*
FROM
temp_table
WHERE
(id, row_no) IN (SELECT id,MAX(row_no) FROM temp_table GROUP BY id)
ORDER BY
id

Tweet

CASE

CASE WHEN NM_INT_ORG_PTY = 'TOTAL'

THEN SUM(T_TRAN.QTY_ACT) OVER (PARTITION BY M_MST.ID_CAL)

ELSE T_TRAN.QTY_ACT

END AS QTY_ACT

Tweet

CREATE TABLE FROM TABLE

CREATE TABLE TBL_TST AS (SELECT DISTINCT ID_OGO_FRM_INT_ORG_PTY FROM DWHDEV.WK_QDDGD11010) WITH DATA

Tweet

REPLACE VIEW

REPLACE VIEW View_name AS

LOCKING TABLE Table_name FOR ACCESS

SELECT

field1

,field2

FROM

Table_name

Tweet

HAVING

SELECT Sum(qty),itm_id FROM table_name GROUP BY itm_id HAVING SUM(qty) >= 6000000;

Tweet

QUALIFY

SELECT
id_ver
,nm_ver
,dttm_ver_upd
FROM
VUGD_VER_INFO
GROUP BY
id_ver
,nm_ver
,dttm_ver_upd
QUALIFY row_number() over(order by dttm_ver_upd DESC) = '1'

Tweet

COALESCE

COALESCE（Field1,Field2）

Tweet

ADD_MONTHS

SELECT ADD_MONTHS(CAST(id_cal_mth AS DATE FORMAT 'YYYYMM'),-5)(FORMAT 'YYYYMM')(VARCHAR(6))

Tweet

Resnode

exec Resnode ('2009-11-19' ,'2009-11-19' ,'17:55:00','17:57:00')

Tweet

AMP SKEW

SELECT
HASHAMP (HASHBUCKET(HASHROW (
dt_dat_cre ,cd_prod_div ,cd_dpt ,cd_main_acct ,cd_dtl_acct ,cd_sub_acct ,cd_bu ,id_itm ,cd_loc ,cd_crcy
))) AS "AMP NO"
,COUNT(*)
FROM dwhtst.TAPD_BLNC_DAY
GROUP BY 1
ORDER BY 2 DESC;

Tweet

DBQLogTbl

2009-09-15

SQL①　SQL発行件数の分布　ユーザー別

sel

cast(StartTime as char(10)) AS TheDate,

substr( cast( StartTime as char(19) ), 12,3)||

(CASE WHEN substr( cast(StartTime as char(16)), 15,2) between '00' and '29' THEN '00' ELSE '30' END)||':00' AS TheTime,

CASE GROUPING(UserName) WHEN 1 THEN '--ALLUSER--' ELSE Trim(UserName) END AS UserName,

count(distinct QueryText) AS QueryCount

from DBC.DBQLogTbl

WHERE cast(StartTime as char(10)) = '2009-09-15'

GROUP BY GROUPING SETS ((TheDate,TheTime,UserName),(TheDate,TheTime))

order by 1,2,3

;

SQL②　SQL発行件数の分布　全ユーザー

※結果を「10.SQL発行件数の分布」シートの表②に貼り付ける。

sel

cast(StartTime as char(10)) AS TheDate,

substr( cast( StartTime as char(19) ), 12,3)||

(CASE WHEN substr( cast(StartTime as char(16)), 15,2) between '00' and '29' THEN '00' ELSE '30' END)||':00' AS TheTime,

'--ALLUSER--' AS UserName,

count(distinct QueryText) AS QueryCount

from DBC.DBQLogTbl

WHERE cast(StartTime as char(10)) = '2009-09-15'

GROUP BY 1,2,3

order by 1,2,3

;

SQL③　処理時間の長いクエリー

基準時間 > 1200

※結果を「11.処理時間の長いクエリー」シートに貼り付ける。

sel

A.UserName,

A.ExeYMD,

A.StartTime,

A.EndTime,

A.ExeTime,

A.SessionID,

A.QueryID,

T3.C_JOB_ID,

( substr( cast(T3.J_END as char(19)), 12, 2)*3600 + substr(cast( T3.J_END as char(19)), 15, 2)*60 + substr( cast(T3.J_END as char(19)), 18, 2) )

-

( substr( cast(T3.J_ST as char(19)), 12, 2)*3600 + substr(cast( T3.J_ST as char(19)), 15, 2)*60 + substr( cast( T3.J_ST as char(19)), 18, 2) )

AS SHELL_ExeTime

from

(

SELECT

Trim(T1.UserName) UserName,

substr( cast( T1.StartTime as char(19) ), 1,10) AS ExeYMD,

substr( cast( T1.StartTime as char(19) ), 12,8) AS StartTime,

substr( cast( T1.LastRespTime as char(19) ), 12,8) AS EndTime,

(

substr( cast(T1.LastRespTime as char(19)), 12, 2)*3600 + substr(cast( T1.LastRespTime as char(19)), 15, 2)*60 + substr( cast(T1.LastRespTime as char(19)), 18, 2)

)

-

(

substr( cast(T1.StartTime as char(19)), 12, 2)*3600 + substr(cast( T1.StartTime as char(19)), 15, 2)*60 + substr( cast( T1.StartTime as char(19)), 18, 2)

) AS ExeTime,

T1.SessionID,

T1.QueryID,

Cast(T1.TotalCPUTime AS INTEGER) AS TotalCPUTime

FROM DBC.DBQLogTbl T1

LEFT OUTER JOIN DBC.DBQLSqlTbl T2

ON (T1.QueryID=T2.QueryID)

WHERE CAST(T1.StartTime AS CHAR(10))='2009-09-15' AND

ExeTime> 1200

GROUP BY 1,2,3,4,5,6,7,8

) A

LEFT OUTER JOIN DWHRUN.BCH_PGM_LOG T3

ON (

A.ExeYMD = substr( cast( T3.J_ST as char(19) ), 1, 10)

AND

A.StartTime

BETWEEN substr( cast( T3.J_ST as char(19) ), 12, 9) AND substr( cast( T3.J_END as char(19) ), 12, 9)

AND

( substr( cast(T3.J_END as char(19)), 12, 2)*3600 + substr(cast( T3.J_END as char(19)), 15, 2)*60 + substr( cast(T3.J_END as char(19)), 18, 2) )

-

( substr( cast(T3.J_ST as char(19)), 12, 2)*3600 + substr(cast( T3.J_ST as char(19)), 15, 2)*60 + substr( cast( T3.J_ST as char(19)), 18, 2) ) > 900

)

where SHELL_ExeTime-ExeTime between 0 and 100

group by UserName,ExeYMD,StartTime,EndTime,ExeTime,SessionID,QueryID,TotalCPUTime,C_JOB_ID,SHELL_ExeTime

ORDER BY StartTime, SHELL_ExeTime desc

;

Tweet

Data Skew

SELECT

trim(DatabaseName) AS DatabaseName,

trim(TableName) AS TableName,

max(CurrentPerm)*1.00/min(CurrentPerm) AS Vproc_Skew,

sum(CurrentPerm) AS AllVproc_Perm,

max(CurrentPerm) AS MaxPerm_PerVproc,

min(CurrentPerm) AS MinPerm_PerVproc

/*

sum(CASE WHEN Vproc=00 THEN CurrentPerm ELSE 0 END) AS Vproc00_CurrentPerm,

sum(CASE WHEN Vproc=01 THEN CurrentPerm ELSE 0 END) AS Vproc01_CurrentPerm,

sum(CASE WHEN Vproc=02 THEN CurrentPerm ELSE 0 END) AS Vproc02_CurrentPerm,

sum(CASE WHEN Vproc=03 THEN CurrentPerm ELSE 0 END) AS Vproc03_CurrentPerm,

sum(CASE WHEN Vproc=04 THEN CurrentPerm ELSE 0 END) AS Vproc04_CurrentPerm,

sum(CASE WHEN Vproc=05 THEN CurrentPerm ELSE 0 END) AS Vproc05_CurrentPerm,

sum(CASE WHEN Vproc=06 THEN CurrentPerm ELSE 0 END) AS Vproc06_CurrentPerm,

sum(CASE WHEN Vproc=07 THEN CurrentPerm ELSE 0 END) AS Vproc07_CurrentPerm,

sum(CASE WHEN Vproc=08 THEN CurrentPerm ELSE 0 END) AS Vproc08_CurrentPerm,

sum(CASE WHEN Vproc=09 THEN CurrentPerm ELSE 0 END) AS Vproc09_CurrentPerm,

sum(CASE WHEN Vproc=10 THEN CurrentPerm ELSE 0 END) AS Vproc10_CurrentPerm,

sum(CASE WHEN Vproc=11 THEN CurrentPerm ELSE 0 END) AS Vproc11_CurrentPerm,

sum(CASE WHEN Vproc=12 THEN CurrentPerm ELSE 0 END) AS Vproc12_CurrentPerm,

sum(CASE WHEN Vproc=13 THEN CurrentPerm ELSE 0 END) AS Vproc13_CurrentPerm,

sum(CASE WHEN Vproc=14 THEN CurrentPerm ELSE 0 END) AS Vproc14_CurrentPerm,

sum(CASE WHEN Vproc=15 THEN CurrentPerm ELSE 0 END) AS Vproc15_CurrentPerm,

sum(CASE WHEN Vproc=16 THEN CurrentPerm ELSE 0 END) AS Vproc16_CurrentPerm,

sum(CASE WHEN Vproc=17 THEN CurrentPerm ELSE 0 END) AS Vproc17_CurrentPerm,

sum(CASE WHEN Vproc=18 THEN CurrentPerm ELSE 0 END) AS Vproc18_CurrentPerm,

sum(CASE WHEN Vproc=19 THEN CurrentPerm ELSE 0 END) AS Vproc19_CurrentPerm

*/

FROM DBC.TableSize

WHERE DatabaseName in ('etlrun', 'dwhrun')

HAVING AllVproc_Perm > 100000000 AND Vproc_Skew >=1.25

Group by DatabaseName, TableName

Order by DatabaseName, TableName, Vproc_Skew

;

Tweet

DBC.DiskSpace

sel

databasename,

sum(maxperm) AS MAXPERM,

sum(currentperm) AS CURRNETPERM,

sum(currentperm)*1.0/nullifzero(sum(maxperm)) AS USAGE

from

DBC.DiskSpace

group by 1

order by 2 desc

Tweet

Res Usage

SELECT
TheDate (FORMAT'yy/mm/dd', TITLE 'Date'),
TheTime (FORMAT'99:99:99', TITLE 'Time'),
/* Percent of time the CPUs were busy doing work */
SUM(CPUBusy) / SUM(Secs) (FORMAT 'ZZ9', TITLE 'CPU使用率(%)'),
/* Parallel efficiency of total pct of time CPUs were busy */
(AVG(CPUBusy)*100)/NULLIFZERO(MAX(CPUBusy)) (FORMAT 'ZZ9', TITLE 'CPUの並列効率(%)'),
/* Percent of time spent waiting for I/O */
SUM(CPUWaitIO)/SUM(Secs) (FORMAT 'ZZ9', TITLE 'I/O待ち(%)'),
/* Pct logical device I/Os done for paging */
SUM(PageOrSwapIO)*100/NULLIFZERO(SUM(LogicalDeviceIO)) (FORMAT 'ZZ9', TITLE 'I/Oに占めるページング・スワップの割合(%)'),
/* Percent of memory that is unused */
SUM(MemFreeKB)*100/NULLIFZERO(SUM(MemSize(FLOAT))*1024) (FORMAT 'ZZ9', TITLE '空きメモリ(%)'),
/* Percent of memory allocation attempts that failed */
SUM(MemFails)*100/NULLIFZERO(SUM(MemFails+MemTextAllocs+MemVprAllocs)) (FORMAT 'ZZ9', TITLE 'メモリ割り当て失敗(%)')
FROM DBC.ResGeneralInfoView
WHERE TheDate BETWEEN '2009-11-13' AND '2009-11-13'
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 4, 2 ) in ('00', '15', '30', '45') */
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 5, 1 ) in ('0', '5') */
GROUP BY TheDate, TheTime
ORDER BY TheDate, TheTime ;

Tweet

ResGeneralInfoView

対象日付 2009-09-15
利用状況作成SQL


SELECT
TheDate (FORMAT'yy/mm/dd', TITLE 'Date'),
TheTime (FORMAT'99:99:99', TITLE 'Time'),
/* Percent of time the CPUs were busy doing work */
SUM(CPUBusy) / SUM(Secs) (FORMAT 'ZZ9', TITLE 'CPU使用率(%)'),
/* Parallel efficiency of total pct of time CPUs were busy */
(AVG(CPUBusy)*100)/NULLIFZERO(MAX(CPUBusy)) (FORMAT 'ZZ9', TITLE 'CPUの並列効率(%)'),
/* Percent of time spent waiting for I/O */
SUM(CPUWaitIO)/SUM(Secs) (FORMAT 'ZZ9', TITLE 'I/O待ち(%)'),
/* Pct logical device I/Os done for paging */
SUM(PageOrSwapIO)*100/NULLIFZERO(SUM(LogicalDeviceIO)) (FORMAT 'ZZ9', TITLE 'I/Oに占めるページング・スワップの割合(%)'),
/* Percent of memory that is unused */
SUM(MemFreeKB)*100/NULLIFZERO(SUM(MemSize(FLOAT))*1024) (FORMAT 'ZZ9', TITLE '空きメモリ(%)'),
/* Percent of memory allocation attempts that failed */
SUM(MemFails)*100/NULLIFZERO(SUM(MemFails+MemTextAllocs+MemVprAllocs)) (FORMAT 'ZZ9', TITLE 'メモリ割り当て失敗(%)')
FROM
DBC.ResGeneralInfoView
WHERE
TheDate BETWEEN '2009-09-15' AND '2009-09-15'
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 4, 2 ) in ('00', '15', '30', '45') */
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 5, 1 ) in ('0', '5') */
GROUP BY
TheDate, TheTime
ORDER BY
TheDate, TheTime ;
※カラムを物理名で表示する場合は下記SQLを実行する。
SELECT
TheDate (FORMAT'yy/mm/dd', TITLE 'Date'),
TheTime (FORMAT'99:99:99', TITLE 'Time'),
/* Percent of time the CPUs were busy doing work */
SUM(CPUBusy) / SUM(Secs) (FORMAT 'ZZ9', TITLE 'CPU Bsy %'),
/* Parallel efficiency of total pct of time CPUs were busy */
(AVG(CPUBusy)*100)/NULLIFZERO(MAX(CPUBusy)) (FORMAT 'ZZ9', TITLE 'CPU Eff %'),
/* Percent of time spent waiting for I/O */
SUM(CPUWaitIO)/SUM(Secs) (FORMAT 'ZZ9', TITLE 'WIO %'),
/* Pct logical device I/Os done for paging */
SUM(PageOrSwapIO)*100/NULLIFZERO(SUM(LogicalDeviceIO)) (FORMAT 'ZZ9', TITLE 'P+S % of IOs'),
/* Percent of memory that is unused */
SUM(MemFreeKB)*100/NULLIFZERO(SUM(MemSize(FLOAT))*1024) (FORMAT 'ZZ9', TITLE 'Fre Mem %'),
/* Percent of memory allocation attempts that failed */
SUM(MemFails)*100/NULLIFZERO(SUM(MemFails+MemTextAllocs+MemVprAllocs)) (FORMAT 'ZZ9', TITLE 'Mem Fai %')
FROM
DBC.ResGeneralInfoView
WHERE
TheDate BETWEEN '2009-09-15' AND '2009-09-15'
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 4, 2 ) in ('00', '15', '30', '45') */
/* and SUBSTR( cast(TheTime AS VARCHAR(8) ), 5, 1 ) in ('0', '5') */
GROUP BY
TheDate, TheTime
ORDER BY
TheDate, TheTime ;
参考： カラム論理名 カラム物理名 観測内容
CPU使用率(%) CPU Bsy % CPUを使用する割合
I/O待ち(%) WIO % I/Oを待つことで、CPUがアイドル状態になる。
CPUの並列効率性 CPU Eff% 複数のノートでバランスよくCPUを使用している。
I/Oに占めるページング・スワップの割合(%) P+S % of Ios ページング・スワップが発生すると、そのオーバーヘッドによりシステムの性能が低下する。
空きメモリ(%) Fre Mem % メモリが少ないと、資源枯渇となる。
メモリ割り当て失敗(%) Mem Fai % メモリ割り当て失敗すると、再割当するため、オーバーヘッドが発生する。

Tweet

VOLATILE TABLE

CREATE MULTISET VOLATILE TABLE VTBL_01,

NO FALLBACK ,

CHECKSUM = DEFAULT,

NO LOG

AS(

SELECT * FROM ${_table_name2} WHERE nm_tran = 'GDN_C_ITEM_BALANCE'

) WITH DATA

PRIMARY INDEX

(dt_dat_cre ,cd_prod_div )

PARTITION BY RANGE_N(dt_dat_cre BETWEEN DATE '1950-01-01' AND DATE '2100-12-31' EACH INTERVAL '1' DAY );

ON COMMIT PRESERVE ROWS;

Tweet

STATISTICS

Collects statistics on columns, group of columns or indexes.

collect statistics on table_1 index (field_1, field_2)

collect statistics on table_name column field_1;

collect statistics on table_name index(field_1 ,field_2 ,field_3)

help statistics table_name

Tweet

Create/Drop SI

CREATE INDEX SI1(COL1_NM) ON TABLE_NM
DROP INDEX SI_NM ON TABLE_NM

---------------------------------

Hi Farid,

you'll find all the details in the manuals:

Database Design

Chapter 11: Secondary Indexes

Unique Secondary Indexes

Nonunique Secondary Indexes

Chapter 15: Database-Level Capacity Planning Considerations

Sizing a Unique Secondary Index Subtable

Sizing a Nonunique Secondary Index Subtable

There's no DDL for a index subtable, but it's similar to base tables with following definition:

USI:

create unique index USI(i int) on tab;

create table USI(

internal_overhead byte(7),

i int,

baseROWID byte(8) -- or byte(10) for a partitioned table)

unique primary index(i);

NUSI:

create index NUSI(i int) on tab;

create table NUSI(

internal_overhead byte(7),

i int,

baseROWIDa array of byte(8) -- or byte(10) for a partitioned table)

[non-hashed AMP-local primary] index(i);

That kind of NUSI-PI can't be created on a base table, that's why there's no easy way to calculate the exact size of a NUSI.

Dieter

Tweet