Oracle DBA Best Practices

UTOUG Training Days 2004 Advanced DBA Best Practices

Michael S. Abbey The Pythian Group [email protected]

Michael S. Abbey – Abbey – Advanced DBA Best Practices

Agenda • Preamble / raison d'être • Monitoring • INIT.ora • Do not wait for waits • Integrity of your backups

2


Agenda • Preamble / raison d'être • Monitoring • INIT.ora • Do not wait for waits • Integrity of your backups

2


Agenda • • • • •

Infrastructure setup Schema environments Cost-based optimizer Distributed computing Application tuning

3


Agenda • Working orking with with Oracle Oracle Support Services • Hodge podge

4


5

Preamble / raison d'être Cryptic

Too many areas

• It's so hard to tune – Cryptic? – Too many areas?

Memory

– Too complex?

Apps

• It’s It’s so easy to tune – Separate components components – Version compatibility compatibil ity – Concept carry-over

I/O

Michael S. Abbey – Advanced DBA Best Practices

Preamble / raison d'être Facts about the Oracle Server • Terminology is portable across version • Same background processes • Method rather than guesswork – Ripple affect – Fix this / break that

• Instance parameters – v$parameter – v$instance

• SQL statement – v$sqlarea – v$sqltext

• Memory structures – v$librarycache

6


7

Preamble / raison d'être Too complex??

• Tune Apps • Tune Memory • Tune I/O

Turn this


Preamble / raison d'être Facts about the Oracle Server

Into this!!!!!

8


Monitoring

Best practices NOW will payoff down the road

9


Caveat You would not believe what the guy at Training Days told me. He said that his approach to monitoring was to ignore situations that he did not deem to be necessary! One person's approach to monitoring may not be the other person's style. Suggestions? Be my guest.

10


Ensure your space monitoring traps ALL space deficiency situations

11


12

Monitoring The dba_free_space poltergeist select distinct a.tablespace_name from dba_tablespaces a, dba_free_space b where a.tablespace_name = b.tablespace_name;

TABLESPACE_NAME -----------------------------AD_DATA AD_INDEX RBS SYSTEM TEMP

select distinct a.tablespace_name from dba_tablespaces a, dba_free_space b where a.tablespace_name = b.tablespace_name (+);

TABLESPACE_NAME -----------------------------AD_DATA AD_INDEX BLINKY RBS SYSTEM TEMP


Monitoring I'm too full set pages 100 col ts_name form a20 head 'Tablespace' col pieces form 9990 head 'Pcs'

Environment

col ts_size form 999,990 head 'SizeMb' col largestpc form 999,990 head 'LrgMB' col totalfree form 999,990 head 'FreeMb' col pct_free form 990 head '%Free' col whatsused form 999,990 head 'Used' col pct_used form 990 head '%Used' col problem head 'Prob??' spool umcdbp1 . . . spool off

Da code goes here

13


14

Monitoring I'm too full select q2.other_tname ts_name, pieces, ts_size ts_size, nvl(largest_chunk,0) largestpc, nvl(total_free,0) totalfree, nvl(round((total_free/ts_size)*100,2),0) pct_free, ts_size-total_free whatsused, nvl(100-round((total_free/ts_size)*100,2),100) pct_used, decode(nvl(100-round((total_free/ts_size)*100,0),100), 85,'+',86,'+',87,'+',88,'+',89,'++',90,'++',91,'++', 92,'++',93,'++',94,'+++',95,'+++',96,'+++',97,'++++', 98,'+++++',99,'+++++',100,'+++++','') problem from (select dfs.tablespace_name,count(*) pieces, round(max(dfs.bytes)/1024/1024,2) largest_chunk, round(sum(dfs.bytes)/1024/1024,2) total_free from dba_free_space dfs group by tablespace_name) q1, (select tablespace_name other_tname, round(sum(ddf2.bytes)/1024/1024,2) ts_size from dba_data_files ddf2 group by tablespace_name) q2 where q2.other_tname = q1.tablespace_name(+) order by nvl(100-round((total_free/ts_size)*100,0),100) desc;

Code


Monitoring Thresholds

• Percentage based on – Growth patterns – Past experiences – Archival habits

• Ignore or not ignore

• LM tablespaces with extent management local …uniform size … • The dba_free_space poltergeist

– Rollback segments – Temp – Non-app related

Infrastructure issues

More of a than anything else

15


Ensure you trap potential object extension problems before your applications

16


17

Monitoring Unable to extend • Less free space available than potential object extension • Objects within 5 extents of their maximum col col col col col col

owner form a5 head Owner segment_type form a5 head Type segment_name form a24 head Name next_extent form 999,999,990 head NextEXT max_extents form 9,999 head MaxEXT extents form 9,999 head CurrEXT

break on owner on segment_type


Monitoring Unable to extend – available extents prompt prompt Objects that cannot extend ... prompt select from where and

owner,segment_type,segment_name,next_extent sys.dba_segments ds segment_type in ('TABLE','INDEX') next_extent > (select max(bytes) from sys.dba_free_space dfs where dfs.tablespace_name = ds.tablespace_name) order by 1,2,3;

18


Monitoring Unable to extend – within 5 prompt prompt Objects within 5 of max extents ... prompt select from where and

owner,segment_type,segment_name,max_extents,extents sys.dba_segments ds segment_type in ('TABLE','INDEX') max_extents - extents <= 5;

• Avoid maxextents unlimited – will never detect objects with too many extents • Number arbitrary -- based on DBA experiences

19


Ensure you are fluent with monitoring locally managed tablespaces

20


Monitoring Locally managed is what!! • Locally managed [permanent] – Extent management LOCAL in DBA_TABLESPACES – Free space tracked in data files by DBA_FREE_SPACE as well – What’s the deal then with locally managed??

• Locally managed temporary – TEMP segment cleanup not one of Oracle’s ―strengths‖ – Easy way to track space released by segments – Easy way to see progress of TEMP-related operations

21


Monitoring Locally managed temp • Use V$TEMP_FILE and V$TEMP_SPACE_HEADER • Useful to assess cleanup of temporary segments col col col col

trname form a23 head File tablespace_name form a7 head TSName bytes form 9,999,999,990 bytes_free like bytes

22


23

Monitoring Locally managed temp select tsh.tablespace_name, '..'||substr(tf.name,length(name)-23) trname, tf.bytes,tsh.bytes_free from v$tempfile tf, v$temp_space_header tsh where tf.file# = tsh.file_id; TS Name -------LOC_TEMP LOC_TEMP LOC_TEMP LOC_TEMP

File BYTES BYTES_FREE ----------------------- -------------- -------------../od01/loc_temp01.dbf 2,098,200,576 1,887,436,800 ../od02/loc_temp02.dbf 2,098,200,576 1,258,291,200 ../od04/loc_temp03.dbf 2,098,200,576 1,887,436,800 ../od02/loc_temp04.dbf 2,098,200,576 1,887,436,800


Ensure you protect what precious space exists in the SYSTEM tablespace

24


25

Monitoring Outa SYSTEM buddy!! • SYSTEM is the last place you can afford object extension • SOURCE$ and OBJ$ love space select username from dba_users where default_tablespace = 'SYSTEM'; select username from dba_users where temporary_tablespace = 'SYSTEM';

What did your DBA tell you?? I would never use SYSTEM!!


Ensure you are in synch with the status of your rollback segments

26


27

Monitoring Rollback segments • ORA-01552: cannot use system rollback segment for non-system tablespace GRID • ONLINE is the only acceptable status • needs recovery or full are a problem • Assumes your utility segment is in SYSTEM tablespace select from where and

segment_name dba_rollback_segs status <> 'ONLINE' tablespace_name <> 'SYSTEM';

RBS01


Detect problems with your job stream before it plagues your application operations

28


29

Monitoring The job stream

• Of interest to the monitoring exercise – What the job does – How often it runs – When it last ran – When it will run next

• Is the job broken – By Oracle or deliberately

dba_jobs dba_jobs_running


30

Monitoring Job run details select schema_user||','||job||','|| to_char(last_date,'mmdd hh24:mi:ss') from dba_jobs where broken='Y' or (last_date < sysdate - 20/(24*60)

Exclusions – and what <> 'abcdefghi'; – and job not in (210,222,388); – and instr (replace (lower(interval, ' ',null)) <> 'sysdate+1';

INTERVAL


Monitoring Unbreakable

• Broken jobs seem to magically fix themselves • Broken job auto fix (next run) create or replace procedure fj (inter in number) is cursor jobstofix is select job,what from user_jobs where last_date > sysdate-inter/1440 and broken = 'Y'; Smarts to ensure you begin do not un-break what is for jobrec in jobstofix supposed to be broken loop dbms_job.run(jobrec.job); end loop; end; /

31


Ensure unusable index partitions are caught by you, not your apps

32


Monitoring Unusable index partitions

Automagically, index partitions are marked unusable • Aborted direct path Loader sessions • Some partition maintenance operations local • Partition maintenance operations global select 'alter index '||owner||'.'||index_name|| ' rebuild partition '||partition_name||';' from dba_ind_partitions where status= 'UNUSABLE';

33


3 "L" words – locks & latches turn into lousy performance if not detected early

34


Monitoring Locks • Two sessions vying for the same resource • catblock.sql – rdbms/admin • Assortment of cryptic lock views • Narrow down sessions and convert to OS pid using v$session and v$process

v$session paddr v$process addr spid

35


Monitoring Latches

• The DBA's nightmare • Low level serializable mechanisms designed to protect global data structures in the SGA • Life expectancy sub-second • Latches are to memory as locks are to disk

36


Monitoring Latches col Requests form 999,999,999,990 head 'Requests' col PctMiss form 99.90 head 'PCTMiss'

select name,gets+misses Requests, round(misses/decode(gets+misses,0,-1, gets+misses)*100,2) PctMiss from v$latch where misses/decode(gets+misses,0,-1,gets+misses)*100 > 10 order by gets desc;

37


Monitoring Latches

select count(*) from x$kglpn; Absolute  over a certain amount (static)

X:Y

Relative  compared to the norm

38


Monitoring Latches

• A healthy environment returns counts well under 200 • Spinning for latches to gain access to precious resources • Tempting to increase latch counts • Low level supervisory mechanism would not allow

39


Detect spikes in system load that affect performance

40


41

Monitoring System load #!/bin/ksh typeset -i THRESHOLD typeset -i LOAD HOUR=$(date +%H) if (( $HOUR > 7 && $HOUR < 17 )); then THRESHOLD=$1 else THRESHOLD=$2 fi tmpuptime=`uptime` tmploads=${tmpuptime##*average:} LOAD=${tmploads%%.*} if (( $LOAD > $THRESHOLD )); then echo "Load of $LOAD exceeded threshold of $THRESHOLD" exit 1 fi exit 0


Detect devices with low free space

42


Monitoring Disk usage

Some locations are crucial to backups

exports

archived redo logs

#!/bin/ksh # $1 - the filesystem to check # $2 - the byte count threshold typeset -i bytesused typeset -i pctused pctused=$(df -k |grep $1 | awk '{print $5}' | sed 's/%//') bytesused=$(df -k |grep $1 | awk '{print $4}') if (( $bytesused < $2 )) then print $1 is $pctused% full, PROBLEM. fi

43


SQL statement contention is a sibling of latch contention

44


Monitoring SQL statement contention

• Concurrent executions • Location of data • Clustering • Default RS locking mode

45


Monitoring SQL statement contention select 'Too many sessions ( '||to_char(count(*))||' running ' substr(sql_text,1,80) query from v$session s, v$sqlarea sa where s.sql_address = sa.address and s.sql_hash_value = sa.hash_value and s.status = 'ACTIVE' and s.audsid != 0 and sql_text not like 'select sql_text%' group by substr(sql_text,1,80) having count(*) > 100;

46


Why do you think Oracle writes that alert log anyways?

47


Monitoring Alert log • Per instance • Significant errors • Lots of meaningless gobbledygook • Establish a marker – Read on – Set new marker

• Ignore list

00600 03113 01142 01598

00604 06512 07445 01659 01146 01545

Err on the side of the client — if the pages are not necessary, add to the exclude list.

48


Who discards the state of those packages anyways?

49


Monitoring Invalid objects • SYS and SYSTEM – Expect DBMS_ invalid packages – Run dbms xxxx ,sql & prvt xxxx.plb – Interdependencies cause one to invalidate other • DBA_DEPENDENCIES

oracle> grep –il dbms_job *sql a0800150.sql catjobq.sql dbmsjob.sql e0800150.sql statsauto.sql statscusr.sql statspack.sql oracle> sqlplus /nolog SQL> connect / as sysdba . . . SQL> @dbmsjob . . . . . . SQL> @prvtjob.plb

50


Ensure you standby your standby

51


Monitoring Standby database – is it up? But I'm still on 8, what should I do??

typeset -i LINES Replace line export log_history=\$log_history 1 with echo "connect / as sysdba internal, and desc v$log_history get rid of the exit"| sqlplus -s /nolog > recid.log /nolog ! LINES=`grep RECID recid.log|wc -l` if (( $LINES = 0 )) then echo Standby down exit 1 else echo Standby OK exit 0 fi

52


Monitoring Standby database – is it up-to-date? typeset -i CHECKER export log_history=\$log_history CHECKER=`echo "connect / as sysdba set echo off feed off pages 0 select (sysdate-(max(first_time)))*24 from v$log_history;" | sqlplus –s /nolog ` if (( $CHECKER > 1 )) then echo Recovery older than 1 hour exit 1 else echo OK exit 0 fi

53


54

Monitoring Is standby useable? • Do an a regular basis • Weekly, bi-weekly, every 48 hours? • Script and inspect output • Significant output (dual??)

*

echo "/ as sysdba alter database open read only; describe dual;

shutdown immediate startup nomount; alter database mount standby database;" | sqlplus -s > roc.log LINES=`grep DUMMY roc.log|wc -l` if [ $LINES = 1 ] then echo OK exit 0 else echo Problem @standby ... exit 1 fi

Only required for 8 i

*


55

Monitoring Direct inserts on master This is a problem on the standby, so ya'd better watch out!

On master, there is no undo or redo for these inserts!

alter session enable parallel dml;

All transactions propagated from master to standby via archived redo logs. Direct inserts not logged!

alter table mailer nologging; insert into mailer select /*+ parallel (mailer,2) */ * from mailer@hasek; commit;

Beware or else


Monitoring Direct inserts on master • mailer table on master row count = 12,345,781 on July 12 • 77,999,201 rows created in mailer on master on July 12 • Standby database activated due to disaster on July 13

select num_rows from user_tables where table_name = 'MAILER'; NUM_ROWS -----------13889778

Over-simplified, but the gist of the problem.

56


57

Monitoring Detection of nologging activities col tablespace_name form a20 head 'Tablespace' col file_name form a30 head 'File' col tablespace_name form a20 head 'Tablespace' col unrecoverable_time form a20 head 'Change Time' select ddf.file_name,ddf.tablespace_name, vd.unrecoverable_time from sys.dba_data_files ddf,v$datafile vd where ddf.file_name = vd.name and nvl(vd.unrecoverable_time,trunc(sysdate+10)) > to_date('&1','DD-MON-YYYY');

Resolution  Babette


Ensure statistics exist for CBO and they are current

58


Monitoring Table statistics -- Not select from where and or

analyzed owner,table_name sys.dba_tables owner not in ('SYS','SYSTEM') nvl(num_rows,0) = 0 last_analyzed is null;

-- Not select from where and and

analyzed for pre-determined number of days owner,table_name sys.dba_tables owner not in ('SYS','SYSTEM') last_analyzed is not null trunc(last_analyzed) < trunc(sysdate)-5;

59


Make sure YOU are not the bottleneck

60


Monitoring The real story

• dba_ and user_ views are expensive • Learn your way around the x$ tables • With credit to Steve Adams (ixora) – Build x_$ views to match their corresponding x$ views – Grant select to public – Reference these views using sys.{view_name}

61


Monitoring The real story set pages 0 lines 999 trimsp on echo off ver off feed off spool xdollars select 'create view x_$'||substr(table_name,3)|| ' as select * from '||table_name||';' from user_tables where table_name like 'X$%'; spool off set echo on feed on spool xdollars.log @xdollars.lst spool off

62


Monitoring Dollar signs

• Familiarize yourself with the v$ views • Available when database mounted and not open • Column names inconsistent with dba_ counterparts select table_name from dict where table_name like 'V$%' order by 1;

63


64

Monitoring Roadmap to $# select ds.owner, ds.segment_name, ds.partition_name, ds.segment_type, . . . from sys.uet$ e, sys.sys_dba_segs ds, sys.file$ f . . . select /*+ ordered use_nl(e) use_nl(f) */ . . . from sys.sys_dba_segs ds, sys.x$ktfbue e, sys.file$ f DONE


Monitoring sys_dba_segs select u.name, o.name, o.subname, . . . from sys.user$ u, sys.obj$ o, sys.ts$ ts, sys.sys_objects so, sys.seg$ s, sys.file$ f . . . select u.name, un.name, NULL, . . . from sys.user$ u, sys.ts$ ts, sys.undo$ un, sys.seg$ s, sys.file$ f . . . select u.name, to_char(f.file#)||'.'||to_char(s.block#), . . . from sys.user$ u, sys.ts$ ts, sys.seg$ s, sys.file$ f

65


Monitoring This is the end

select select select select select select select select select

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

from from from from from from from from from

sys.tab$ t sys.tabpart$ tp sys.clu$ c sys.ind$ i sys.indpart$ ip sys.lob$ l sys.tabsubpart$ tsp sys.indsubpart$ isp sys.lobfrag$ lf

66


Monitoring Caveat at upgrade time

67


Monitoring So many alerts!!

68


The autoextensible black hole revealed

69


Monitoring Autoextension

• Default for many installations • 2 serious issues when requested extension – over 2Gb limit – cannot be physically accommodated one device

• Oracle is known to not do a very good job of recovering from aborted autoextension • Cleanup after abend not handled well

70


Monitoring Autoextension create table aemon ( file_name sdate bytes insert select from where

varchar2(30), date number);

into aemon file_name,sysdate,bytes dba_data_files autoextensible = 'YES';

71


Monitoring Autoextension select fst.file_name, fst.bytes, fsy.bytes, round((fst.bytes-fsy.bytes)/fsy.bytes *100,2) from aem fsy, aem fst where fst.file_name = fsy.file_name and trunc(fst.sdate) = trunc(sysdate) and trunc(fsy.sdate) = trunc(sysdate-1) and fst.bytes > fsy.bytes;

72


Monitoring Best practices

• Connection testing

• Page and/or email

• Anticipate common problematic areas

• Document via some form of tracking system

• Be thorough and proactive • Email gateway • Share the load

• MetaLink MetaLink MetaLink MetaLink MetaLink MetaLink

73


INIT.ora

Best practices NOW will payoff down the road road

74


INIT.ora • Organization • Compatibility • Shared pool • Ramifications • Undocumented parameters

75


INIT.ora must be clean and readable

76


77

INIT.ora Organization

• Not a fan of ifile • Alphabetical order sort < initumc.ora > sio cp sio initumc.ora rm sio

• Toggles stay in place • Comment changes

• Clean up dbs directory – What is soxx??

Significant defaults – Hard-code – Movements between versions

• Show parameters


INIT.ora Compatibility

• Hard-code – Surprises during upgrades – Are you going far enough

• New features – Can I use them? – Can I go back?

• Access to new features • Changed behaviour of old features

78


Wading through the shared pool

79


INIT.ora Shared pool

• Increasing demands as version increases • CURSOR_SHARING in 8.1.6 is a big deal – Default is EXACT – Preferred is FORCE – Linux 8.1.7 – Alpha 8.1.6

• Bigger is not always better • 4031 errors centre around 4096 – _shared_pool_reserved_min_alloc

at 4000 • Flushing the pool – Manual or transparently – A double-edged sword

• Get yourself in a bind

80


INIT.ora

81


Aware of ramifications when changing values

82


INIT.ora Ramifications of higher values • Semaphores are advisory locking mechanisms that ensure a server completes certain tasks before beginning another. • Some parameter values drain semaphores and others look for larger portions of shared memory DB_CACHE_SIZE LARGE_POOL_SIZE SHARED_POOL_SIZE OPEN_CURSORS PROCESSES

set set set set set set set set set

shmsys:shminfo_shmmax =4294967295 shmsys:shminfo_shmmin =1 shmsys:shminfo_shmmni =100 shmsys:shminfo_shmseg =10 semsys:seminfo_semmni=100 semsys:seminfo_semmsl=500 semsys:seminfo_semmns=500 semsys:seminfo_semopm =1000 semsys:seminfo_semvmx=32767

83


More than meets the eye

84


85

INIT.ora Undocumented parameters x$ksppi Name

x$ksppcv Type

Name

Type

--------- ---------------ADDR RAW(4) INDX NUMBER

----------- -------------ADDR RAW(4) INDX NUMBER

INST_ID KSPPINM

NUMBER VARCHAR2(64)

INST_ID KSPPSTVL

NUMBER VARCHAR2(512)

KSPPITY KSPPDESC

NUMBER VARCHAR2(64)

KSPPSTDF KSPPSTVF

VARCHAR2(9) NUMBER

KSPPIFLG

NUMBER

KSPPSTCMNT

VARCHAR2(255)


86

INIT.ora Undocumented parameters select ksppinm,ksppstvl,ksppdesc from x$ksppi x,x$ksppcv y where x.indx = y.indx and translate(ksppinm,'_','#') like '#%' order by 1 KSPPINSM

KSPPSTVL

KSPPDESC

-------------------------------- --------- --------------------------------_allow_resetlogs_corruption

FALSE

allow resetlogs even if it will cause corruption

_corrupted_rollback_segments

corrupted undo segment list

_db_handles_cached

5

Buffer handles cached each process

_shared_pool_reserved_min_alloc

4400

minimum allocation size in bytes for reserved area of shared pool


87

INIT.ora Best practices • Readable is crucial • Comment changes • Cleanup the dbs directory • Be familiar with memory consumers • Semaphore limitations

• Familiarize yourself with dynamic memory structures in 9i • Block size caches 9i • Watch out for shared_pool_size > 90000000 Issues  Rich


Do not wait for WAITs


88


Are wait situations detracting from overall performance?

89


Do not wait for WAITs • Wait situations are the biggest performance detractor • Latches — protect the integrity of shared memory structures. • Locks — protect the integrity of your data.

• V$ dynamic performance views: – v$waitstat – v$session_wait

• DBA_ views – dba_blockers (catblock.sql)

90


Do not wait for WAITs v$session_wait select event,count(*),sum(seconds_in_wait) siw from v$session_wait group by event; EVENT COUNT(*) SIW --------------------------- -------- --------SQL*Net message from client SQL*Net message to client db file sequential read log file parallel write pmon timer rdbms ipc message smon timer

1192 2 5 1 1 24 1

747,039 0 0 0 9,062 128 908

91


Where has all the CPU gone?

92


Do not wait for WAITs Who's doing what

• Map running SQL to user sessions • Zero in on CPU consumers select sql_text, sid, serial# from v$session s, v$sqlarea sa where s.sql_address = sa.address and s.sql_hash_value = sa.hash_value and s.status = 'ACTIVE' and s.audsid != 0 and sql_text not like 'select sql_text%';

93


Do not wait for WAITs Relationship to system load

• Get a handle on "normal" system load using uptime command • Spool to log file in the crontab, running every 5 minutes */5 /oracle/bin/uptime.sh >> /oracle/logs/uptime.sh.log 2>&1

• High load averages almost always map to SQL*Net message from client wait situations db_file sequential read db file scattered read

94


95

Do not wait for WAITs utllockt.sql set charwidth 17 select lpad(' ',3*(level-1))||waiting_session, waiting_session,lock_type,mode_requested, mode_held,lock_id1,lock_id2 from lock_holders connect by prior waiting_session = holding_session start with holding_session is null; WAITING_SESSION

TYPE MODE REQUESTED

MODE HELD

LOCK ID1 LOCK ID2

----------------- ---- ----------------- ----------------- -------- -------553

NONE None

None

0

0

TX

Share (S)

Exclusive (X)

34888

39

378

RW

Exclusive (X)

S/Row-X (SSX)

33255666

2

3928

RW

Exclusive (X)

S/Row-X (SSX)

3255666

2

213


96

Do not wait for waits Best practices • Waiting drains system resources

• SQL statement contention

• Wait events cause destructive spinning

• Concurrent user requests for the same data in the cache

• Minimizing potential waits foremost in your minds

• DO not exploit multitasking

• Commit often Issues  Rich


Integrity of your backups


97


Integrity of your backups Media recovery enabled (archivelog mode)

5. INIT.ora

1. Image backups

7. tnsnames.ora

2. Export – No indexes / No constraints – No rows

3. Control file

6. listener.ora

Assume these are run nightly

98


Backup is nada without integrity and recovery testing

99


Integrity of your backups Image backups

Best case scenario • Separate server to restore image • Deliberate complete media recovery – recover database . . . – recover datafile . . . – recover tablespace . . .

• Deliberate incomplete recovery at the database level

100



• Mock disaster recovery • Ideally on the same machine – With no downtime, involves an ORACLE_SID change • using backup controlfile is not the Real McCoy

– Plan during a window of opportunity

• Make sure you have performed the recovery portion of your image is NOT tested

101



• NFS device is no substitute for offline storage but … – Mount points all over the corporate server environment – Target of all images and exports

• Run the NFS out to secondary storage • root user buy-in

102


Tested rebuild from full database exports

103


104

Integrity of your backups Test rebuild from full database export

• Create the database (see Infrastructure section)

in

• Run 3 or 4 full database imports 1

Just definitions NOT data

4

inctype=system

rows=n ignore=y

online rollback

2

segments just built

3

inctype=restore

rows=n ignore=y


Standby assisting checkup of recovery process

105


Integrity of your backups Testing using standby technology

• Shutdown your standby • Re-instantiate (hot  /u02/hot) set pages 0 trimsp on lines 999 echo off feed off select 'scp /u02/hot'||file_name|| ' [email protected]:'||file_name from dba_data_files; set pages 0 trimsp on lines 999 echo off feed off select 'gzip –d '||file_name||'.gz' from dba_data_files;

106


Integrity of your backups Testing using standby technology

• Re-get standby control file • Copy to appropriate control_files = locations alter database create standby controlfile as 'standby.ctl';

set pages 0 trimsp on lines 999 echo off feed off select 'cp standby.ctl '||name from v$controlfile;

107


Data integrity of your exports

108


109

Integrity of your backups How's export written

• Mismatch between UNIX environment and Server character set

Wake-up

• Determined by NLS_LANG Export: Release 9.2.0.5 - Production on Sat Sep 31 16:47:52 2004 (c) Copyright 2001 Oracle Corporation.

All rights reserved.

Connected to: Oracle9i Enterprise Edition Release 9.2.0.5 - Production With the Partitioning option JServer Release 9.2.0.5 - Production Export done in WE8DEC character set and AL16UTF16 NCHAR character set server uses US7ASCII character set (possible charset conversion)


Integrity of your backups Character set conversion

Collège Brébeuf Frhre Norman: Mrs. Jelinski and I were very pleased with our visit to Collhge Bribeuf, and are pleased with what we saw. We are especially happy that we see eye to eye on the preservation of the native Francophone culture with something as simple as accent retention in our Oracle9 i database!

110


Compression checking your export/import engine

111


Integrity of your backups Gee, zip? rm export.pipe > /dev/null 2>&1 mkfifo export.pipe gzip < export.pipe > fulldb.dmp.gz & sleep 2 exp parfile= fulldb.parfile

Usable in its compressed state? 8i

userid=/ full=y log=fulldb indexes=n constraints=n

triggers=n

buffer=5000000 triggers=n file=export.pipe

9i

112


Integrity of your backups Gee, unzip? rm import.pipe > /dev/null 2>&1 mkfifo import.pipe gunzip < fulldb.dmp.gz > import.pipe & sleep 2 imp userid=/ file=import.pipe full=y log=fulldb_in buffer=50000000

Better find out sooner than during a disaster recovery!! IMP-00037: Character set marker unknown IMP-00000: Import terminated unsuccessfully IMP-00009: abnormal end of export file IMP-00000: Import terminated unsuccessfully

113


Bottom- line … can I read that export file?

114


Integrity of your backups Read this, import! imp userid=/ full=y indexfile=fulldb.sql log=fulldb_if Import: Release 9.2.0.5 - Production on Sat Jun 30 22:13:49 2004 (c) Copyright 2001 Oracle Corporation. All rights reserved. Connected to: Oracle9i Enterprise Edition Release 9.2.0.5 - Production With the Partitioning option JServer Release 9.2.0.5 - Production Export file created by EXPORT:V09.20.00.05 via conventional path import done in US7ASCII character set and AL16UTF16 NCHAR character set Import terminated successfully without warnings.

115


Integrity of your backups Best practices • Only 10% of the way there without testing recovery

• Missing control file(s)

• Full and partial components of database

• Test clone on another server

• Missing archived redo logs

• Test rebuild from export

• Role played by your standby

116


Infrastructure setup The

Oracle 9i


117


Infrastructure setup A word on creating a database I've been an Oracle DBA for 2,000 years (US) and have never created a database; why start now?? I always upgrade

No thanks, sonny, I went to a user group meeting. As soon as I got back to the office I started creating databases. I learned SO MUCH!! You ever tried it?? You need the support N more than me!!

Please sit

o

118


Infrastructure setup Creating the database

• 9i Oracle Managed files by specifying – db_create_file_dest – db_create_online_log_destn – undo_management = auto – undo_tablespace = rollback – undo_retention = 2000 create database umc undo tablespace rollback default temporary tablespace temp;

• Do it yourself – Create database • • • • • •

controlfile character set maxlogfiles maxloghistory maxlogmembers maxdatafiles

– Datafile – Log file(s)

119


Step-by-step database creation; the players

120


Infrastructure setup Creating the database yourself

• The SYSTEM tablespace – systemnn.dbf ~ 250m – Set pctincrease 1 (later)

• Redo log groups – At least dual twomembered groups – logmn_gn.ora (member number / group number)

• Construct first nonSYSTEM rollback segment – Acquire 20 extents of ~ 100k each

• Run admin scripts • Test creation

121


Infrastructure setup Create database – phase 1 startup nomount create database umc controlfile reuse maxlogfiles 16 maxlogmembers 2 maxdatafiles 2048 maxloghistory 1000 character set we8iso8859p1 datafile '/data01/oracle/umc/dbs1_umc.dbf' size 250m logfile group 1 ('/redo01/umc/log1_g1.dbf', '/redo02/umc/log2_g1.dbf') size 200m, group 2 ('/redo02/umc/log1_g2.dbf', '/redo01/umc/log2_g2.dbf') size 200m;

122


123

Infrastructure setup Create database (1) – important points 1. For this or create controlfile 2. Will not error out if files already exist 3. Make an artificially large # 4. Do not allow to default – WE good place to start 5. No smaller 6. 2 dual membered groups

1

startup nomount create database umc controlfile reuse

2

maxlogfiles 16 maxlogmembers 2 maxdatafiles 30

3

maxloghistory 1000 character set

4

datafile '/data01/…' size 250m logfile group 1 ('…1_g1.dbf',

5 6

'…2_g1.dbf') size 200m, group 2 ('…1_g2.dbf', '…2_g2.dbf') size 200m;


Infrastructure setup A word on sizing online redo logs

Two choices:

1. Small online redo logs to minimize time to push to standby 2. Customary large redo logs with forced log switches

124


Infrastructure setup Create database – phase 2 shutdown startup alter tablespace system default storage (pctincrease 1); create rollback segment rb_temp tablespace system storage (initial 50k next 50k minextents 20 maxextents 20); alter rollback segment rb_temp online;

125


126

Infrastructure setup Create database (2) – important points 1. No more to be done nomount – rest of work requires instance to be open.

shutdown

1

startup alter tablespace system default storage (pctincrease 1);

2

create rollback segment rb_temp

3

tablespace system storage (initial 50k

2. Permit pmon to automatically coalesce free space in SYSTEM 3. First non-SYSTEM rollback segment (~1Mb)

next 50k minextents 20 maxextents 20);

alter rollback segment rb_temp online;

4. Rollback segments created OFFLINE by default; edit INIT.ora at the same time so you do not forget

4


Infrastructure setup Create database – phase 3 @?/rdbms/admin/catalog

Build the data dictionary

@?/rdbms/admin/catproc

Install PL/SQL

@?/rdbms/admin/catrep

Replication

@?/rdbms/admin/catsnap

Snapshot specifics

@?/rdbms/admin/dbmsutil

Popular utility packages

@?/rdbms/admin/prvtutil.plb @?/rdbms/admin/dbmssql @?/rdbms/admin/prvtsql.plb

"" SQL utilities ""

127


128

Infrastructure setup Create database – phase 4 Right option(s) Version numbering EE/Standard Expected feedback


Infrastructure setup

Database is created Data dictionary has been populated First non-system rollback segment exists and is online

129


Infrastructure setup pupbld.sql

130


Fluency in rollback segment setup

131


Infrastructure setup Rollback segments • 4 to 6 equally-sized segments

• Large segment for large transactions

• minextents and maxextents the same at creation time

– 2 to 4Gb

• 500k to 2Mb extents

– Cap at creation time

• Not in locally managed tablespaces • Mentioned in INIT.ora

– Larger extent size • No guarantee unless the only one online

132


133

Infrastructure setup Rollback segment tablespace(s)

• Dedicated tablespace • Appropriately named

File names contain tablespace name.

create tablespace rollback datafile '/u3/oradata/umc/ rollback01.dbf' size 1024m, '/u5/oradata/umc/ rollback02.dbf' size 1024m, '/u5/oradata/umc/ rollback03.dbf' size 1024m, '/u5/oradata/umc/ rollback04.dbf' size 1024m;


Infrastructure setup Rollback segment creation create rollback segment rbs01 tablespace storage (initial 2m next 2m minextents 499 maxextents 499); create rollback segment rbs02 tablespace storage (initial 2m next 2m minextents 499 maxextents 499); create rollback segment rbs03 tablespace storage (initial 2m next 2m minextents 499 maxextents 499); create rollback segment rbs04 tablespace storage (initial 2m next 2m minextents 499 maxextents 499);

rollback

rollback

rollback

rollback

ROLLBACK_SEGMENTS = (rbs01,rbs02,rbs03,rbs04)

134


Empowering the techies with do-it-yourself

135


136

Infrastructure setup v$dba • v$ dynamic performance views • Over 200 with 9i – most accessible when database mounted • DBA_ dictionary views similar to their USER_ counterparts with ownership column • There's nothing to hide • Grant access but no public synonyms

Mother may I ??


Infrastructure setup v$dba connect / as sysdba set echo off pages 0 lines 999 trimsp on feed off spool veedba select 'grant select on '||view_name||' to public;' from user_views where view_name like 'DBA%' or view_name like 'V_$%';

• Often used here and there by applications • V$ public synonyms probably exist • Granting must be done on view, not public synonym for V$ to avoid ORA-02030: can only select from fixed tables/views

137


Fluency in locally managed

138


Infrastructure setup Locally managed tablespaces • Different syntax than DM counterparts • Temporary files end up in DBA_TEMP_FILES • Free space bitmap in datafile headers • Reduced recursive I/O

• Extent management local – autoallocate turns Oracle loose with extent sizing • Objects of differing sizes • Lots of different sized extents

– uniform size {} defaults to 1Mb

139


Infrastructure setup Locally managed tablespaces create tablespace loc_geo_index datafile '/u01/oradata/loc_geo_index01.dbf' size 2000m, '/u04/oradata/loc_geo_index01.dbf' size 2000m extent management local autoallocate ;

create tablespace loc_geo_index datafile '/u01/oradata/loc_geo_index01.dbf' size 2000m, '/u04/oradata/loc_geo_index01.dbf' size 2000m extent management local uniform size 20m;

140


Infrastructure setup LM temp tablespace create temporary tablespace loc_temp datafile '/u01/oradata/loc_temp01.dbf' size 2000m, '/u04/oradata/loc_temp02.dbf' size 2000m extent management local uniform size 20m;

• Keep extents in TEMP below 1000 • Space allocated then controls uniform size • Great deal of the work done at startup involves reconciliation and transfer of block ids from free to unallocated ( fet$ to uet$)

141


Infrastructure setup Best practices • Make a point of starting from the beginning (create database)

• Practice practice practice

• Pay attention to each piece in the puzzle

• Know your friendly neighbourhood admin scripts

• Completeness checking at the end

• Do temp and rollback right

142


Schema Environments


143


Establish and follow guidelines

144


Schema environments Guidelines • Thou shalt not give user an overrich environment • Thou shalt not sort or default to SYSTEM • Thou shalt occupy space in appropriately-named tablespaces • Thou shalt not have DBA, CONNECT, or RESOURCE

145


Smart when setting up your user environments

146


147

Schema environments User creation • Creating a user • Temporary tablespace • Use O/S authorization for all host-based users • Connection and privileges depend on function

create user sample identified by sx_p0 temporary tablespace loc_temp; alter user sample identified externally;

-- Jack and Jill users grant create session to jj_user;


Schema environments Application data users • Give necessary privileges and no more grant create table, create view, create procedure, create trigger, create synonym, create snapshot, create database link to sample;

• Who has more than they need?? • Who has been given DBA • Are the any privileges given out correctly?

148


Schema environments Who has more than they need?? select grantee, privilege -- SAMPLE is a schema from sys.dba_sys_privs -- deliberately set up to where privilege not in -- set the standard (select privilege from sys.dba_sys_privs where grantee = 'SAMPLE') UNION select grantee, granted_role from sys.dba_role_privs where granted_role in ('DBA','CONNECT', 'RESOURCE','EXP_FULL_DATABASE', 'IMP_FULL_DATABASE') order by 1;

149


Schema environments Who has more than they need?? GRANTEE -----------------ABRAMSON ABRAMSOM JACKSON JACKSON JONES REDMOND THOMSON THOMSON THOMSON UQUART UQUART

PRIVILEGE ----------------------DBA DROP PUBLIC SYNONYM CONNECT RESOURCE DBA SELECT ANY TABLE DBA INSERT ANY TABLE UPDATE ANY TABLE DBA UNLIMITED TABLESPACES

150


Adopt conventions that make sense

151


Schema environments Naming conventions

• Plan for the future • Picture yourself at 4:15 am weeding through cryptic names • Bind tablespace names to owner USERNAME • Marry data and index tablespace names

152


Schema environments Naming tablespaces

• Based on USERNAME create tablespace DELIVERY datafile -- Data segments '/u01/oracle/oradata/delivery01.dbf' size 1024m, '/u05/oracle/oradata/delivery02.dbf' size 1024m default storage (initial 100m next 100, pctincrease 0);

• Based on function (data or index) create tablespace DELIVERYX datafile -- Index segments '/u02/oracle/oradata/deliveryx01.dbf' size 1024m, '/u03/oracle/oradata/deliveryx02.dbf' size 1024m default storage (initial 100m next 100, pctincrease 0);

153


Schema environments Naming tablespaces - violators

Define rules first, implement, and track – Data segment names = USERNAME – Index segment names = USERNAME with X – Partitioned objects = USERNAME followed by underscore ~ X for index partitions ~ PTS for partitioned tablespace select from where or

distinct tablespace_name sys.dba_segments tablespace_name not in (owner,owner||'X') (tablespace_name not in ('%'||owner||'%X%'||'%PTS%') or tablespace_name not in ('%'||owner||'%PTS%'));

154


Schema environments Naming partitions • Object name • Index non-unique or unique indicator + sequential number • Underscore separators • Pnnnn forcing fixedlength names for sorts • Abbreviate carefully

• Imbed hp for hashpartitioned objects – Affects syntax for maintenance – Affects how operations are carried out

• Fix names after merges and splits

155


Schema environments Data partitions create table mail (originator varchar2(40), created date, . . . . . . read varchar2(1)) storage (initial 200m next 200m pctincrease 0) partition by range (created) (partition post_mail_p0001 values less than (to_date('01-JAN-2001','DD-MON-YYYY')) tablespace post_mail_pts_0001, partition post_mail_p0002 values less than (to_date('01-JUL-2001','DD-MON-YYYY')) tablespace post_mail_pts_0002, partition post_mail_p0003 values less than (to_date('01-JAN-2002','DD-MON-YYYY')) tablespace post_mail_pts_0003, . . . . . . partition post_mailing_pmax values less than (maxvalue) tablespace post_mail_pts_max);

156


Schema environments Index partitions create index mail_n1 on mail (originator) (partition post_mail_n1_p0001 values less than tablespace post_mailx_pts_0001, partition post_mail_n1_p0002 values less than tablespace post_mailx_pts_0002, partition post_mail_n1_p0003 values less than tablespace post_mailx_pts_0003, . . . . . . partition post_mail_n1_pmax values less than (maxvalue) tablespace post_mailx_pts_max);

157


Schema environments Naming indexes

Table name Abbreviation(s) for column(s) in index

Table name N for non-unique indexes U for unique indexes Separate series for each type 2 schools of thought

158


Key index decisions

159


Schema environments Unique index vs. constraint

• Primary key constraints – Always name with _PK suffix – With using index are the same as unique indexes – Allow foreign key references alter table mailer add constraint mailer_pk primary key (id) using index storage (initial 300m next 300m pctincrease 0) tablespace loc_mail_indx;

160


Schema environments Best practices • Users allowed to do what they need and nothing more • Segregate users by function • Adopt and follow naming conventions

• Trap and report on violators • 30 character limit – Abbreviate smartly – Abbreviate consistently – Ensure all players are aware

161


Distributed computing


162


Manage (not mis-manage) your replication environment

163


Distributed computing Snapshots are a snap

• • • • •

Use same name as master table Give them storage parameters just like tables Negotiate refresh interval Ensure there are no premature refreshes Existing refresh intervals . . .

select job,what,interval from user_jobs where lower(what) like 'dbms_refresh%';

164


Distributed computing Indexes on snapshots

• Primary key snapshots preferred • ROWID (à la Oracle7) still supported in 9i • PK constraint created with create snapshot • Check for secondary indexes on snapshots by referring back to master • Do not check and forget — do so on a regular basis

165


Distributed computing Indexes on snapshots On master create snapshot log on vendor_prod storage (initial 2m next 2m pctincrease 0) tablespace sn_logs;

On remote

snapshot + secondary indexes

create snapshot vendor_prod refresh fast start with sysdate next sysdate+10/1440 as select * from vendor_prod@product tablespace loc_snaps;

166


Thoroughness of indexing your replicated objects

167


Distributed computing Indexes on snapshots —completeness check col inm form a10 head Index col column_name form a7 head Column col column_position form 999 head Pos break on inm select from where minus select from where order

index_name inm,column_position,column_name user_ind_columns table_name = upper('&1') index_name inm,column_position,column_name user_ind_columns@product table_name = upper('&1') by 1,2;

168


Many many listeners

169


Distributed computing Some port with that?

• Use different ports for different connection requests • Separate by source – By web server – By connection type ( jvm, cgi)

• Speak with the SA before choosing ports • Must be reflected on client(s)

170


171

Distributed computing Multiple ports / multiple listeners LISTENER1521 = (DESCRIPTION_LIST = (DESCRIPTION = (ADDRESS_LIST = (ADDRESS = (PROTOCOL = IPC)(KEY = EXTPROC))) (ADDRESS_LIST = (ADDRESS = (PROTOCOL = TCP)(HOST = mofo1)(PORT = 1521))))) LISTENER1522 = (DESCRIPTION_LIST = (DESCRIPTION =

tnsnames.ora aliases must reflect right port #s

(ADDRESS_LIST = (ADDRESS = (PROTOCOL = IPC)(KEY = EXTPROC))) (ADDRESS_LIST = (ADDRESS = (PROTOCOL = TCP)(HOST = mofo1)(PORT = 1522)))))


172

Distributed computing Best practices • Know how to replicate • Fundamental to the Internet environment • Offloading of processing crucial to life expectancy of sites with very large user community

• Completeness in your replication environment • Multiple listeners • Listener per snapshot (if enough to go around) • Fluency with DBMS_JOB package


Cost-based optimizer

Best practices NOW will payoff down the road road

173


174

Cost-based optimizer DBA's best-friend • Collecting of statistics like – Number of rows

• Frequency decisions – Consult IT personnel C B

– Average row length – Distinct column values – High value / low value

• Run unattended periodically from crontab

O

– Stale statistics more harm than good

• Using SQL*Plus analyze command OR PL/SQL packages • Inspect last_analyzed


Decisions / approaches

175


176

Cost-based optimizer Setup • analyze is not going any where (yet) with 9i and 10g • Perform from a central user • Preferably O/S authenticated [externally]

create user analyzer identified by pw; C alter user analyzer identified externally; B grant create session to analyzer; grant analyze any to analyzer; O

We always identify externally


177

Cost-based optimizer How often

• Size based – Pick a size – Treat those above threshold differently

• Usage based – Track most frequently changed objects – Exclude hot objects / treat differently


Ensure you begin at the beginning

178


Cost-based optimizer In the beginning

• Delete statistics EVERYWHERE • Flip or set entry in INIT.ora • Liaise with user community • Pick 2 or 3 days a week • Assess run time • Assess impact on system

179


180

Cost-based optimizer Deleting statistics set echo off pages 0 trimsp on lines 999 spool stat_del.sql select 'exec dbms_stats.delete_schema_stats (ownname=>'|| ''''||username||''''||')' from sys.dba_users where username not in ('SYS','SYSTEM'); spool off set echo on feed on spool stat_del @stat_del spool off


Cost-based optimizer Method

• Tables – Estimate sample {2-10} percent – Populates USER_TABLES

• Indexes – Compute – Populates USER_IND_COLUMNS

• Nothing short of a windfall

181


Always be aware of your options

182


Cost-based optimizer Using PL/SQL procedure PROCEDURE GATHER_SCHEMA_STATS Argument Name Type ----------------- ---------OWNNAME VARCHAR2 ESTIMATE_PERCENT NUMBER BLOCK_SAMPLE BOOLEAN METHOD_OPT VARCHAR2 DEGREE NUMBER GRANULARITY VARCHAR2 CASCADE BOOLEAN STATTAB VARCHAR2 STATID VARCHAR2 OPTIONS VARCHAR2 OBJLIST DBMS_STATS STATOWN VARCHAR2

In/Out -----IN IN IN IN IN IN IN IN IN IN OUT IN

Default? -------DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT DEFAULT

183


Four-part approach 1. Initialization parameters 2. Place objects in monitoring mode 3. Place a stake in the ground for each schema 4. Schedule regular statistic collection I. Do not use for Oracle Applications II. Do not use for third party apps – liaise with vendors

184


1 – INIT.ora

185


2 - Turn on monitoring set pages 0 lines 999 trimsp on feed off spool monon select 'alter table '||owner||'.'||table_name|| ' monitoring;' from sys.dba_tables where owner not in ('SYS','SYSTEM') and nvl(duration,'X') <> 'SYS$TRANSACTION' and nvl(iot_type,'X') <> 'IOT_OVERFLOW' and nvl(temporary,'X') <> 'Y' and monitoring <> 'YES'; spool off

186


187

3 – Stake in the ground set lines 999 trimsp on pages 0 spool sitg select unique 'exec dbms_stats.gather_schema_stats (ownname=>'|| ''''||owner||''''||',estimate_percent=>2,'|| 'block_sample=>false,method_opt=>'|| ''''||'for all indexed columns size 1'|| ''''||',degree=>8,granularity=>'||''''||'ALL'||''''|| ',cascade=>true);' from sys.dba_tables where owner not in ('SYS','SYSTEM'); spool off


4 – Schedule regular collection • turn on monitoring – for tables not already in monitoring mode

• gather empty – new tables created since last run – tables inadvertently taken out of monitoring mode

• gather stale – those that have changed >= 10%

188


4.1 – Turn on monitoring set pages 0 lines 999 trimsp on feed off spool monon select 'alter table '||owner||'.'||table_name|| ' monitoring;' from sys.dba_tables where owner not in ('SYS','SYSTEM') and nvl(duration,'X') <> 'SYS$TRANSACTION' and nvl(iot_type,'X') <> 'IOT_OVERFLOW' and nvl(temporary,'X') <> 'Y' and monitoring <> 'YES'; spool off

189


190

4.2 – gather empty select unique 'exec dbms_stats.gather_schema_stats '|| '(ownname=>'||''''||owner||''''|| ',estimate_percent=>1,block_sample=>false,'|| 'method_opt=>'||''''|| 'for all indexed'|| ' columns'||' size 1'||''''||',options=>'|| ''''||'gather empty'||''''||',degree=>8,'|| 'granularity=>'||''''||'ALL'||''''|| ',cascade=>true);' from sys.dba_tables where owner not in ('SYS','SYSTEM');


191

4.3 – gather stale select unique 'exec dbms_stats.gather_schema_stats '|| '(ownname=>'||''''||owner||''''|| ',estimate_percent=>1,block_sample=>false,'|| 'method_opt=>'||''''|| 'for all indexed'|| ' columns'||' size 1'||''''||',options=>'|| ''''||'gather stale'||''''||',degree=>8,'|| 'granularity=>'||''''||'ALL'||''''|| ',cascade=>true);' from sys.dba_tables where owner not in ('SYS','SYSTEM');


Rationalize the approach break on owner on table_name set pages 70 col table_name form a24 col partition_name form a24 col owner form a16 col pct form 999,999.9 select a.owner,a.table_name,b.partition_name, a.num_rows, (inserts+updates+deletes)/num_rows*100 pct, last_analyzed from sys.dba_tables a,sys.dba_tab_modifications b where a.table_name = b.table_name and a.owner = b.table_owner and (inserts+updates+deletes)/num_rows > .1 and nvl(num_rows,0) <> 0 order by 1,2

192


Tried, tested, and true - analyze (but not recommended unless you have no choice)

193


Cost-based optimizer Using analyze set echo off pages 0 feed off lines 999 trimsp on spool anaall.sql {analyze tables and indexes code} spool off set echo on timi on feed on spool anaall @anaall spool off

Put the pieces together over the next 3 slides

194


Cost-based optimizer Using analyze on non-partitioned tables

select 'analyze table '||owner||'.'||table_name|| ' estimate statistics for table sample '|| '20 percent;' from sys.dba_tables where owner not in ('SYS','SYSTEM') and partitioned = 'NO' order by owner,table_name;

195


Cost-based optimizer Using analyze on non-partitioned indexes select 'analyze index '||owner||'.'|| index_name|| ' compute statistics;' from sys.dba_indexes where owner not in ('SYS','SYSTEM') and partitioned = 'NO' order by owner,index_name;

196


Cost-based optimizer Using analyze on partitioned tables select 'analyze table '||table_owner||'.'|| '.'||table_name|| ' partition ('||partition_name||')'|| ' estimate statistics sample 2 percent;' from sys.dba_tab_partitions where table_owner not in ('SYS','SYSTEM') order by table_owner,table_name,partition_name;

197


Cost-based optimizer Using analyze on partitioned indexes select 'analyze index '||index_owner|| '.'||index_name|| ' partition ('||partition_name||')'|| ' compute statistics;' from sys.dba_ind_partitions where index_owner not in ('SYS','SYSTEM') order by index_owner,index_name,partition_name;

198


Not interfering with replication

199


Cost-based optimizer Exclusions with snapshots and (owner,table_name) not in (select distinct log_owner,master from sys.dba_snapshot_logs union select log_owner,log_table from sys.dba_snapshot_logs) and (owner,index_name) not in (select a.owner,a.index_name from sys.dba_indexes a, sys.dba_snapshot_logs b where a.table_name = b.master)

200


201

Cost-based optimizer Exclusions with snapshots

• Analyzing collides with snapshot refreshes • Restrictive lock inhibits job execution • System load can skyrocket • Horrific latch contention 6:57pm up 6 day(s), 7:12, 9 users, load average: 1.06, 1.08, 1.10

normal during analyze 4:13am up 8 day(s), 7:12, 9 users, load average: 341.06, 309.08, 301.10


Stuff just too big to analyze

202


Cost-based optimizer Exclusions – large objects

• DBA nightmare • exclude based on row count • gather stale takes care of most very big tables • if still required – store large table names somwhere – join with dba_tables when building calls to DBMS_STATS

203


204

Cost-based optimizer Implementation without study

XDBA

Hastily implemented collection plan


Cost-based optimizer Best practices • Know your options

• The whole schema

• Adopt an approach

• Ensure new objects are analyzed

• Liaise with consumer • No cowboy implementation • Staleness of statistics

– Whose responsibility – Step in promotion of code and dependent objects

• ONLY option

205


Application tuning


206


Allow everyone to EXPLAIN themselves

207


Application tuning plustrce role • Create at once when setting up a new instance or inheriting an old one • plustrce.sql as SYS from ?/sqlplus/admin • PLAN_TABLE

beforehand using ?/rdbms/admin/utlxplan

• There's nothing to hide • Access to dynamic performance tables used for – v_$sessstat – v_$statname – v_$session

• Give access to all

208


AUTOTRACE and

beyond

209


210

Application tuning Empowering the developers • Want to write good code • Arm them with the tools • Imbed testing into coding – education fosters growth

set autot trace exp SP2-0613: Unable to verify PLAN_TABLE format or existence SP2-0611: Error enabling EXPLAIN report

• Centrally managed PLAN_TABLE with public or private synonyms


211

Application tuning Foster smart coding habits SQL> set autot trace exp SQL> select * from mailing partition (mailing_p04); Execution Plan ---------------------------------------------------------0 1

SELECT STATEMENT Optimizer=CHOOSE (Cost=1 Card=164 Bytes=337676) 0

TABLE ACCESS (FULL) OF 'MAILING' (Cost=1 Card=164 Bytes=337676)

SQL> select * from mailing; Execution Plan ---------------------------------------------------------0

SELECT STATEMENT Optimizer=CHOOSE (Cost=2 Card=984 Bytes=2026056)

1

0

2

1

PARTITION RANGE (ALL) TABLE ACCESS (FULL) OF 'MAILING' (Cost=2 Card=984 Bytes=2026056)


Understanding the cost

212


Application tuning Cost analysis set echo off term off feed off ver off select decode(id,0,'', lpad(' ',2*(level-1))||level||'.'||position)||' '|| operation||' '||options||' '||object_name||' '|| object_type||' '|| decode(id,0,'Cost = '||position) Query_plan from plan_table connect by prior id = parent_id and statement_id = upper('&1') start with id = 0 and statement_id = upper('&1');

213


Application tuning Listen and listen closely

You can only compare costs between different wordings of the same SQL statement NOT across statements.

You won't get away with it!

214


Find potentially problematic SQL statements before it's too late

215


Application tuning Buffer gets measurement

select buffer_gets,sql_text,executions, buffer_gets/executions from v$sqlarea where buffer_gets > 200000

Good place to start

and v$sqlarea.executions != 0 order by buffer_gets/executions desc

216


Assess I/O balance while applications interact with instance

217


218

Application tuning I/O balance - primer col file_name form a30 head File col tablespace_name form a20 head Tbsp col a new_value preads col c new_value pwrites select sum(phyrds) a,sum(phywrts) c from v$filestat;

/u02 /u04

select file_name,tablespace_name, phyrds/&b*100 pctrd,phywrts/&d*100 pctwrt from sys.dba_data_files ddf,v$filestat vf where ddf.file_id = vf.file# and (phyrds/&preads*100 > 5 or phywrts/&pwrites*100 > 5) order by 3;


Application tuning Best practices • Share the responsibility • Make inroads with management • Balanced I/O patterns • Arm the developers so they can help themselves

• Compare relative costs within different wordings of the same statement • Imbed in program development • Pick reasonable measurement indicators

219


Working with OSS


220


Working with OSS MetaLink

• The first place to start • Be smart with your search criteria • Save yourself and OSS time • Someone else's problem yesterday is yours today • Learn from others' experience

221


Working with OSS Be a smart DBA

• Use profiles – 10 per account – keep them up-to-date – organize in 1 central account

• Familiarize yourself with updates – current O/S? uname – a – current Oracle version tool herald

222


Arm yourself with the necessary backup to your request

223


224

Working with OSS Profiles

Organize by client and O/S


Working with OSS Be specific Problem with rman Oracle error when doing a list backup command

Unique constraint violation when trying to list backup

225


Working with OSS TAR intake

• Keyword scan of free form entry • Routed to subject experts • Response time – 30-60 minutes – Severity 1 next to immediate

• Preferred contact method is email – ensure OSS has correct address – check incoming mail regularily

226


227

Working with OSS Tools of the trade

lert log

init.ora

ode

racle error(s)

xplain plan

race file(s)


Working with OSS Best practices • Have realistic expectations • Do your homework first (MetaLink) • Correct contact information • Be prepared

• Use email — best communications performance • Respond to requests • No games

228


Working with OSS

229


Hodge-podge


230


Hodge-podge Admin scripts

• ?/rdbms/admin • Familiarize yourself with the contents of this directory in your spare time • Two part naming convention – dbmsabcd.sql – prvt abcd.plb

231


Hodge-podge Hard-core education Just what do you think you're doing Dave?

232


233

Hodge-podge Smart professionals

$$


234

Hodge-podge In a bind • Work with your developers to bind everything – perl – SQL*Plus – AOWBI

• Re-usable SQL • Computation of hash value very restrictive

select from where group

sql_text,count(*) v$sqlarea instr(sql_text,'":SY"') > 0 by sql_text;

select count(*) from v$sqlarea; select count(*) from v$sqlarea where sql_text like '%'||''''||'%';


235

Hodge-podge Index foreign keys • Development requires primary/foreign key relationships to protect data integrity • Primary key constraints using index oops! • Never think of or take the time to index foreign key columns • Locks on deleting from parent

primary key index foreign key INDEX

HU GE locking problems

col table_name form a20 head Table col column_name form a30 heading 'Missing index for FK'


Hodge-podge Finding non-indexed foreign keys select dc.table_name,dcc.column_name from sys.dba_constraints dc, sys.dba_cons_columns dcc where dc.constraint_type = 'R' and dc.constraint_name = dcc.constraint_name and dc.owner not in ('SYS','SYSTEM') and not exists (select ' ' from sys.dba_indexes di,sys.dba_ind_columns dic where di.index_name = dic.index_name and di.table_name = dc.table_name and dic.column_name = dcc.column_name);

236


Hodge-podge Manage RI before it manages you • Frustrating venture • Endless Oracle errors • Protection as well as integrity

• constraint_type – C for check – R for reference – P for primary

• r_constraint_name • r_owner

237


Hodge-podge Manage RI before it manages you

SYS_ SYS_

alter table … add constraint …_pk primary key (…) using index storage (initial N next N pctincrease 0) tablespace …; alter table … add constraint …_fk foreign key (…) references … (…);

Thou shalt name your constraints Thou shalt alter table … add constraint

238


239

Hodge-podge Saving foreign key constraints – part 1 select 'alter table '||uc1.owner||'.'||uc1.table_name|| chr(10)||' drop constraint '|| uc1.constraint_name||';' from user_constraints uc1, -- foreign key constraint def user_constraints uc2, -- primary key constraint def user_cons_columns ucc1, -- columns in foreign key user_cons_columns ucc2 -- columns in primary key where uc1.r_constraint_name = upper(uc2.constraint_name) and ucc2.constraint_name = upper(uc2.constraint_name) and ucc1.constraint_name = uc1.constraint_name and uc2.table_name in ({list_of_tables});

DBAs worth their weight in gold


Hodge-podge Saving foreign key constraints – part 2 select 'alter table '||uc1.owner||'.'||uc1.table_name|| ' add constraint '||chr(10)||' '|| uc1.constraint_name||chr(10)||' foreign key ('|| ucc1.column_name||')'|| ' references '||uc2.owner||'.'||uc2.table_name|| ' ('||ucc2.column_name||');' from user_constraints uc1, -- foreign key constraint def user_constraints uc2, -- primary key constraint def user_cons_columns ucc1, -- columns in foreign key user_cons_columns ucc2 -- columns in primary key where uc1.r_constraint_name = upper(uc2.constraint_name) and ucc2.constraint_name = upper(uc2.constraint_name) and ucc1.constraint_name = uc1.constraint_name and uc1.table_name in ({list_of_tables});

240


241

Hodge-podge Extract source code – procedures / functions

• Be prepared / be smart • Quicker and more useful than alter procedure GEO_MAINT compile; sho errors package geo_maint set lines 9999 trimsp on pages 0 echo off feed off ver off sqlbl on spool pf select decode(line,1,'create or replace '||text,text) from user_source where name = upper('&1') order by line; prompt / spool off


242

Hodge-podge Extract source code - packages set lines 9999 trimsp on pages 0 echo off feed off ver off sqlbl on spool pkg select decode(line,1,'create or replace '||text,text) from user_source where name = upper('&1') and type = 'PACKAGE' order by line; prompt / prompt select decode(line,1,'create or replace '||text,text) from user_source where name = upper('&1') and type = 'PACKAGE BODY' order by line; prompt / spool off


Hodge-podge Global partitioned indexes

• Do not have the same partition implementation as their data • Boundaries specified on creation • Must be rebuilt from scratch when performing partition maintenance – adding or dropping – splitting or merging

243


Hodge-podge Avoid global index rebuilds

• List or range partitioning • Painstakingly adopted naming convention • Date or surrogate key based approach easiest to work with • Ensure there are no rows in last 2 partitions

244


245

Hodge-podge Avoid global index rebuilds

• Partition naming - table – table name + highest possible date – e.g., SUBS_20020831 • format of date mask important • allows for pseudo-sequential queries

it's 3AM – your planning will pay off

• Partition naming – index – table name + index name + U/N/F identifier – unique / non-unique / function-based


Hodge-podge Avoid global index rebuilds set pages 0 lines 999 trimsp on ver off echo off col a new_value mxname select from where and

max(partition_name) a dba_tab_partitions table_owner = upper('&1') table_name = upper('&2');

select 'Partitioned table '||upper('&1')||'.'|| upper('&2')||' has '||count(*)|| ' rows in last partition ('||'&mxname'||')' from &1..&2 partition (&mxname) having count(*) = 0; set lines 75

246


Hodge-podge Avoid global index rebuilds set echo off feed off pages 0 select distinct '@lp_look '||table_owner|| ' '||table_name from sys.dba_tab_partitions order by table_owner,table_name; . . . SQL> @lp_look TERA WORK SQL> @lp_look TERA YIELD . . . Partitioned table TERA.WORK has 3 rows in last partition (TERA_WORK_P20030630) Partitioned table TERA.YIELD has 8 rows in last partition (TERA_YIELD_P20030831)

247


Hodge-podge Shutdown abort When running with or without media recovery enabled

SQL> shutdown abort ORACLE instance shutdown. SQL> startup restrict . . . SQL> shutdown . . . ORACLE instance shutdown.

248


Hodge-podge A word on OEM

249


250

Hodge-podge Weaning yourself of Server Manager

• No more svrmgrl with release 9i • No more internal user • Done exclusively in SQL*Plus, unlike 8i with either/or • Close the one remaining inconsistency between the way these 2 tools behave • Ripple affect on backup scripts

9i


Hodge-podge Ensuring tools behave exactly the same way SVRMGR> select count(*), cus_id 2> from reader 3>

SQL> select count(*), cus_id 2> from reader 3>

4>

group by cus_id

5>

having count(*) > 1;

SQL>

• Server Manager tolerates

blank lines, whereas SQL*Plus didn't until

251


252

Hodge-podge Change in logon procedures

connect internal

8i

9i /nolog connect / as sysdba 10g


253

Hodge-podge Online redo log maintenance

• To protect you from yourself, rename the members … Before … select group#,member from v$logfile order by group#,member GROUP# MEMBER ------ ----------------------1 /redo1/beg9/redo01a.log 1 /redo2/beg9/redo01b.log 2 /redo2/beg9/redo02a.log 2 /redo1/beg9/redo02b.log

After … select group#,member from v$logfile order by group#,member GROUP# MEMBER ------ ----------------------1 /redo1/beg9/log1_g1.log 1 /redo2/beg9/log2_g1.log 2 /redo2/beg9/log1_g2.log 2 /redo1/beg9/log2_g2.log


254

The chicken or the egg?? practices allow your systems to

Better systems

=

users contribute to a smarter DBA

better.

users.

Oracle DBA Best Practices

Recommend Documents