6212883126866262792 performance testing_cloud

Oracle Database 19c
Roy Swonger, Mike Dietrich & Daniel Overby Hansen
Oracle Database Upgrade, Utilities, Cloud Migration and Patching
Copyright © 2021, Oracle and/or its affiliates
1
Performance Testing in the Cloud
Updated: May 13, 2021
Photo
by
David
Becker
on
Unsplash

Vice President
Database Upgrade,
Utilities & Patching
Roy F. Swonger
@RoyFSwonger

Distinguished Product Manager
Database Upgrade
and Migrations
Mike Dietrich
https://MikeDietrichDE.com
@MikeDietrichDE
mikedietrich

Senior Principal Product Manager
Cloud Migration
Daniel Overby Hansen
https://dohdatabase.com
@dohdatabase
dohdatabase

5
https://DOHdatabase.com

Recorded Web Seminars
https://MikeDietrichDE.com/videos/
https://dohdatabase.com/webinars
6

7
The
Usual
Testing
Challenges
Photo
by
Sigmund
on
Unsplash

Testing | Usual Challenges
Only 10% of real data used
Artificially created data sets
Outdated data
Tests done on a laptop
No testing tools used
No stale statistics refreshed
Testing?? Maybe next time …
Photo
by
Elijah
O'Donnell
on
Unsplash
"

In
95%
Help me - I have an upgrade problem …
Photo
by
nikko
macaspac
on
Unsplash
“ ”
of all cases, "upgrade problem" in reality is a performance issue
after upgrade. Or not database related.
There is exactly one way to mitigate the risk.
TESTING!

Testing | Pareto Principle – 80/20 Rule
Costs
Completeness
low high
0%
100%
80%

11
But what if
you get only old and outdated test hardware?
https://unsplash.com/photos/Zd6PL6PSW5E

12
And no proper testing tools?
Photo
by
Fredy
Jacob
on
Unsplash

Step 2
Download and
install newest RU
MOS Note: 2118136.2
your key to
Successful Database Upgrades / Migrations
Step 1
Download and
install Oracle 19c
eDelivery.oracle.com
Step 3
Download and use
AutoUpgrade
MOS Note: 2485457.1
Step 4
Performance Stability
with SPM, STA and RAT
13

Hardware
Challenge #1
14

When you provision a DBCS in OCI
the license includes some of the
Options and Management Packs?
15 Copyright © 2021, Oracle and/or its affiliates
Why Use Cloud | License Included
Did you know?

© 2020 Oracle Corporation
DBCS EE DBCS EE-HP DBCS EE-EP ExaCS
Diagnostics Pack x x x x
Tuning Pack x x x x
Real Application Testing x x x x
Advanced Compression x x x
Oracle Database
Offerings
and licenses included

Why Use Cloud | Use Cases
Diagnostics Pack
• Database Replay Compare Period Report
• AWR
• ASH
• Performance Hub
Tuning Pack
• SQL Access Advisor
• SQL Tuning Advisor
• SQL Profiles
• Real-time SQL and PL/SQL Monitoring

Why Use Cloud | Use Cases
Real Application Testing
• Database Replay
(still need license for capture)
• SQL Performance Analyzer
Advanced Compression
• Compress your Data Pump export
(importing a compressed dump file is not licensed)
• Compress your backups of your OCI database using RMAN compression
Many more Options and Packs are included

Creating Database | Multitenant
Multitenant is the default in OCI
What if ... you are not there yet

Database Cloud Service | Virtual Machines
Entry-level, provision with GI or LVM (fast-provision)
Restrictions:
• Only one CDB - the pre-created one
• Can't install another Oracle Home
• COMPATIBLE is always default - unless
• Max. storage 40 TB
To create non-CDB
• Migrate your non-CDB using ZDM
• Restore a non-CDB on top of the pre-created CDB
• DB_NAME and DB_UNIQUE_NAME must match exactly

Database Cloud Service | Bare Metal
Mid-level, provision with GI
Restrictions:
• As many databases as you want
• DATA disk group, max. 16 TB
• Only one database edition
• Only one database per Oracle Home
To create non-CDB
• It is not possible to create a non-CDB via the console - use dbcli
• Before using dbcli it is a good idea to update it: cliadm update-dbcli

22
[root]$ dbcli create-database
--dbname DOH19
--databaseUniqueName DOH191
--no-cdb
--dbconsole
--adminpassword "<secret-password>"
--dbshape odb4
--dbtype SI
--version 19.10.0.0
Job details
----------------------------------------------------------------
ID: 6d87fbf0-2b03-417c-aec9-09b69b0d6222
Description: Database service creation with db name: DOH19
Status: Created
Created: May 9, 2021 8:33:11 AM UTC
Progress: 0%
Message:

23
[root]$ dbcli describe-job -i 6d87fbf0-2b03-417c-aec9-09b69b0d6222
Job details
----------------------------------------------------------------
ID: 6d87fbf0-2b03-417c-aec9-09b69b0d6222
Description: Database service creation with db name: DOH19
Status: Running
Created: May 9, 2021 8:33:11 AM UTC
Progress: 0%
Message:
Task Name Start Time End Time Status
--------------------------- -------------------------- ---------------------------------- --------
Pre Database Creation Tasks May 9, 2021 8:33:11 AM UTC May 9, 2021 8:33:13 AM UTC Success
Database Home Creation May 9, 2021 8:33:13 AM UTC May 9, 2021 8:33:13 AM UTC Running

Database Cloud Service | Exadata
World's best database machine, provision with GI
Restrictions:
• As many databases as you want
• Non-CDB are supported only for 12.1.0.2 and 19c
• Max. storage up to 598 TB (depend on Exadata version and configuration)
To create non-CDB
• It is not possible to create a non-CDB via the console - use dbaasapi
• Creating non-CDB databases on the Service in OCI (Doc ID 2528257.1)

Tools
Challenge #2
25

Remember?
26

27
1.
Collect
2.
Compare
3.
Analyze
4.
Tune
5.
Manage
6.
Test
Performance Stability Prescription

Performance Stability Prescription | Adoption
28
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Performance Stability Prescription | Toolbox
AWR
Cursor Cache
SQL Tuning Set
Staging Tables
SQL Plan Management
SQL Tuning Advisor
SQL Performance Analyzer
Database Capture and Replay
Photo
by
jesse
orrico
on
Unsplash

Migration
Challenge #3
30

Migration | Move to the Cloud

Migrate | Overview
When migrating for test
• Don't worry about downtime
• Make it simple
• Avoid affecting production database

Migrate | Data Pump
• Run at off-peak hours
• Use parallel at off-peak hours
• Remember to make entire export consistent - flashback_time=systimestamp
• Use Data Pump compression (license required)
- else use OS utilities for compressing
• Export from a snapshot standby database
• Database Replay - recover standby to desired SCN, then run Data Pump export

Migrate | Data Pump From Specific SCN
Export from standby database to:
• Offload production database
• Avoid ORA-01555: snapshot too old
1. Stop redo apply on standby database
2. Finish workload on primary database, e.g. prepare for Database Capture
3. Recover standby database to desired SCN
DGMGRL> edit database 'stdby1' set state='apply-off';
SQL> alter database recover managed standby database until change 16489354;

Migrate | Data Pump From Specific SCN
4. Convert to snapshot standby. It opens the database in read write mode which is required
by Data Pump
5. Export your data with Data Pump. It is not necessary to use flashback_scn or
flashback_time because no one is using the system
6. Convert your standby back into a physical standby database
$ expdp system schemas=SH parallel=8 directory=mydir dumpfile=sh%U.dmp
DGMGRL> convert database 'stdby1' to snapshot standby;
Pro tip: Snapshot Standby is
part of Enterprise Edition
DGMGRL> convert database 'stdby1' to physical standby;

Migrate | Existing Backups
• Use existing backups
• If backup sets are not compressed, consider using OS utilities
• No effect on production database
• Allows you to test your recovery protocols
• Cloud is an attractive Disaster Recovery strategy - give it a try

Migrate | Multipart Uploads
Recommended for files larger than 100 MB
Use OCI CLI
Max part size is 50 GB
OCI CLI installation guide
oci os object put
--namespace ... -bn ... --file ... --name ...
--part-size 1024
--parallel-upload-count 4

Migrate | Bulk Uploads
Recommended for many files
Use OCI CLI
Does multipart and parallel uploads automatically
Optionally,
• Finetune uploads using --parallel-upload-count and --part-size
• Prefix all file names with --object-prefix
• Include or exclude files selectively using patterns and --include and --exclude
oci os object bulk-upload
-ns ... -bn ... --src-dir ...

Upgrade | Overview
Upgrading in the Cloud

Upgrade | Overview
40
Cloud Tooling
AutoUpgrade
Data Pump

Upgrade | Overview
41
Cloud Tooling
• One-Button approach
• Requires fully supported system
• Slower than other options
• Not an option you can use for production upgrade

Upgrade | Overview
42
AutoUpgrade
• Breaks the cloud tooling
• Familiar option
• Likely what you will use for production upgrade
• Customizable
• Test your database during upgrade

Upgrade | Overview
43
Data Pump
• Migrate directly into new release
• Familiar option
• Probably not what you will use for production upgrade
• Customizable

Migrate | Your Data
44
BRING ALL YOUR DATA

Migrate | Storage Cap
45
40 TB
16 TB
598 TB

Migrate | Options
46
ALL DATA STRUCTURE ONLY
Migrate structure and data
Allows full testing - including the application
Requires more resources
Preferred method
Migrate structure only
Allows only partial testing
Can use the smallest possible shape
Alternative method

Migrate | Structure Only
Data Pump can export the structure but skip the actual data
On import
• Statistics are maintained
If the source table had 100 rows, the statistics will reflect this
• Statistics are locked
$ expdp .... schemas=SH content=metadata_only

Migrate | Structure Only
Alternatively, truncate tables and shrink data files
• Tablespace must be completely empty to resize the data files to a minimum
• Optionally, use the cascade keyword to truncate parent and child tables
• Statistics are cleared - import production stats afterwards
• Don't do this in production!
SQL> truncate table sh.sales;
SQL> truncate table sh.costs cascade;
...
SQL> alter database datafile 10 resize 5m;

Migrate | Subsetting
Why not Data Subsetting? It is an option - but data subsetting is hard!
• Which data to remove?
• By age
• Random
• Integrity
• Referential integrity
• Logical integrity
Pro tip: Explore the options in
Data Masking and Subsetting Guide

Perhaps some data isn't needed for testing
• Audit
• Historical
• Logging
• Temporary
Data Pump can exclude tables
If you need the tables, but want them empty:
$ expdp .... schemas=SH exclude=table:in('COSTS')
$ expdp .... schemas=SH exclude=table:in('COSTS')
$ expdp .... schemas=SH include=table:in('COSTS') content=metadata_only

During restore or duplicate RMAN can exclude tablespaces
After restore/duplicate
• Tablespaces are offline
• Any tables stored in the tablespace are still defined in data dictionary
• Any tables stored in the tablespace can't be accessed
• They are not to be considered as empty tables
RMAN> restore database skip tablespace audit, historical, logging ...
RMAN> duplicate ... skip tablespace audit, historical, logging ...

Setup
Challenge #4
52

All tests need to be repeatable
Guaranteed Restore Points
Snapshot Standby
53
Photo
by
Nick
Fewings
on
Unsplash

Setup | Guaranteed Restore Point
Set a Guaranteed Restore Point (GRP)
Flashback to the GRP
Test
Flashback
GRP

Setup | Guaranteed Restore Point
Set a Guaranteed Restore Point (GRP)
Flashback to the GRP
Before Test After Test
CREATE RESTORE POINT grpt
GUARANTEE FLASHBACK DATABASE;
SHUTDOWN IMMEDIATE
STARTUP MOUNT;
FLASHBACK DATABASE TO RESTORE POINT grpt;
SHUTDOWN IMMEDIATE
STARTUP MOUNT;
ALTER DATABASE OPEN RESETLOGS;
TEST

Setup | Snapshot Standby
Use a physical standby for testing
Test
Convert back to Physical Standby
GRP
Log Shipping
PROD PHYS
Convert to
Snapshot Standby

Statistics
Challenge #5
57

Statistics
Photo
by
Veri
Ivanova
on
Unsplash
Things to know

Statistics | Overview
59
Which statistics matter for our purpose?
• System statistics
• Measure performance of system CPU and storage subsystem
• Schema statistics
• User object statistics per schema
• Database statistics?
• Includes schema stats, dictionary stats, fixed objects stats, system stats

System Statistics | Overview
60
The system statistics describe hardware characteristics such as I/O and CPU
performance and utilization.
System statistics enable the query optimizer to more accurately estimate I/O and
CPU costs when choosing execution plans.
" Database 19c SQL Tuning Guide, chapter 10
That sounds like a good idea

System Statistics | Recommendation
61
... in most cases you should use the defaults and not gather system statistics.
"Databases supporting a pure data warehouse workload on an Oracle Exadata
Database Machine can benefit from system statistics gathered using the EXADATA
option
... if the workload is mixed or you are not in a position to test the effect of using
EXADATA system statistics, then stick to the defaults even on this platform.
Nigel Bayliss, Optimizer blog

System Statistics | Reference
To delete system statistics (and revert to defaults)
References:
• Optimizer blog, Should You Gather System Statistics?
• SQL Tuning Guide, System Statistics
• SQL Tuning Guide, Guidelines for Gathering Optimizer Statistics Manually
• Database Performance Tuning Guide, Session and System Statistics
SQL> EXEC DBMS_STATS.DELETE_SYSTEM_STATS

Statistics
Transporting
Photo
by
Clay
Banks
on
Unsplash

Transporting Statistics | Overview
Pro tip: You can read more about transporting
statistics in the SQL Tuning Guide
When you transport optimizer statistics between databases, you must use
DBMS_STATS to copy the statistics to and from a staging table, and tools to make
the table contents accessible to the destination database.
You can transport the following statistics
• Schema
• Table
• Database (rare)
• Dictionary and fixed objects (rare)

Restore complete database
Data Pump
• Full database export
• Subset or schema only
• METADATA_ONLY
• Best case
• All stats including COL_USAGE$ are included
• Statistics need to be transported
• Especially COL_USAGE$ will be missing
• Consider METHOD_OPT=SKEWONLY
Transporting Statistics | Use Cases
65

Transporting Statistics | Workflow
Source database
User data (SALESAPP)
Data dictionary (SYS)
This is where optimizer
stats are located

Transporting System Statistics | Workflow
Create staging table
Export statistics to staging table
Export staging table using Data Pump
SQL> EXEC DBMS_STATS.CREATE_STAT_TABLE (
ownname => 'SYSTEM',
stattab => 'SYS_STATS_STG');
SQL> EXEC DBMS_STATS.EXPORT_SYSTEM_STATS (
stattab => 'SYS_STATS_STG',
statown => 'SYSTEM');
$ expdp SYSTEM
DIRECTORY=DATA_PUMP_DIR
DUMPFILE=sys_stats_stg.dmp
TABLES=SYS_STATS_STG

Transporting System Statistics | Workflow
Target database
Upload dump file to the object storage
Import staging table using Data Pump
Update statistics when you transport from a different version
"Import" statistics from staging table
You may repeat this for schema stats as well
$ impdp SYSTEM
DIRECTORY=mydirectory
DUMPFILE=sys_stats_stg.dmp
TABLES=SYS_STATS_STG
SQL> EXEC dbms_stats.upgrade_stat_table (
ownname => 'SYSTEM',
stattab => 'SYS_STATS_STG');
SQL> EXEC DBMS_STATS.IMPORT_SYSTEM_STATS (
stattab => 'SYS_STATS_STG',
statown => 'SYSTEM');

Transporting Statistics | Demo
Watch on YouTube

Transporting Statistics | Nice to Know
• The optimizer does not use statistics stored in a user-owned table - only from dictionary
• Importing statistics makes them current (i.e., not stale)
• You can transfer to a higher version - potentially the stats table must be upgraded
• Incremental statistics: optionally export synopses as well
SQL> EXEC DBMS_STATS.IMPORT_SCHEMA_STATS ( ...
ORA-20002: Version of statistics table "SALESAPP"."OPT_STATS_STG" is too old
SQL> EXEC DBMS_STATS.UPGRADE_STAT_TABLE ('SALESAPP', 'OPT_STATS_STG');
SQL> EXEC DBMS_STATS.EXPORT_SCHEMA_STATS (
...
stat_category => 'OBJECT_STATS, REALTIME_STATS, SYNOPSES');

Transporting Statistics | Nice to Know - 2
• Gather stats on the staging table after Data Pump import
• Also, gather on staging table indexes
• Before executing DBMS_STATS.IMPORT_SCHEMA_STATS / IMPORT_TABLE_STATS
• If enabled, imported statistics will be added as pending stats until you publish them

Statistics | Refresh?
Should you refresh object statistics when you upgrade or migrate to Oracle 19c?
• It depends
• Don't do it if you don't have the exact amount of data to test with
• Do it if you have the same amount of data
• Be cautions
• When your source is 11.2, histograms can change
• Avoid gradual change of plans when stats become stale
• Workload is required to create correct statistics

Statistics | Disable Auto Tasks
In case you import your stats, better disable the Auto Task
• Check:
SQL> EXEC DBMS_AUTO_TASK_ADMIN.DISABLE( -
client_name => 'auto optimizer stats collection', -
operation => NULL, -
window_name => NULL);
SQL> SELECT client_name, status FROM dba_autotask_client;
CLIENT_NAME STATUS
------------------------------------ --------
sql tuning advisor ENABLED
auto optimizer stats collection DISABLED
auto space advisor ENABLED

Statistics | Gather Stats on Load
Since Oracle 12.1.0.2, statistics are gathered automatically on load
• CREATE TABLE AS SELECT (CTAS)
• INSERT AS SELECT (IAS)
• Feature can be turned off
_optimizer_gather_stats_on_load = FALSE
insert /*+append NO_GATHER_OPTIMIZER_STATISTICS*/ into MYTAB select …

Should you lock statistics?
Photo
by
Carlos
Muza
on
Unsplash
Photo
by
Jen
Theodore
on
Unsplash

Locking Statistics | Use Cases
You can lock statistics to prevent them from changing.
• Certain static environments
• Highly volatile tables
• Enable use of dynamic statistics
• ... and all the exceptions

Locking Statistics | How to
Lock table statistics
You can also lock at:
• Schema-level
• Partition-level
You can also unlock statistics
SQL> EXEC DBMS_STATS.LOCK_TABLE_STATS(ownname=>'MYAPP', tabname=>'MY_VOLATILE_TAB1');
Pro tip: Locking table statistics also
lock index and partition statistics

Locking Statistics | Worth mentioning
• Locking and unlocking statistics causes cursor invalidation
• To achieve plan stability, consider SQL Plan Management
• Statistics advisor will warn you about locked statistics
• Locking information is not exported

Workload
Challenge #6
79

Workload | Collect Information on Source
80
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Workload | Objective
81
Collect workload information on source database
1. Create SQL Tuning Set (STS)
2. Capture workload from
a) AWR
b) Cursor cache
SQL Tuning Set

An SQL Tuning Set (STS) enables you to group SQL statements and related
metadata in a single database object, which you can use to meet your tuning goals.
Specifically, SQL tuning sets achieve the following goals:
• Providing input to the performance tuning advisors
• Transporting SQL between databases
SQL Tuning Set | Definition
82
Database 19c SQL Tuning Guide, chapter 23
"

SQL Tuning Set | Definition
83
SQL statement
Context
Statistics
Plans

SQL Tuning Set | Create
84
S T S
First, create a SQL Tuning Set
SQL> BEGIN
DBMS_SQLSET.CREATE_SQLSET (
sqlset_name => 'UPG_STS_1',
description => 'For upgrade - from source'
);
END;
/
Pro tip: You can also use DBMS_SQLTUNE
to create a SQL Tuning Set

SQL Tuning Set | Capture
85
Next, capture statements from AWR
SQL> DECLARE
begin_id number;
end_id number;
cur sys_refcursor;
BEGIN
SELECT min(snap_id), max(snap_id) INTO begin_id, end_id
FROM dba_hist_snapshot;
open cur for
select value(p) from table(dbms_sqltune.select_workload_repository(
begin_snap => begin_id,
end_snap => end_id,
basic_filter => 'parsing_schema_name not in (''SYS'')',
ranking_measure1 => 'elapsed_time',
result_limit => 5000,
attribute_list => 'ALL')) p;
dbms_sqltune.load_sqlset('UPG_STS_1', cur);
close cur;
END;
/
Pro tip: Consider excluding other internal
schemas like DBSNMP, ORACLE_OCM,
LBACSYS, WMSYS, XDB, SYSTEM
S T S
SQL> DECLARE
begin_id number;
end_id number;
cur sys_refcursor;
BEGIN
open cur for
end_snap => end_id,
close cur;
END;
/
SQL> DECLARE
begin_id number;
end_id number;
cur sys_refcursor;
BEGIN
open cur for
end_snap => end_id,
close cur;
END;
/
SQL> DECLARE
begin_id number;
end_id number;
cur sys_refcursor;
BEGIN
open cur for
end_snap => end_id,
close cur;
END;
/

SQL Tuning Set | Capture
86
Optionally, capture statements from cursor cache
Careful - puts load on your system
SQL> BEGIN
DBMS_SQLSET.CAPTURE_CURSOR_CACHE_SQLSET(
time_limit => 900,
repeat_interval => 60,
capture_option => 'MERGE',
capture_mode => DBMS_SQLTUNE.MODE_ACCUMULATE_STATS,
sqlset_owner => NULL,
recursive_sql => 'HAS_RECURSIVE_SQL');
END;
/
Pro tip: SQL Tuning Guide shows how to
load all statements from a given schema
S T S

Workload | Transport SQL Tuning Sets
87
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Workload | Objective
88
Transport SQL Tuning Set
1. Pack SQL Tuning Set into staging table
2. Export staging table
3. Upload to object storage
4. Import into test database in the cloud
SQL Tuning Set + Data Pump

SQL Tuning Set | Transport
89
S T S
Pack into staging table on source database
SQL> BEGIN
DBMS_SQLTUNE.CREATE_STGTAB_SQLSET (
table_name => 'UPG_STGTAB_1');
DBMS_SQLTUNE.PACK_STGTAB_SQLSET (
staging_table_name => 'UPG_STGTAB_1');
END;
Export with Data Pump
$ expdp user
directory=mydirectory
dumpfile=upg_stgtab_1.dmp
tables=UPG_STGTAB_1
Optionally, use DBMS_SQLTUNE.REMAP_STGTAB_SQLSET
to remap between CON_DBID

90
S T S
Optionally, compress with Data Pump (license required)
$ expdp ... compression=all compression_algorithm=medium
Or, use OS utilities
$ zip upg_stgtab_1.zip upg_stgtab_1.dmp

91
S T S
1. Transfer dump file directly via scp or rsync
2. Or, use OCI CLI
$ oci os object put --namespace oradbclouducm
-bn zdm-oss
--file upg_stgtab_1.dmp
--part-size 100
--parallel-upload-count 4
Create Pre-Authenticated Request (PAR) and download
$ wget https://objectstorage... /o/upg_stgtab_1.dmp

92
S T S
Import with Data Pump to target database
$ impdp user
directory=mydirectory
dumpfile=upg_stgtab_1.dmp
tables=UPG_STGTAB_1
Unpack staging table
SQL> BEGIN
DBMS_SQLTUNE.UNPACK_STGTAB_SQLSET (
sqlset_name => '%'
replace => true
staging_table_name => 'UPG_STGTAB_1'
);
END;
/

SQL Tuning Set | License
SQL Tuning Sets can also be accessed by way of database server APIs and
command-line interfaces. Usage of any subprograms in the DBMS_SQLSET package
to manage SQL Tuning Sets is part of the EE and EE-ES offerings.
"
Database 19c Database Licensing Information User Manual
SQL Tuning Sets can also be accessed by way of database server APIs and
command-line interfaces. Usage of any subprograms in the DBMS_SQLSET package
to manage SQL Tuning Sets is part of the EE and EE-ES offerings.
In addition, the following subprograms, part of the DBMS_SQLTUNE package,
provide an older interface to manage SQL Tuning Sets and are also part of the EE
and EE-ES offerings:
ADD_SQLSET_REFERENCE CAPTURE_CURSOR_CACHE_SQLSET CREATE_SQLSET
CREATE_STGTAB_SQLSET DELETE_SQLSET DROP_SQLSET
LOAD_SQLSET PACK_STGTAB_SQLSET REMOVE_SQLSET_REFERENCE
SELECT_CURSOR_CACHE SELECT_SQLSET SELECT_WORKLOAD_REPOSITORY
UNPACK_STGTAB_SQLSET UPDATE_SQLSET

SQL Tuning Set | Recommendation
Always capture workload data into SQL Tuning Sets

Comparison
Challenge #7
95

Comparison| AWR Snapshots
96
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Comparison | Objective
97
Compare workload data
1. Identify two snapshots for a representative workload
2. Export AWR
3. Upload to object storage
4. Import AWR into test database in the cloud
5. Run representative workload in the cloud
6. Compare AWR snapshots
AWR snapshots and diff report
Due to completely different specs and load scenarios, the result may be misleading or not useful

Comparison | AWR Snapshots
98
Test Database
On-Prem Database
awrextr.sql awrload.sql
Copy export dump to Object Storage
awrddrpi.sql

Comparison | AWR Diff Report
99
Use script awrddrpi.sql Instances in this Workload Repository schema
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
DB Id Inst Num DB Name Instance Host
------------ ---------- --------- ---------- ------
72245725 1 UPGR UPGR hol.localdom
* 753780962 1 CDB2 CDB2 hol.localdom
Database Id and Instance Number for the First Pair of Snapshots
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Enter value for dbid:

Comparison | SQL Performance Analyzer
100
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Comparison | Objective
101
Find bad plans by simulating SQL execution
1. Build ANALYSIS task
2. Simulate SQL statements captured on production
a) AWR
b) Cursor Cache
3. Compare results
4. Remedy issues
5. Repeat analysis
SQL Performance Analyzer

SQL Performance Analyzer | SPA
" Oracle Database Real Application Testing Data Sheet
SPA provides fine-grained assessment of environment changes on SQL execution
plans and statistics by running the SQL statements both in isolation and serially
manner in before-change and after-change environments.
SPA functionality is well integrated with existing SQL Tuning Set (STS), SQL Tuning
Advisor, and SQL Plan Management functionality.

SPA | Concept
11.2.0.4
COMPARE
COMPARE
SQL Tuning Set
Production
workload
Upgrade /
Migrate
19c
SQL,
schema,
binds,
...
SPA
Test execution
After upgrade
Before upgrade
Plans
Elapsed, CPU time,
buffer gets, disk reads ...
Pro tip: For migrations,
import STS into target database

SPA | Regressed Report
From production
workload From test
execution

SPA | Continuous Improvement
19c
SPA
Test execution
Implement
change
COMPARE
COMPARE

Some ideas for testing with SPA
• init.ora Parameters
• New statistics
• Optimizer settings
• DBMS_OPTIM_BUNDLE _fix_control settings
• Histogram types
• New synopsis for incremental statistics
• Specific patches
109
Photo
by
Andrew
George
on
Unsplash

SPA | Validate Rows and Values
NEW FEATURE since Oracle 18c
exec DBMS_SQLPA.SET_ANALYSIS_TASK_PARAMETER( -
task_name => 'my_spa_task', -
parameter => 'COMPARE_RESULTSET', -
value => 'TRUE');

SPA | Scripts
Find the scripts to try out SPA on:
https://MikeDietrichDE.com/scripts/

Comparison | Database Capture and Replay
112
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Database Replay | Overview
113
" Database 19c Testing Guide, chapter 9
You can use Database Replay to capture a workload on the production system and
replay it on a test system with the exact timing, concurrency, and transaction
characteristics of the original workload.
This enables you to test the effects of a system change without affecting the
production system.
Requires Real Application Testing on source

114
11.2.0.4 19c
Capture files
Workload
capture
Test
system
Preprocess
Replay files
Replay
analysis

115
Ulrike Schwinn on blogs.oracle.com

1. Platform independent
2. RAC compliant
3. Per-PDB capture/replay
4. Capture and replay across database releases
5. Download scripts
Database Replay | Facts
116

1. Workload capture restrictions
2. Not suitable with external dependencies
• Database link, external tables, UTL_HTTP
3. Recommended to restart database before capture
• Startup in restricted mode, capture automatically sets unrestricted mode
4. Work best from dedicated SCN
• Data Pump FLASHBACK_SCN or restore to specific SCN
Database Replay | To Consider
117

Optimization
Challenge #8
118

Optimization | SQL Tuning Advisor
119
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

SQL Tuning Advisor
120
SQL Tuning Advisor is SQL diagnostic software in the Oracle Database Tuning Pack.
...
SQL Tuning Advisor is a mechanism for resolving problems related to suboptimally
performing SQL statements.

Types of findings:
1. Collection of object statistics
2. Creation of indexes
3. Rewriting SQL statements
4. Creation of SQL profiles
.... and more
SQL Tuning Advisor | Findings
121
Pro tip: SQL Developer has a good
interface to SQL Tuning Advisor

SQL Profiles
122
A SQL profile is a database object that contains auxiliary statistics specific to a SQL
statement.
...
The corrected statistics in a SQL profile can improve optimizer cardinality estimates,
which in turn leads the optimizer to select better plans.

1. Part of Tuning Pack
• Included in some cloud offerings
2. Stores a set of hints that causes the optimizer to select a plan
3. Affects one statement only
SQL Profiles | Facts
123

4. You can enable/disable a profile
5. Transparent to application
• Does not require application changes
6. Persistent and transportable
• Documentation
7. Useful with literals using FORCE_MATCH=TRUE
SQL Profiles | Facts
124

SQL Profile | Testing
125
1. Enable profile for selected environments only
2. Verify the profile – it doesn't get used by the optimizer in the live environment
3. Accept and make visible to all sessions ('DEFAULT')
SQL> exec :p_name := dbms_sqltune.accept_sql_profile(
task_name=>'TASK_21944',
name=>'XT_PROFILE',
category=>'TEST_ENV');
SQL> alter session set sqltune_category='TEST_ENV';
SQL> exec dbms_sqltune.alter_sql_profile(
name=>'XT_PROFILE',
attribute_name=>'CATEGORY',
value=>'DEFAULT');

Optimization | SQL Plan Management
126
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

SQL Plan Management | SPM
127
SQL plan management uses a mechanism called a SQL plan baseline, which is a set
of accepted plans that the optimizer is allowed to use for a SQL statement.
SQL plan management uses a mechanism called a SQL plan baseline, which is a set
of accepted plans that the optimizer is allowed to use for a SQL statement.
...
SQL plan management prevents performance regressions caused by plan changes.

SPM | Concept
128
Plan baseline
Repeatable SQL
OPTIMIZER_CAPTURE_SQL_PLAN_BASELINES=TRUE
Filter
Filter on:
• Schema
• SQL text
• Module
• Action
Plan A
Plan A
Accepted
No plans
in baseline
Next execution
Plan B
Something changed
• New statistics
• New parameters
• Upgrade
Plan History
Plan history
- Dedicated part of SQL plan baseline
Plan B
Not accepted
OPTIMIZER_USE_SQL_PLAN_BASELINES=TRUE
Plan A is used

The plans in a SQL plan baseline can be:
• Enabled
• Accepted
• Fixed
To change status use DBMS_SPM.ALTER_SQL_PLAN_BASELINE
You can also prevent plans from getting purged
by setting the autopurge property.
SPM | Plans
129
Pro tip: The Accepted attribute can
only be set by a test execution

SPM | Load from STS
130
Plan A
Accepted
Plan History
Plan B
Not accepted
Plan C
SQL> DECLARE
cnt number;
BEGIN
cnt := DBMS_SPM.LOAD_PLANS_FROM_SQLSET('UPG_STS_1');
END;
/
Plan C
Accepted
SQL Tuning set
Automatically
accepted

SQL Plan Management in a system with literals is not a good fit
• Many distinct statements
• CURSOR_SHARING = FORCE? No!
• SQL profiles can do force matching
Optimal solution: Change your application to use bind variables
SPM | What if ... literals
131

Plan baseline
SPM | Use Case
132
Source
11.2.0.4
Test
19c
Plans from production
Plans from tuning
Production
19c
Automatically
accepted
Load into
SQL plan baseline

Plan baseline
SPM | Use Case
133
SQL> DECLARE
plans_loaded NUMBER;
filter VARCHAR2(255);
BEGIN
filter := 'sql_id=''czzzubf8fjz96'' AND plan_hash_value=''1165613724''';
SQL> DECLARE
plans_loaded NUMBER;
filter VARCHAR2(255);
BEGIN
filter := 'sql_id=''czzzubf8fjz96'' AND plan_hash_value=''1165613724''';
plans_loaded := DBMS_SPM.LOAD_PLANS_FROM_SQLSET (
basic_filter => filter
);
END;
/
Pro tip: The function LOAD_PLANS_
FROM_SQLSET can also fix the plans

Transport
Challenge #9
134

Transport | SQL Plan Baselines
135
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Transporting | SQL Plan Baseline
Prepare
Extract
Transfer
Load
SQL Plan Baselines are stored in data dictionary
To transfer - information must be converted to a transportable
format and stored in a staging table
SQL> BEGIN
DBMS_SPM.CREATE_STGTAB_BASELINE (
table_name => 'SPB_STAGING',
table_owner => 'SPM');
END;
/

Prepare
Extract
Transfer
Load
Select the baselines that you want to transfer
To extract the fixed and accepted plans
SQL> DECLARE
l_count NUMBER;
BEGIN
l_count := DBMS_SPM.PACK_STGTAB_BASELINE (
table_owner => 'SPM',
enabled => 'YES',
fixed => 'YES');
END;
/
Pro tip: You can also use
dba_sql_plan_baselines to find plans

Prepare
Extract
Transfer
Load
Use Data Pump to transfer that single table
Optionally,
• Export to dump file
• Transfer using scp or rsync
• Import from dump file
SQL> CREATE DATABASE LINK cloud_link ... ;
$ impdp system network_link=cloud_link tables=SPM.SPB_STAGING
...

Prepare
Extract
Transfer
Load
Finally, load the baselines from the staging table into the data
dictionary
• This requires Enterprise Edition license on-prem
SQL> DECLARE
l_count NUMBER;
BEGIN
l_count := DBMS_SPM.UNPACK_STGTAB_BASELINE (
table_owner => 'SPM');
END;
/
Pro tip: You can apply filters to
limit the baselines to import

Additional resources
Documentation
Technical Brief, SQL Plan Management in Oracle Database 19c
Database 19c, SQL Tuning Guide, Overview of SQL Plan Management

Transport | SQL Profiles
141
Test Database
On-Prem Database
Compare
Analyze
Tune
Manage
Transport
Collect
Transport

Transporting | SQL Profiles
Prepare
Extract
Transfer
Load
SQL Profiles are stored in data dictionary
To transfer - profiles must be converted to a transportable format
and stored in a staging table
SQL> BEGIN
DBMS_SQLTUNE.CREATE_STGTAB_SQLPROF (
table_name => 'STAGING',
table_owner => 'SQLPROFILES');
END;
/

Prepare
Extract
Transfer
Load
Select the profiles that you want to transfer
To extract all profiles from DEFAULT category
SQL> BEGIN
DBMS_SQLTUNE.PACK_STGTAB_SQLPROF (
staging_table_name => 'STAGING',
staging_schema_owner => 'SQLPROFILES');
END;
/
Pro tip: You can filter on profile_name
and profile_category as well

Prepare
Extract
Transfer
Load
Use Data Pump to transfer that single table
SQL> CREATE DATABASE LINK cloud_link ... ;
$ impdp system network_link=cloud_link
tables=SQLPROFILES.STAGING ...

Prepare
Extract
Transfer
Load
Finally, load the profiles from the staging table into the data
dictionary
• This requires a Tuning Pack license on-prem
SQL> BEGIN
DBMS_SQLTUNE.UNPACK_STGTAB_SQLPROF (
staging_table_name => 'STAGING',
staging_schema_owner => 'SQLPROFILES',
replace => TRUE);
END;
/
Pro tip: You can load a SQL profile
into the same or higher release

Additional resources
Documentation
Database 19c, SQL Tuning Guide, Transporting a SQL Profile
How to Move SQL Profiles from One Database to Another (Including to Higher Versions)
(Doc ID 457531.1)

147
Photo
by
Eric
Nopanen
on
Unsplash

Want to Know More?
Photo
by
Chris
Briggs
on
Unsplash
Webinar: Performance Stability
Recording
Slides
Photo
by
Dušan
veverkolog
on
Unsplash

149
https://DOHdatabase.com

YouTube | Oracle Database Upgrades and Migrations
150
YouTube Channel

Recorded Web Seminars
https://MikeDietrichDE.com/videos/
151

HOL | https://MikeDietrichDE.com

6212883126866262792 performance testing_cloud

More Related Content

What's hot (20)

Similar to 6212883126866262792 performance testing_cloud (20)

Recently uploaded (20)

6212883126866262792 performance testing_cloud