DOAG2021 DataPumpDeepDive

Deep Dive into Oracle Data Pump Internals
Bill Beauregard Roy Swonger

Senior Principal Product Manager
Vice President
Data Pump and Utilities
Database Upgrade, Patching & Utilities
Oracle
Oracle
Roy F. Swonger
Vice President
Database Upgrade,
Utilities & Patching
@RoyFSwonger
royfswonger
William Beauregard
Senior Principal Product Manager

Data Pump and Utilities
william-beauregard-3053791
• How Data Pump works
• Parallel Operations
Data Pump • Data Pump Data Movement Methods
Internals
• Oracle Data Pump CONTROL Table
• Marker Objects and Excluding Statistics
Topics • Oracle Data Pump Export & Import Restart

• Include and Exclude Objects in the Same Job
• Helpful Interactive Commands
• Troubleshooting Oracle Data Pump Jobs Using logfile
diagnostics and the TRACE Parameter
How Data Pump
Works
Copyright © 2021, Oracle and/or its affiliates

How Data Pump Works
3 main User Components
• The Command-line clients, expdp and impdp
• The DBMS_DATAPUMP PL/SQL package, the Data Pump API
• The DBMS_METADATA PL/SQL package, the Metadata API
6 Copyright © 2021, Oracle and/or its affiliates

How Data Pump Works
The Clients
• expdp and impdp invoke the Export and Import utilities, respectively
• A shadow process is created when the client logs in to database
• The shadow process calls the DBMS_DATAPUMP.OPEN API.

• Creates the control table and control process.
• The clients use the Data Pump API to execute export and import commands
• DBMS_METADATA used for extracting, manipulating, & re-creating dictionary metadata
Note: DBMS_DATAPUMP & DBMS_METADATA can be used independent of the clients

Object Paths
How is metadata referred to?
An “Object Path” describes a particular category of metadata, such as
• TABLE_EXPORT/TABLE/STATISTICS/TABLE_STATISTICS
• SCHEMA_EXPORT/SYSTEM_GRANT
• DATABASE_EXPORT/AUDIT_UNIFIED/AUDIT_POLICY
Used in INCLUDE, EXCLUDE, and internally when distributing work to processes
See these views for full list of available paths

• <TABLE|SCHEMA|DATABASE>_EXPORT_OBJECTS

Data Pump: Component Block Diagram

Data Pump
Parallel
Operations

Data Pump: 2 kinds of Parallelism for Dumpfiles
Overview
• Inter-partition parallelism
• For tables with partitions below a particular threshold in size
• One connection and worker per partition or sub partition
• Parallel query (PQ) intra-partition parallelism

• Invoked for larger tables or partitions
• Worker process is the query coordinator, and not included in the DOP limit

Data Pump: Parallel Operations
Overview
• A job starts w/ a minimum of 2 processes: a Control Process (CP) + 1 Worker
• The user can specify a Degree of Parallelism (DOP) for the Data Pump job
• DOP = maximum number of ACTIVE parallel processes (Active workers + PQ processes)
• DOP does not include the CP or shadow process
• Any additional workers are idle and wait for work
• Data Pump will only start the number of threads needed to complete the task
• Control Process: Verifies parameters & job description, controls & assigns work items to
workers

Data Pump 19c | Set Max Data Pump Jobs & Parallelism
Control the maximum number of jobs and degree of parallelism to control resource
• MAX_DATAPUMP_JOBS_PER_PDB database parameter (Changed)
• Default: 100, Range: 0 to 250 or Auto - 50% of SESSIONS
• Must be same for each RAC instance, dynamic, and modifiable per-PDB
• MAX_DATAPUMP_PARALLEL_PER_JOB database parameter (New)

• Default: 50, Range: 1 to 250 or Auto - 25% of SESSIONS
• Can be different for each RAC instance, dynamic, and modifiable per-PDB
Documentation: Database Utilities Guide
Copyright © 2021 Oracle and/or its affiliates.

Data Pump: Parallel Worker Processes
What do the parallel worker processes do?
Export Import
• Collects object metadata w/ Converts metadata from XML to
DBMS_METADATA API DDL & PL/SQL w/
Metadata
• Writes metadata to a DBMS_METADATA API
dumpfile as XML
Data • Uses the Data Layer to unload Uses the Data Layer to load data
data
• Network mode Import uses DBMS_METADATA API to convert source database metadata
for execution and moves data to the target

Data Pump: TABLE_DATA & Metadata Units of Work
How is work assigned to parallel Worker processes?
• TABLE_DATA work unit is:
• Subpartition for subpartitioned tables
• Partition for partitioned tables
• Table for non-partitioned tables
• Specify all data for the table as one work unit, regardless of partitioning w/
DATA_OPTIONS=GROUP_PARTITION_TABLE_DATA
• The metadata work unit is a single object path

Data Pump
Parallel Export

Data Pump: How Export works
Overview
• Export (and network mode import) work in multiple phases
• Metadata
• Each object path is a metadata work unit
• Workers begin collecting metadata for export
• Metadata unloaded in any order with parallel worker processes
• Data
• Metadata for TABLE_DATA items contain a size estimate that drives order of export
• Data work items are scheduled as worker processes become available.
Pro Tip: current dictionary and object statistics
are crucial to data pump export performance!

Data Pump | Parallel Metadata Export Comparison
Metadata export happens concurrent w/ estimate phase for table data
Most metadata & data objects are exported in parallel when PARALLEL=2 or greater
Pre-12.2 Since 12.2

• Start with ESTIMATE phase • Export collects information as it progresses
•
• Gather table data objects • Determines number of workers for metadata
• export
• Other workers remain idle until data
objects are gathered • Reserves workers for data export & begins
• metadata export
• Metadata exported serially
• • Simultaneously begins estimating data export
Data exported in parallel
• Documentation: Database Utilities Guide

Data Pump 12.2 | Parallel Metadata Export
Logfile Differences
12.1.0.2 12.2.0.1

Data Pump | Parallel Metadata Export
Details
• Source database must be 12.2 or higher
• ESTIMATE phase uses STATISTICS only (pre-12.2 it counted blocks)
• Restart works as always
• Database metadata objects that have dependencies are imported serially
• Types (due to inheritance), schemas and procedural actions.

Data Pump | Export Best Practices
Details
1. Collect statistics using the dbms_stats package: gather_table_stats,
gather_schema_stats, gather_database_stats,gather_dictionary_stats
2. Convert LOBs to SecureFiles LOBs
3. Include wildcard characters in the dumpfile filename specification
• %U for 1 to 99 dumpfiles (2-digit, fixed-width, incrementing integer)
• %L for 1 to more than 99 dumpfiles (3-digit to 10-digit, variable-width integer)
• Always use wildcards when exporting with PARALLEL
4. Specify maximum dumpfile size using FILESIZE

Data Pump | Export Example
Processes started for a large data export using PARALLEL = 6
PROCESS ACTIVITY PARALLELISM

CP Control process is not parallelized NA
Worker1 Unload metadata 1 parallel process
Worker2 Unload sub-partition 2 degrees of parallelism

user1:tab1:subpart1
• 1 parallel PX process
• Worker2_PX1: Unload data • 1 parallel PX process
• Worker2_PX2: Unload data
Worker3 Unload subpartition 1 degree of parallelism
user1:tab1:subpart2
Worker4 Unload subpartition 2 degrees of parallelism
user1:tab1:subpart3
• 1 parallel PX process
• Worker4_PX1 – Unload data • 1 parallel PX process
• Worker4_PX2 – Unload data
Data Pump
Parallel Import

Data Pump: How Import works
• No estimate phase
• First loads the metadata, mostly in parallel
• Builds multiple indexes simultaneously w/ data pump workers
• Individual indexes are built serially.
• Can be a performance issue for indexes on very large tables
• Loads package bodies in parallel using multiple workers
• Loads data in parallel using multiple workers and multiple PQ processes.

Data Pump | Parallel Metadata Import
Metadata export happens concurrent w/ estimate phase for table data
Documentation:Pre-12.2
Database Utilities Guide Since 12.2
• One worker per partition/subpartition • Added parallel import of most other metadata
• PQ used if partitions are large enough objects
•
• Package bodies loaded in parallel
• Some exceptions
• •
With patch for bug 22273229 • Types (due to inheritance)
• •
• Indexes built in parallel • Schemas
• •
• Constraints created in parallel • Procedural actions
•
• Backport to 12.1.0.2, 11.2.0.4

Internals
XML Doc 1
• Metadata is exported in XML documents into dumpfile User1
• Each XML document contains N objects of a given type User2 Worker 1
• One XML document allocated to a worker at a time …
• Example: 161 USER objects to import User80
• 2 workers import 80 users per XML document XML Doc 2
User81
• 1 worker imports 1 user, other workers may be idle Worker 2
User82
• Notes: …
• Works for conventional (dumpfile) jobs (& TTS since 21c) User160
• Restart works same as always XML Doc 1
• Status command will show multiple workers on metadata User161 Worker 3

Pro Tip: export and import
parallelism do not have to match
Logfile
• Comparison with PARALLEL=8 for 27586 object grants and METRICS=Y
• 12.1.0.2
• 12.2.0.1

Performance
• Examples from E-Business Suite test database *import time in seconds

Data Pump Data
Movement
Methods

Data Pump | Data Movement Methods
Overview
• 3 most common approaches
• Direct Path
• External Table
• Insert as Select over a network link.
• Conventional access method used in some import edge cases
• Syntax (defaults to AUTOMATIC)
ACCESS_METHOD=[AUTOMATIC | DIRECT_PATH | EXTERNAL_TABLE]

Internals
• External table method does not support moving LONG or LONG RAW
• Direct path not allowed when importing into pre-existing partitioned tables
• CP chooses optimal method based on table size, job size & DOP
• One large table is faster w/ external tables, PX processes & maximum parallel value
• Many smaller tables are usually faster to use Direct Path
• Desirable to export the larger tables first especially if parallel value > 1
Pro Tip: current table stats will help

Data Pump make the right choice!

What if you choose the wrong method?
Example: table with a LONG RAW column
impdp system/manager tables=scott.foo_long directory=dpump_dir \
dumpfile=s.dmp access_method=external_table
…
…
Processing object type TABLE_EXPORT/TABLE/TABLE_DATA
ORA-31696: unable to export/import TABLE_DATA:“SCOTT"."FOO_LONG" using client
specified EXTERNAL_TABLE method

Data Pump
CONTROL Table

Data Pump | Control Table
Used to keep track of Data pump job progress
• Record of job parameters, current status, object information, index into the dump files
• Can be accessed from SQL*Plus
• Upon successful job completion
• CONTROL TABLE deleted by default, Use KEEP_MASTER=Y to keep anyway
• Table written to the last dump file(s) in the set (depends on size of table & file size)
• First object imported to determine what was exported & what to import
• Table includes rows describing objects, type completions, and job info

Data Pump | Control Table
Benefits
• Can be queried for job status
• Information in the table accelerates importing a subset of the objects
• Allows job restart if stopped or encounters fatal error, such as insufficient disk space

Data Pump Marker
Objects and
Excluding Statistics

Data Pump | Marker Objects
Purpose
• Used by Export & Import to order (sequence) actions for other object types
• Indicate a point during the job where some action is required
• Cannot be an excluded object type b/c they are used for several internal purposes
• Examples:
• Mark the dependency that package bodies have on package specifications
• If statistics are included in the export a marker is set to trigger a call to the
DBMS_STATS package to import statistics on objects imported.
• If EXCLUDE=STATISTICS is specified on export all objects in an object path for
statistics are excluded, including the STATISTICS marker

Data Pump
Export Restart

Data Pump | Export Restart
Internals
• Export job can usually be restarted unless the estimate phase is not completed
• Requires the CONTROL TABLE
• Workers create & update “type completion” rows for each object type w/ start &
completion time stamps
• During a restart, a worker process looks for rows with no completion time.
• In the table below “USER” is incomplete, so Worker
• Removes user objects from dumpfile
• Configures the Metadata API to start after TABLESPACE
OBJECT_TYPE START_TIME COMPLETION_TIME

TABLESPACE 12-SEP-2021:9:04.01 12-SEP-2021:9:05.23
USER 12-SEP-2012:9:05.27

Data Pump
Import Restart

Data Pump | Import Restart
Internals
• Import job can be restarted using the CONTROL TABLE
• Workers check object rows’ state & status fields to detect what’s been imported
OBJECT OBJECT_SCHEMA OBJECT_NAME PROCESSING_STATE PROCESSING_STATUS
TABLE SCOTT FOO W C

TABLE SCOTT BAR U C
INDEX SCOTT FOO_IND R C
INDEX SCOTT BAR_IND R C
R = objects were Retrieved (exported)
C = objects are Current (successfully imported)
W = objects are written (imported)
U = objects are unknown (import started but did not finish)

Data Pump
Include and
Exclude Objects

Data Pump 21c | Include & Exclude objects in Same Job
Be more specific about job contents
• Include and exclude objects within the same export or import job
• Example:
expdp hr DIRECTORY=dpump_dir1 DUMPFILE=exp%u.dmp SCHEMAS=hr,oe
include=table exclude=statistics
• Details:
• INCLUDE parameter(s) processed first, include all objects identified by the parameter
• EXCLUDE parameter(s) processed next. It removes any objects in the list of include
objects

Data Pump 21c | Include & Exclude objects in Same Job
Example
include=table
exclude=statistics
…
Processing object type SCHEMA_EXPORT/TABLE/TABLE_DATA
Total estimation using BLOCKS method: 448 KB
Processing object type SCHEMA_EXPORT/TABLE/TABLE
Processing object type SCHEMA_EXPORT/TABLE/PRE_TABLE_ACTION
Processing object type SCHEMA_EXPORT/TABLE/GRANT/OWNER_GRANT/OBJECT_GRANT
Processing object type SCHEMA_EXPORT/TABLE/COMMENT
Processing object type SCHEMA_EXPORT/TABLE/INDEX/INDEX
Processing object type SCHEMA_EXPORT/TABLE/CONSTRAINT/CONSTRAINT
Processing object type SCHEMA_EXPORT/TABLE/INDEX/STATISTICS/INDEX_STATISTICS
Processing object type SCHEMA_EXPORT/TABLE/CONSTRAINT/REF_CONSTRAINT
Processing object type SCHEMA_EXPORT/TABLE/TRIGGER
Processing object type SCHEMA_EXPORT/TABLE/STATISTICS/TABLE_STATISTICS
Processing object type SCHEMA_EXPORT/TABLE/POST_TABLE_ACTION

Data Pump
Helpful Interactive
Commands

Data Pump | Helpful interactive commands
Type ^c while the job is running to interact with the job via the interactive CLI
• Get the job and worker status
• Includes Operation, Mode, State, Percent Done, and Current Parallelism
Import> STATUS
• Change the parallelism for current job
• Increases almost immediately. Decreases after the current assignment is complete
Import> PARALLEL=n
• Add tracing to a running job
• Useful if if a job seems to be hung or performing slowly
Import> trace=XXXXXX

Troubleshooting Jobs
Using Logfile Diagnostics

Data Pump | Logfile Diagnostics
Log the number of objects exported and elapsed time
•Add diagnostics to your logfile
expdp system/manager full=y directory=dpump_dir dumpfile=full.dmp\ metrics=yes
logtime=all
• Without METRICS=Y:
Processing object type DATABASE_EXPORT/FINAL_POST_INSTANCE_IMPCALLOUT/MARKER
Processing object type DATABASE_EXPORT/AUDIT_UNIFIED/AUDIT_POLICY_ENABLE
Processing object type DATABASE_EXPORT/POST_SYSTEM_IMPCALLOUT/MARKER
. . exported "SYS"."KU$_USER_MAPPING_VIEW" 5.890 KB 25 rows
. . exported "SYSTEM"."REDO_DB" 25.59 KB 1 rows
• With METRICS=Y:
DATABASE_EXPORT/FINAL_POST_INSTANCE_IMPCALLOUT/MARKER
02-NOV-21 19:43:56.064: W-1 Completed 1 MARKER objects in 0 seconds
02-NOV-21 19:43:59.171: W-1 Processing object type DATABASE_EXPORT/AUDIT_UNIFIED/AUDIT_POLICY_ENABLE
02-NOV-21 19:43:59.195: W-1 Completed 2 AUDIT_POLICY_ENABLE objects in 0 seconds
02-NOV-21 19:43:59.380: W-1 Processing object type DATABASE_EXPORT/POST_SYSTEM_IMPCALLOUT/MARKER
02-NOV-21 19:43:59.387: W-1 Completed 1 MARKER objects in 0 seconds
02-NOV-21 19:43:59.830: W-1 . . exported "SYS"."KU$_USER_MAPPING_VIEW" 5.890 KB 25 rows in 0 seconds using external_table
02-NOV-21 19:43:59.923: W-1 . . exported "SYSTEM"."REDO_DB" 25.59 KB 1 rows in 0 seconds using direct_path

Troubleshooting Jobs
Using the TRACE
Parameter

Data Pump | TRACE
How to create a Data Pump trace file
• Requires privileged user - DBA role or EXP[/IMP]_FULL_DATABASE Role
• ensure MAX_DUMP_FILE_SIZE is large enough to capture the trace (default=unlimited)
• Use parameter: TRACE= 8-digit hexadecimal mask

• First three digits specify a Data Pump component (case insensitive)
• Last four digits are usually: 0300
• Leading zero can be omitted
• MOS Note:
Export/Import DataPump Parameter TRACE How to Diagnose Oracle Data Pump [ID
286496.1]

Data Pump | TRACE Bit Masks
Bit Mask for tracing each component
Trace DM DW ORA Lines
level trc trc trc in
(hex) file file file trace Purpose
------- ---- ---- ---- ------ -----------------------------------------------
10300 x x x SHDW: To trace the Shadow process (API) (expdp/impdp)
20300 x x x KUPV: To trace Fixed table
40300 x x x 'div' To trace Process services
80300 x KUPM: To trace Master Control Process (MCP) (DM)
100300 x x KUPF: To trace File Manager
200300 x x x KUPC: To trace Queue services
400300 x KUPW: To trace Worker process(es) (DW)
800300 x KUPD: To trace Data Package
1000300 x META: To trace Metadata Package
2000300 x PLTS: To trace Pluggable Tablespaces
4000300 x DPEX: To trace Callouts
8000300 x META: To trace Utilities
10000300 x DPEX: To trace extensibility (procedural actions)
------- 'Bit AND'
1FFF0300 x x x 'all' To trace all components (full tracing)
---------
See: Export/Import DataPump Parameter TRACE - How to Diagnose Oracle Data Pump (Doc ID 286496.1)

Data Pump | TRACE
Start by tracing the Control Process and the Worker Process(es)
expdp <LOGIN>/<PASSWORD> DIRECTORY=<DIRECTORY_NAME>
DUMPFILE=<DUMP_NAME>.dmp LOGFILE=<LOG_NAME>.log SCHEMAS=<SCHEMA_NAME>\
TRACE=480300
• 'Bit AND’ of Control Process 80300 & Worker 400300 components
Creates 2 trace files:

• <SID>_dm<number>_<process_id>.trc - Control process trace
• <SID>_dw<number>_<process_id>.trc - Worker trace file (one for each worker)
Files are written to the directories specified by the init.ora/spfile initialization parameters
BACKGROUND_DUMP_DEST and USER_DUMP_DEST or in <ADR_HOME>/trace

Data Pump | TRACE
Tracing ALL components (if Oracle Support requests this )
impdp <LOGIN>/<PASSWORD> DIRECTORY=<DIRECTORY_NAME> \

DUMPFILE=<DUMP_NAME>.dmp LOGFILE=<LOG_NAME>.log FULL=y \
TRACE=1fff0300
Creates 2 trace files

• <SID>_dm<number>_<process_id>.trc - Control process trace
• <SID>_dw<number>_<process_id>.trc - Worker trace file
Also creates 1 trace file in USER_DUMP_DEST:

• <SID>_ora_<process_id>.trc - Shadow Process trace file

Data Pump | How to start tracing the Data Pump job
Use the TRACE parameter in the job command
expdp <LOGIN>/<PASSWORD> DIRECTORY=<DIRECTORY_NAME> \

DUMPFILE=<DUMP_NAME>.dmp LOGFILE=<LOG_NAME>.log \
TABLES=<SCHEMA_NAME>.<TABLE_NAME> \
TRACE=480300

Add tracing with an interactive command
Type ^c while the job is running to interact with the job via the interactive CLI
• Restart the job and add tracing
• Useful if if a job seems to be hung or performing slowly
• Can be changed in the middle of a job
impdp user/password attach=user.imp_job_1 trace=xxxxxx

Import> start_job [=SKIP_CURRENT=YES]

Use a database init.ora/spfile event
• Add the following line to init.ora parameter file
EVENT="39089 trace name context forever, level 0x300"

• Restart the database
• Start the Export Data Pump or Import Data Pump job
Event 39089 can also be enabled and disabled at system level
ALTER SYSTEM SET EVENTS = '39089 trace name context forever, level 0x300' ;
ALTER SYSTEM SET EVENTS = '39089 trace name context off' ;

Data Pump | SQL Tracing
Create a SQL trace file for an Export Data Pump or Import Data Pump job
• SQL trace file created for a specific process (usually the Worker process).
• Ensure directory has enough free space using init.ora/spfile parameter BACKGROUND_DUMP_DEST
• Event Levels
• event 10046, level 1 = enable standard SQL_TRACE functionality
• event 10046, level 4 = as level 1, plus trace the BIND values
• event 10046, level 8 = as level 1, plus trace the WAITs
• event 10046, level 12 = as level 1, plus trace the BIND values and the WAITs
A level 8 or 12 SQL trace file requires init.ora/spfile parameter TIMED_STATISTICS set to TRUE before the
event is set and before the Data Pump job is started

Data Pump | SQL Tracing
Create a SQL trace file for an Export Data Pump or Import Data Pump job
Example:
• Trace Worker process (400300) with standard SQL_TRACE functionality (1)
expdp <LOGIN>/<PASSWORD> DIRECTORY=<DIRECTORY_NAME> DUMPFILE=<DUMP_NAME>.dmp
\
LOGFILE=<LOG_NAME>.log TABLES=<SCHEMA_NAME>/<TABLE_NAME> \
TRACE=400301

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

YouTube | Oracle Database Upgrades and Migrations
https://MikeDietrichDE.com
https://DOHdatabase.com
YouTube Channel
Oracle LiveLabs hands-on workshops

https://livelabs.oracle.com

Thank you!

DOAG2021 DataPumpDeepDive

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

DOAG2021 DataPumpDeepDive

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DOAG2021 DataPumpDeepDive

Uploaded by

Copyright:

Available Formats

Deep Dive into Oracle Data Pump Internals

Bill Beauregard Roy Swonger

Senior Principal Product Manager

Topics • Oracle Data Pump Export & Import Restart

Copyright © 2021, Oracle and/or its affiliates

• The DBMS_DATAPUMP PL/SQL package, the Data Pump API

• The DBMS_METADATA PL/SQL package, the Metadata API

6 Copyright © 2021, Oracle and/or its affiliates

• A shadow process is created when the client logs in to database

• The shadow process calls the DBMS_DATAPUMP.OPEN API.

• DBMS_METADATA used for extracting, manipulating, & re-creating dictionary metadata

Note: DBMS_DATAPUMP & DBMS_METADATA can be used independent of the clients

7 Copyright © 2021, Oracle and/or its affiliates

Used in INCLUDE, EXCLUDE, and internally when distributing work to processes

See these views for full list of available paths

8 Copyright © 2021, Oracle and/or its affiliates

9 Copyright © 2021, Oracle and/or its affiliates

Copyright © 2021, Oracle and/or its affiliates

• Parallel query (PQ) intra-partition parallelism

11 Copyright © 2021, Oracle and/or its affiliates

12 Copyright © 2021, Oracle and/or its affiliates

• MAX_DATAPUMP_PARALLEL_PER_JOB database parameter (New)

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

13 Copyright © 2021, Oracle and/or its affiliates

• The metadata work unit is a single object path

14 Copyright © 2021, Oracle and/or its affiliates

Copyright © 2021, Oracle and/or its affiliates

16 Copyright © 2021, Oracle and/or its affiliates

Pre-12.2 Since 12.2

Copyright © 2021 Oracle and/or its affiliates.

Copyright © 2021 Oracle and/or its affiliates.

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

PROCESS ACTIVITY PARALLELISM

Worker1 Unload metadata 1 parallel process

Worker2 Unload sub-partition 2 degrees of parallelism

Copyright © 2021, Oracle and/or its affiliates

23 Copyright © 2021, Oracle and/or its affiliates

Copyright © 2021 Oracle and/or its affiliates.

Documentation: Database Utilities Guide

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

Copyright © 2021, Oracle and/or its affiliates

Documentation: Database Utilities Guide

Copyright © 2021 Oracle and/or its affiliates.

Pro Tip: current table stats will help

30 Copyright © 2021, Oracle and/or its affiliates

31 Copyright © 2021, Oracle and/or its affiliates

Copyright © 2021, Oracle and/or its affiliates

33 Copyright © 2021, Oracle and/or its affiliates

34 Copyright © 2021, Oracle and/or its affiliates

Copyright © 2021, Oracle and/or its affiliates