Wednesday, May 21, 2014

What is difference between Row Chaning and Row Migration and how to resolve it?

Row migration & Row chaining mainly depends on Orale DB Block Size will be 8K, will ranging from 2K to 32K, which will depend on OS. It occurs at a time when table row data is INSERTED or UPDATED. 

Row migration is typically caused by an UPDATE & Row chaining is typically caused by an INSERT.

Row Migration:

When we performing any Update statement on a table row, if the content of new data is not able to fit in one database block. It means in update of a larger content, overall row length increases and free space of database block is already full.

It replaces content of original block to a pointer address - rowid and entire row is moved to the rowid defined in original block.

Row Migration is caused due to lesser value of PCTFREE for the table.There is not enough place in the BLOCK for UPDATES.

To avoid row migration, table should have PCTFREE value to be set in such a way that there is some enough space within the BLOCKS for UPDATES.

select table_name,chain_cnt,pct_free,pct_used from dba_tables where table_name = table_name ;

alter table table_name pctfree 30;

Row Chaining:

When a row data is too large to fit into a single database block. For example, if your database is of 8K db block size, and we need to insert a row of 20K into it. At this time, Oracle will use 3 db blocks and store the row in pieces.

Below are the reasons at a time when row chaining is occurs:
  • Tables having rowsize exceeds the db blocksize.
  • Tables with LONG and LONG RAW or any other LOB columns.
  • Tables with more then 255 columns.
Oracle internally stores the data in multiple blocks, one block stores all forwarding address - rowids and the data on another two or more blocks.

Row Chaining is UNAVOIDABLE but Row Migration can be MANAGED and resolved through reorganization.Whenever a row migration or row chaining will occur, I/O performance will be decrease as Oracle has to scan more than one data block to retrieve data for the row.

We can identify Row Changing and Row Migration by querying table CHAINED_ROWS, USER_TABLES and V$SYSSTAT or in an ADDM/Statspack report and using ANALYZE command.

Run below command on user to create a system table named - CHAINED_ROWS.

@$ORCL_HOME\ADMIN\utlchain.sql

select owner_name,table_name,count(head_rowid) row_count from CHAINED_ROWS where table_name = table_name GROUP BY owner_name,table_name;

SELECT owner, table_name, chain_cnt , initial_extent, num_rows, blocks, empty_blocks, avg_space, avg_row_len 
FROM user_tables WHERE chain_cnt > 0 and table_name = table_name;

SELECT name, value FROM v$sysstat WHERE name = 'table fetch continued row';

analyze table table_name list chained rows;

select count(*) from chained_rows where table_name = table_name ;


To eliminate row migration / row chaining .

alter table table_name MOVE;

This statement does is to move the table to a new segment. Oracle reads the data from current table and rebuilds the table completely. During this reorganization all migrated rows get fixed.

This causes indexes on the table to be marked UNUSABLE, because it changes the rowids of rows in the table. It generates ORA-01502 error, before REBUILDING of a table .

Do remember that you have to REBUILD all your indexes and repopulate table/index statistics in DB before using the table.

We can perform Import/Export activity as an approach to eliminate the migrated rows from a table.

Friday, May 9, 2014

Please explain how WITH Clause working in a sql query?

An inline view or a query uses subquery in it.
Subquery execution done more efficiently rather by using the query repetadly.
On case by case the performance varies with WITH clause in a sql query.

Below example shows how to reduce repetition of query and simplify complex SQL statements.

SELECT e.ename AS employee_name,
       dc.d_count AS emp_dept_count
FROM   emp e,
       (SELECT deptid, COUNT(*) AS d_count
        FROM   emp
        GROUP BY deptid) dc
WHERE  e.deptid = dc.deptid;

Query will be like as below by using a WITH clause.

WITH d_count AS (
  SELECT deptid, COUNT(*) AS dept_count
  FROM   emp
  GROUP BY deptid)
SELECT e.ename AS employee_name,
       dc.dept_count AS emp_dept_count
FROM   emp e,
       d_count dc
WHERE  e.deptid = dc.deptid;

This way we can make multiple references to the subquery more efficient.

Sometime,WITH clause can be used for a subquery to use as a global temporary table or the query to be use as inline query.

What are Private procedure / functions in a Package ?

A function or procedure in a package which is not declared in package specification, its called as private function / procedure.

Scope of private function /procedure is very limited - on the package itself only.

While in declaring the private procedure/function in package, need to take care that it must be declare first in package declaration part of the package.

create or replace package pack_p1 is
    procedure proc_1;
end pack_p1;

create or replace package body pack_p1 is


    procedure test_proc_1 is

    begin
        dbms_output.put_line('package pack_p1 private procedure');
    end test_proc_1;

    function func_1 is

    begin
        dbms_output.put_line('package pack_p1 private function');
    end func_1;

    procedure proc_1 is

    begin
        dbms_output.put_line('package pack_p1 - proc 1');
        test_proc_1;
    end proc_1;

end pack_p1;

Thursday, May 8, 2014

What to do I have to do when query is too slow, which was earlier running fine?

You need to know why its became slow suddenly. Need to check through following ways, rewrite the query or applying oracle hints on query or using parallellization. So the first step of yours should be to investigate the root cause.

Widely used tools are:
- dbms_profiler
- Explain plan
- SQL*Trace / TKPROF
- Statspack

dbms_profiler:

Use dbms_profiler, if you want to know where time is being spent in PL/SQL code.
Statspack is also a good tool if you are a DBA and you want to check query performance for entire database.

For a single query , explain plan & SQL*Trace & TKPROF are useful tools for query execution.

Explain plan:

in SQL*Plus you have to type:

explain plan for ;
select * from table(dbms_xplan.display);


if you are using older version of plan_table, you have to run the script $ORACLE_HOME\RDBMS\ADMIN\utlxplan.sql.

To get plan table output on serial or parallel execution using below scripts:
$ORACLE_HOME\RDBMS\ADMIN\utlxpls.sql - plan table output with serial execution.
$ORACLE_HOME\RDBMS\ADMIN\utlxplp.sql - plan table output with parallel execution.


The output is shows you what the query optimzer expects.It gives us an idea on choosing correct access path.

SQL*Trace/TKPROF:

Open a new SQL*Plus:
- open a new db session
- alter session set sql_trace true;
- run your query
- disconnect session
- identify your trace file in the server directory as specified in the parameter - user_dump_dest
select value from v$parameter where name = 'user_dump_dest'; 

OR 
Show parameter user_dump_dest;
- on your operating system: tkprof query_op.txt
query_op.txt will now give you valuable information on what has actually happened.

By comparing the output from explain plan with the output from TKPROF, We can able to identify problematic areas of the query.

What actions-factors you need to take care while creating index?

Creation of index on table depends on size of table, volume of data.
If table size is large and we need only few data while selecting data then create index. How frequently column is used in where condition, its also important.

if table has already a primary key, then create a unique key for the required column. Creating a primary key or unique key automatically create unique index on the column.

It is also not advised to create so many indexes as it impacts on query performance of DML statments as every DML statment need to go through with various index and table segments simultaneously.

We can create index or rebulid online also. It enables us to update base tables on the sametime.

creating any index parallelly is not supported while creating and rebuilding index online.

CREATE INDEX idx_emp_mgr ON employee (empid,mgrid) ONLINE;

Monday, January 13, 2014

/*+ APPEND */ hint in serial & parallel mode in oracle.

/*+ APPEND */ hint enables the optimizer to use direct-path insert particularly in insert statements in oracle

In serial mode, direct path can be used only if you include the APPEND hint.
Conventional insert is the default in serial mode. 
It means that NO UNDO will be written. 
UNDO is backed by REDO, so the REDO volume will be accordingly reduced.

Direct path insert is the default in parallel mode. 
In parallel mode, conventional insert can be used only if you specify the /*+ NOAPPEND */ hint.
data is appended to the end of the table,rather than using existing space currently allocated to the table, means after HWM of the current block. 

Direct-path inserts are considerably faster than conventional inserts.

Sunday, December 8, 2013

What are the Bind Variables & Substitution Variables?

Bind Variables are also called host variables, which is accessed even after the PL/SQL block is executed.
It can be create using VARIABLE keyword and referenced with a preceding colon-:.

ex.
VARIABLE sal number;
begin
select salary into :sal from emp where emp_id = 11;
end;
/
select firstname , lastname from emp where salary = :sal;

Substitution Variables are used to get user input at run time to avoid hard-coding of values that we can provide run time also.
We can reference the same with in PL/SQL block with symbol ampersand-&.

VARIABLE sal number;
accept empid prompt 'provide emp id' 
declare
v_empid number := &empid; 
begin
select salary into :sal from emp where emp_id = v_empid;
end;
/
select firstname , lastname from emp where salary = :sal;