Oracle DBA Checklist

Version 1.4 Generic

Revised: 23-Aug-2000

Authors: Thomas B. Cox, with Christine Choi

Purpose: This document gives details for performing daily, weekly, and monthly checks of the status of one or more Oracle databases. All SQL and PL/SQL code for the listed checks can be found in the appendix.

The latest version of this paper should always be available on the primary author's home page, <http://www.geocities.com/tbcox23>.

Change Notes: 1.1: Typo in 'existext.sql' identified by Steve DeNunzio, fixed

1.2: Typos fixed

1.3 Gnu Public License added; pctincr 0 in rebuild index added

1.4 Added pointer to newest version on Geocities home page, http://www.geocities.com/tbcox23
Fixed pointer to 'orapub' web site
Added nightly checklist and volumetrics

Support Information (customize for your site):

Help Desk: <phone>

Physical DBA: <name> <phone>

Application DBA: <name> <phone>

Oracle Support: CSI: <#> <phone>

Acknowledgements:

This paper was inspired by the work of David Cook (see References), and Version 1.0 was largely fleshed out by Christine Choi of Hewlett-Packard (Components Group), San Jose, California. I am grateful to both for their contributions to this document.

Please send your corrections, suggestions, and feedback to me at the address below, with your return address so I may credit your contribution. Thank you.

-Thomas B. Cox, <tbcox@att.net>

This document is free; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2

of the License, or (at your option) any later version.

This document is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

For a copy of the GNU General Public License write to the Free Software

Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

Index

I. Daily Procedures 3

A. Verify all instances are up 3

B. Look for any new alert log entries 3

C. Verify DBSNMP is running 3

D. Verify success of database backup 3

E. Verify success of database archiving to tape 3

F. Verify enough resources for acceptable performance 3

G. Copy Archived Logs to Standby Database and Roll Forward 5

H. Read DBA manuals for one hour 5

II. Nightly Procedures 6

A. Collect volumetric data 6

III. Weekly Procedures 7

A. Look for objects that break rules 7

B. Look for security policy violations 7

C. Look in SQL*Net logs for errors, issues 7

D. Archive all Alert Logs to history 7

E. Visit home pages of key vendors 8

IV. Monthly Procedures 9

A. Look for Harmful Growth Rates 9

B. Review Tuning Opportunities 9

C. Look for I/O Contention 9

D. Review Fragmentation 9

E. Project Performance into the Future 9

F. Perform Tuning and Maintenance 9

V. Appendix 10

A. Daily Procedures 10

B. Nightly Procedures 12

C. Weekly Procedures 14

VI. References 17

I. Daily Procedures

A. Verify all instances are up

Make sure the database is available. Log into each instance and run daily reports or test scripts. Some sites may wish to automate this.

Optional implementation: use Oracle Enterprise Manager's 'probe' event.

B. Look for any new alert log entries

· Connect to each managed system.

· Use 'telnet' or comparable program.

· For each managed instance, go to the background dump destination, usually $ORACLE_BASE/<SID>/bdump. Make sure to look under each managed database's SID.

· At the prompt, use the Unix ‘tail’ command to see the alert_<SID>.log, or otherwise examine the most recent entries in the file.

· If any ORA-errors have appeared since the previous time you looked, note them in the Database Recovery Log and investigate each one. The recovery log is in <file>.

C. Verify DBSNMP is running

1. Log on to each managed machine to check for the 'dbsnmp' process.

For Unix: at the command line, type ps –ef | grep dbsnmp. There should be two dbsnmp processes running. If not, restart DBSNMP. (Some sites have this disabled on purpose; if this is the case, remove this item from your list, or change it to "verify that DBSNMP is NOT running".)

D. Verify success of database backup

E. Verify success of database archiving to tape

F. Verify enough resources for acceptable performance

1. Verify free space in tablespaces.

For each instance, verify that enough free space exists in each tablespace to handle the day’s expected growth. As of <date>, the minimum free space for <repeat for each tablespace>: [ < tablespace > is < amount > ]. When incoming data is stable, and average daily growth can be calculated, then the minimum free space should be at least <time to order, get, and install more disks> days’ data growth.

a) Go to each instance, run free.sql to check free mb in tablespaces.

Compare to the minimum free MB for that tablespace. Note any low-space conditions and correct.

b) Go to each instance, run space.sql to check percentage free in tablespaces.

Compare to the minimum percent free for that tablespace. Note any low-space conditions and correct.

2. Verify rollback segment.

Status should be ONLINE, not OFFLINE or FULL, except in some cases you may have a special rollback segment for large batch jobs whose normal status is OFFLINE.

a) Optional: each database may have a list of rollback segment names and their expected statuses.

b) For current status of each ONLINE or FULL rollback segment (by ID not by name), query on V$ROLLSTAT.

c) For storage parameters and names of ALL rollback segment, query on DBA_ROLLBACK_SEGS. That view’s STATUS field is less accurate than V$ROLLSTAT, however, as it lacks the PENDING OFFLINE and FULL statuses, showing these as OFFLINE and ONLINE respectively.

3. Identify bad growth projections.

Look for segments in the database that are running out of resources (e.g. extents) or growing at an excessive rate. The storage parameters of these segments may need to be adjusted. For example, if any object reached 200 as the number of current extents, AND it's an object that is supposed to get large, upgrade the max_extents to unlimited.

a) To gather daily sizing information, run analyze5pct.sql. If you are collecting nightly volumetrics, skip this step.

b) To check current extents, run nr_extents.sql

c) Query current table sizing information

d) Query current index sizing information

e) Query growth trends

4. Identify space-bound objects.

Space-bound objects’ next_extents are bigger than the largest extent that the tablespace can offer. Space-bound objects can harm database operation. If we get such object, first need to investigate the situation. Then we can use ALTER TABLESPACE <tablespace> COALESCE. Or add another datafile.

a) Run spacebound.sql. If all is well, zero rows will be returned.

5. Processes to review contention for CPU, memory, network or disk resources.

a) To check CPU utilization, go to x:\web\phase2\default.htm =>system metrics=>CPU utilization page. 400 is the maximum CPU utilization because there are 4 CPUs on phxdev and phxprd machine. We need to investigate if CPU utilization keeps above 350 for a while.

G. Copy Archived Logs to Standby Database and Roll Forward

If you have a Standby Database, copy the appropriate Archived Logs to the expected location on the standby machine and apply those logs (roll forward the changes) to the standby database. This keeps the standby database up-to-date.

The copying of logs, the applying of them, or both, can in some cases be automated. If you have automated them, then your daily task should be to confirm that this happened correctly each day.

H. Read DBA manuals for one hour

Nothing is more valuable in the long run than that the DBA be as widely experienced, and as widely read, as possible. Readings should include DBA manuals, trade journals, and possibly newsgroups or mailing lists.

II. Nightly Procedures

Most production databases (and many development and test databases) will benefit from having certain nightly batch processes run.

A. Collect volumetric data

This example collects table row counts. This can easily be extended to other objects such as indexes, and other data such as average row sizes.

1. Analyze Schemas and Collect Data.

The idea here is to use the more time consuming and more accurate ANALYZE COMPUTE command and save the results, which show up in the data dictionary, to a more permanent store.

a) If you havent' yet, create the volumetrics table with mk_volfact.sql

b) To gather nightly sizing information, run analyze_comp.sql.

c) To collect the resulting statistics, run pop_vol.sql

d) Examine the data at your leisure, probably weekly or monthly.

I use MS Excel and an ODBC connection to examine and graph data growth.

III. Weekly Procedures

A. Look for objects that break rules

For each object-creation policy (naming convention, storage parameters, etc.) have an automated check to verify that the policy is being followed.

1. Every object in a given tablespace should have the exact same size for NEXT_EXTENT, which should match the tablespace default for NEXT_EXTENT. As of 12/14/98, default NEXT_EXTENT for DATAHI is 1 gig (1048576 bytes), DATALO is 500 mb (524288 bytes), and INDEXES is 256 mb (262144 bytes).

a) To check settings for NEXT_EXTENT, run nextext.sql.

b) To check existing extents, run existext.sql

2. All tables should have unique primary keys.

a) To check missing PK, run no_pk.sql.

b) To check disabled PK, run disPK.sql.

c) All primary key indexes should be unique. Run nonuPK.sql to check.

3. All indexes should use INDEXES tablespace. Run mkrebuild_idx.sql.

4. Schemas should look identical between environments, especially test and production.

a) To check data type consistency, run datatype.sql.

b) To check other object consistency, run obj_coord.sql.

c) Better yet, use a tool like Quest Software's Schema Manager.

B. Look for security policy violations

C. Look in SQL*Net logs for errors, issues

1. Client side logs

2. Server side logs

D. Archive all Alert Logs to history

E. Visit home pages of key vendors

1. Oracle Corporation

http://www.oracle.com
http://technet.oracle.com
http://www.oracle.com/support
http://www.oramag.com

2. Quest Software

http://www.quests.com

3. Sun Microsystems

http://www.sun.com

IV. Monthly Procedures

A. Look for Harmful Growth Rates

1. Review changes in segment growth when compared to previous reports to identify segments with a harmful growth rate.

B. Review Tuning Opportunities

1. Review common Oracle tuning points such as cache hit ratio, latch contention, and other points dealing with memory management. Compare with past reports to identify harmful trends or determine impact of recent tuning adjustments.

C. Look for I/O Contention

1. Review database file activity. Compare to past output to identify trends that could lead to possible contention.

D. Review Fragmentation

1. Investigate fragmentation (e.g. row chaining, etc.).

E. Project Performance into the Future

1. Compare reports on CPU, memory, network, and disk utilization from both Oracle and the operating system to identify trends that could lead to contention for any one of these resources in the near future.

2. Compare performance trends to Service Level Agreement to see when the system will go out of bounds

F. Perform Tuning and Maintenance

1. Make the adjustments necessary to avoid contention for system resources. This may include scheduled down time or request for additional resources.

V. Appendix

A. Daily Procedures

1. Free.sql

-- free.sql

-- To verify free space in tablespaces

-- Minimum amount of free space

-- document your thresholds:

-- <tablespace_name> = <amount> m

SELECT tablespace_name, sum ( blocks ) as free_blk , trunc ( sum ( bytes ) / (1024*1024) ) as free_m

, max ( bytes ) / (1024) as big_chunk_k, count (*) as num_chunks

FROM dba_free_space

GROUP BY tablespace_name

2. Space.sql

-- space.sql

-- To check free, pct_free, and allocated space within a tablespace

-- 11/24/98

SELECT tablespace_name, largest_free_chunk

, nr_free_chunks, sum_alloc_blocks, sum_free_blocks

, to_char(100*sum_free_blocks/sum_alloc_blocks, '09.99') || '%'

AS pct_free

FROM ( SELECT tablespace_name

, sum(blocks) AS sum_alloc_blocks

FROM dba_data_files

GROUP BY tablespace_name

)

, ( SELECT tablespace_name AS fs_ts_name

, max(blocks) AS largest_free_chunk

, count(blocks) AS nr_free_chunks

, sum(blocks) AS sum_free_blocks

FROM dba_free_space

GROUP BY tablespace_name )

WHERE tablespace_name = fs_ts_name

3. analyze5pct.sql

-- analyze5pct.sql

-- To analyze tables and indexes quickly, using a 5% sample size

-- (do not use this script if you are performing the overnight

-- collection of volumetric data)

-- 11/30/98

BEGIN

dbms_utility.analyze_schema ( '&OWNER', 'ESTIMATE', NULL, 5 ) ;

END ;

4. nr_extents.sql

-- nr_extents.sql

-- To find out any object reaching <threshold>

-- extents, and manually upgrade it to allow unlimited

-- max_extents (thus only objects we *expect* to be big

-- are allowed to become big)

-- 11/30/98

SELECT e.owner, e.segment_type , e.segment_name , count(*) as nr_extents , s.max_extents

, to_char ( sum ( e.bytes ) / ( 1024 * 1024 ) , '999,999.90') as MB

FROM dba_extents e , dba_segments s

WHERE e.segment_name = s.segment_name

GROUP BY e.owner, e.segment_type , e.segment_name , s.max_extents

HAVING count(*) > &THRESHOLD

OR ( ( s.max_extents - count(*) ) < &&THRESHOLD )

ORDER BY count(*) desc

5. spacebound.sql

-- spacebound.sql

-- To identify space-bound objects. If all is well, no rows are returned.

-- If any space-bound objects are found, look at value of NEXT extent

-- size to figure out what happened.

-- Then use coalesce (alter tablespace <foo> coalesce;).

-- Lastly, add another datafile to the tablespace if needed.

-- 11/30/98

SELECT a.table_name, a.next_extent, a.tablespace_name

FROM all_tables a,

( SELECT tablespace_name, max(bytes) as big_chunk

FROM dba_free_space

GROUP BY tablespace_name ) f

WHERE f.tablespace_name = a.tablespace_name

AND a.next_extent > f.big_chunk

B. Nightly Procedures

1. mk_volfact.sql

-- mk_volfact.sql (only run this once to set it up; do not run it nightly!)

-- -- Table UTL_VOL_FACTS

CREATE TABLE utl_vol_facts

(

table_name VARCHAR2(30),

num_rows NUMBER,

meas_dt DATE

)

TABLESPACE platab

STORAGE (

INITIAL 128k

NEXT 128k

PCTINCREASE 0

MINEXTENTS 1

MAXEXTENTS unlimited

)

-- Public Synonym

CREATE PUBLIC SYNONYM utl_vol_facts FOR &OWNER..utl_vol_facts

-- Grants for UTL_VOL_FACTS

GRANT SELECT ON utl_vol_facts TO public

2. analyze_comp.sql

-- analyze_comp.sql

BEGIN

sys.dbms_utility.analyze_schema ( '&OWNER','COMPUTE');

END ;

3. pop_vol.sql

-- pop_vol.sql

insert into utl_vol_facts

select table_name

, NVL ( num_rows, 0) as num_rows

, trunc ( last_analyzed ) as meas_dt

from all_tables -- or just user_tables

where owner in ('&OWNER') -- or a comma-separated list of owners

commit

C. Weekly Procedures

1. nextext.sql

-- nextext.sql

-- To find tables that don't match the tablespace default for NEXT extent.

-- The implicit rule here is that every table in a given tablespace should

-- use the exact same value for NEXT, which should also be the tablespace's

-- default value for NEXT.

-- This tells us what the setting for NEXT is for these objects today.

-- 11/30/98

SELECT segment_name, segment_type, ds.next_extent as Actual_Next

, dt.tablespace_name, dt.next_extent as Default_Next

FROM dba_tablespaces dt, dba_segments ds

WHERE dt.tablespace_name = ds.tablespace_name

AND dt.next_extent !=ds.next_extent

AND ds.owner = UPPER ( '&OWNER' )

ORDER BY tablespace_name, segment_type, segment_name

2. existext.sql

-- existext.sql

-- To check existing extents

-- This tells us how many of each object's extents differ in size from

-- the tablespace's default size. If this report shows a lot of different

-- sized extents, your free space is likely to become fragmented. If so,

-- this tablespace is a candidate for reorganizing.

-- 12/15/98

SELECT segment_name, segment_type

, count(*) as nr_exts

, sum ( DECODE ( dx.bytes,dt.next_extent,0,1) ) as nr_illsized_exts

, dt.tablespace_name, dt.next_extent as dflt_ext_size

FROM dba_tablespaces dt, dba_extents dx

WHERE dt.tablespace_name = dx.tablespace_name

AND dx.owner = '&OWNER'

GROUP BY segment_name, segment_type, dt.tablespace_name, dt.next_extent

3. No_pk.sql

-- no_pk.sql

-- To find tables without PK constraint

-- 11/2/98

SELECT table_name

FROM all_tables

WHERE owner = '&OWNER'

MINUS

SELECT table_name

FROM all_constraints

WHERE owner = '&&OWNER'

AND constraint_type = 'P'

4. disPK.sql

-- disPK.sql

-- To find out which primary keys are disabled

-- 11/30/98

SELECT owner, constraint_name, table_name, status

FROM all_constraints

WHERE owner = '&OWNER' AND status = 'DISABLED’ AND constraint_type = 'P'

5. nonuPK.sql

-- nonuPK.sql

-- To find tables with nonunique PK indexes. Requires that PK names

-- follow a naming convention. An alternative query follows that

-- does not have this requirement, but runs more slowly.

-- 11/2/98

SELECT index_name, table_name, uniqueness

FROM all_indexes

WHERE index_name like '&PKNAME%'

AND owner = '&OWNER' AND uniqueness = 'NONUNIQUE'

SELECT c.constraint_name, i.tablespace_name, i.uniqueness

FROM all_constraints c , all_indexes i

WHERE c.owner = UPPER ( '&OWNER' ) AND i.uniqueness = 'NONUNIQUE'

AND c.constraint_type = 'P' AND i.index_name = c.constraint_name

6. mkrebuild_idx.sql

-- mkrebuild_idx.sql

-- Rebuild indexes to have correct storage parameters

-- 11/2/98

SELECT 'alter index ' || index_name || ' rebuild '

, 'tablespace INDEXES storage '

|| ' ( initial 256 K next 256 K pctincrease 0 ) ; '

FROM all_indexes

WHERE ( tablespace_name != 'INDEXES'

OR next_extent != ( 256 * 1024 )

)

AND owner = '&OWNER'

7. datatype.sql

-- datatype.sql

-- To check datatype consistency between two environments

-- 11/30/98

SELECT

table_name,

column_name,

data_type,

data_length,

data_precision,

data_scale,

nullable

FROM all_tab_columns -- first environment

WHERE owner = '&OWNER'

MINUS

SELECT

table_name,

column_name,

data_type,

data_length,

data_precision,

data_scale,

nullable

FROM all_tab_columns@&my_db_link -- second environment

WHERE owner = '&OWNER2'

order by table_name, column_name

8. obj_coord.sql

-- obj_coord.sql

-- To find out any difference in objects between two instances

-- 12/08/98

SELECT object_name, object_type

FROM user_objects

MINUS

SELECT object_name, object_type

FROM user_objects@&my_db_link

VI. References

1. Loney, Kevin Oracle8 DBA Handbook

2. Cook, David Database Management from Crisis to Confidence
[http://www.orapub.com/]

3. Cox, Thomas B. The Database Administration Maturity Model

Monday, 7 November 2011

dba check list

1. Log on to each managed machine to check for the 'dbsnmp' process.

1. Verify free space in tablespaces.

a) Go to each instance, run free.sql to check free mb in tablespaces.

b) Go to each instance, run space.sql to check percentage free in tablespaces.

2. Verify rollback segment.

a) Optional: each database may have a list of rollback segment names and their expected statuses.

b) For current status of each ONLINE or FULL rollback segment (by ID not by name), query on V$ROLLSTAT.

c) For storage parameters and names of ALL rollback segment, query on DBA_ROLLBACK_SEGS. That view’s STATUS field is less accurate than V$ROLLSTAT, however, as it lacks the PENDING OFFLINE and FULL statuses, showing these as OFFLINE and ONLINE respectively.

3. Identify bad growth projections.

a) To gather daily sizing information, run analyze5pct.sql. If you are collecting nightly volumetrics, skip this step.

b) To check current extents, run nr_extents.sql

c) Query current table sizing information

d) Query current index sizing information

e) Query growth trends

4. Identify space-bound objects.

a) Run spacebound.sql. If all is well, zero rows will be returned.

5. Processes to review contention for CPU, memory, network or disk resources.

a) To check CPU utilization, go to x:\web\phase2\default.htm =>system metrics=>CPU utilization page. 400 is the maximum CPU utilization because there are 4 CPUs on phxdev and phxprd machine. We need to investigate if CPU utilization keeps above 350 for a while.

1. Analyze Schemas and Collect Data.

a) If you havent' yet, create the volumetrics table with mk_volfact.sql

b) To gather nightly sizing information, run analyze_comp.sql.

c) To collect the resulting statistics, run pop_vol.sql

d) Examine the data at your leisure, probably weekly or monthly.

1. Every object in a given tablespace should have the exact same size for NEXT_EXTENT, which should match the tablespace default for NEXT_EXTENT. As of 12/14/98, default NEXT_EXTENT for DATAHI is 1 gig (1048576 bytes), DATALO is 500 mb (524288 bytes), and INDEXES is 256 mb (262144 bytes).

a) To check settings for NEXT_EXTENT, run nextext.sql.

b) To check existing extents, run existext.sql

2. All tables should have unique primary keys.

a) To check missing PK, run no_pk.sql.

b) To check disabled PK, run disPK.sql.

c) All primary key indexes should be unique. Run nonuPK.sql to check.

3. All indexes should use INDEXES tablespace. Run mkrebuild_idx.sql.

4. Schemas should look identical between environments, especially test and production.

a) To check data type consistency, run datatype.sql.

b) To check other object consistency, run obj_coord.sql.

c) Better yet, use a tool like Quest Software's Schema Manager.

1. Client side logs

2. Server side logs

1. Oracle Corporation

2. Quest Software

3. Sun Microsystems

1. Review changes in segment growth when compared to previous reports to identify segments with a harmful growth rate.

1. Review common Oracle tuning points such as cache hit ratio, latch contention, and other points dealing with memory management. Compare with past reports to identify harmful trends or determine impact of recent tuning adjustments.

1. Review database file activity. Compare to past output to identify trends that could lead to possible contention.

1. Investigate fragmentation (e.g. row chaining, etc.).

1. Compare reports on CPU, memory, network, and disk utilization from both Oracle and the operating system to identify trends that could lead to contention for any one of these resources in the near future.

2. Compare performance trends to Service Level Agreement to see when the system will go out of bounds

1. Make the adjustments necessary to avoid contention for system resources. This may include scheduled down time or request for additional resources.

1. Free.sql

2. Space.sql

3. analyze5pct.sql

4. nr_extents.sql

5. spacebound.sql

1. mk_volfact.sql

2. analyze_comp.sql

3. pop_vol.sql

1. nextext.sql

2. existext.sql

3. No_pk.sql

4. disPK.sql

5. nonuPK.sql

6. mkrebuild_idx.sql

7. datatype.sql

8. obj_coord.sql

1 comment: