Gleb Otochkin presented on connecting Oracle databases to Big Data platforms. He discussed using Oracle GoldenGate for real-time replication from Oracle to HDFS, Kafka, Flume, HBase, MongoDB and Cassandra. For batch loads, tools include Sqoop and Oracle Loader for Hadoop. Oracle Data Integrator and Oracle SQL Connector for HDFS can query Big Data sources. Business cases for integration include growing data volumes, unstructured data, data retention policies and cost savings.

1. There and back again.
or how to connect Oracle and Big Data.
SESSION ID
GLEB OTOCHKIN

2. Gleb Otochkin
Started to work with data in 1992
At Pythian since 2008
Area of expertise:
● Data Integration
● Oracle RAC
● Oracle engineered systems
● Virtualization
● Performance tuning
● Big Data
otochkin@pythian.com
@sky_vst
Principal Consultant
© The Pythian Group Inc., 2017 2

3. ABOUT PYTHIAN
Pythian’s 400+ IT professionals
help companies adopt and
manage disruptive technologies
to better compete
© 2016 Pythian. Confidential 3

4. Systems currently
managed by Pythian
EXPERIENCED
Pythian experts
in 35 countries
GLOBAL
Millennia of experience
gathered and shared
over 19 years
EXPERTS
11,800 2400
© The Pythian Group Inc., 2017 4

5. Journey of our data begins here.
© The Pythian Group Inc., 2017 5

6. AGENDA
© The Pythian Group Inc., 2017 6
• What do we call by name “Big Data”?
• Business cases.
• Real time replication from Oracle to Big Data.
• How to use Big Data in Oracle.
• What is next.
• QA

7. Big Data?
© The Pythian Group Inc., 2017 7

8. CLICK TO ADD
TITLE
Big Data ecosystem
Big Data?
Here in Texas we call it
just Data.

9. CLICK TO ADD
TITLE
SUBTITLE
© The Pythian Group Inc., 2017
Big Data ecosystem
VOLUME
ANALYSIS
ANY
STRUCTURE
PROCESSINGGROWS
COMPLEX
DISCOVERY
BIG

10. Some Big Data tools and terms.
Some terms and tools used in Big DATA:
• HDFS - Hadoop Distributed File System.
• Apache Hadoop - framework for distributed storage and processing.
• HBase - non-relational, distributed database.
• Kafka - streaming processing data platform.
• Flume - streaming, aggregation data framework.
• Cassandra - open-source distributed NoSQL database.
• Mongodb - open-source cross-platform document-oriented database.
• Hive - query and analysis data framework on top of Hadoop.
• Avro - data format widely used in BD (JSON+binary)
© The Pythian Group Inc., 2017 10

11. Business cases
© The Pythian Group Inc., 2017 11

12. Why do we need replication?
Business cases.

13. Business cases.
Operation activity on DB with rest on a Data Lake.
© The Pythian Group Inc., 2017 13
RDBMS
OLTP
RDBMS
OLTP
BD platform
Data
Preparation
Engine
BI &
Analysis

14. Business cases.
Operation activity on DB with Data in Big Data and BI on a RDBMS.
© The Pythian Group Inc., 2017 14
RDBMS
OLTP
RDBMS
OLTP
BD platform
Data
Preparation
Engine
BI &
AnalysisRDBMS

15. Business cases.
Operation and BI activity on DB with main Data body in a Big Data platform.
© The Pythian Group Inc., 2017 15
RDBMS
OLTP
RDBMS
OLTP
BD platform
BI &
Analysis
ODI & BD SQL

16. Business cases.
Operation and BI activity on DB with main Data body in a Big Data platform.
© The Pythian Group Inc., 2017 16
RDBMS
OLTP
RDBMS
OLTP
BD platforms
BI &
Analysis
Kafka KafkaKafka
RDBMS

17. Business cases.
Short summary
• How we connect different platforms:
•Data replication tools.
▪Oracle GoldenGate.
▪DBVisit.
▪Shareplex.
▪ELT tools
•Batch load tools.
▪Sqoop.
▪Oracle loader for Hadoop
•Data Integration tools.
▪Oracle Data Integrator.
•Presenting Big Data to RDBMS:
▪Oracle Big Data SQL.
▪Gluent.
• Reasons:
• Growing Data Volume.
• Unstructured data.
• Data retention policy.
• Cost.
© The Pythian Group Inc., 2017 17

18. Oracle to big data.
Real time replication OLTP data to a Big Data platform.

19. And what is wrong with Sqoop?

20. Batch offloading from Oracle to Big Data.
To HDFS.
• Sqoop one of the oldes.
• Works in both ways.
• Gluent is more than a just
batch tool.
© The Pythian Group Inc., 2017 20
SQOOP HDFS
HDFS
HDFS
HDFS
HDFS
Database

21. Replication to Big Data.
Sqoop import.
© The Pythian Group Inc., 2017 21
App
App
App
App Database
App
Sqoop HDFS
HDFS
HDFS
HDFS
HDFS
MapReduceJDBC
Hive
Generate MapReduce Jobs.
Connects through JDBC.
Good for batch processing.

22. Replication to Big Data.
Oracle Goldengate:
• Proc.
•Real time streaming.
•No or minimal impact to the source.
•Supports most of the data types.
•Different formats and conversion.
•Enterprise support
• Cons.
•Licensing fee.
•Closed source code.
© The Pythian Group Inc., 2017 22
Why Goldengate?

23. Replication to Big Data.
Replication to HDFS by Oracle GoldenGate.
© The Pythian Group Inc., 2017 23
App
App
App
App Database Tran
Log
App
Oracle
Goldengate
Oracle
Goldengate
HDFS
HDFS
HDFS
HDFS
HDFS
OGG Trail

24. Replication to Big Data.
Source side.
• Oracle 11.2.0.4 and up
• Archivelog mode.
• Supplemental logging:
•Minimal od DB level.
•On schema or table level.
• OGG user in database.
© The Pythian Group Inc., 2017 24
App
App
App
App Database Tran
Log
App

25. Replication to Big Data.
Oracle GoldenGate.
• OGG 12.2.0.1 and up.
• BD adapters with replicat.
• Supported BD targets:
•HDFS
•Kafka
•Flume
•HBase
•Mongodb (from 12.3)
•Cassandra (from 12.3)
• Different formats.
© The Pythian Group Inc., 2017 25
Tran
Log
Manager
OGG Trail
Extract
OGG Trail
Data Pump
Manager
Replicat

26. Replication to Big Data.
To HDFS.
• OGG 12.2.0.1 and up.
• BD adapters with replicat.
• Different formats.
• DML and DDL.
© The Pythian Group Inc., 2017 26
OGG Trail
Manager
Replicat
HDFS Client
HDFS
HDFS
HDFS
HDFS
HDFS

27. Replication to Big Data.
Test two.
• New columns:
•Operation type.
•Table name
•Local and UTC timestamp
orcl> select * from ggtest.test_tab_2;
ID RND_STR USE_DATE
1 BGBXRKJL 02/13/16 08:34:19
2 FNMCEPWE 08/17/15 04:50:18
© The Pythian Group Inc., 2017 27
hive> select * from BDTEST.TEST_TAB_2;
OK
I BDTEST.TEST_TAB_2 2016-10-25 01:09:16.000168 2016-10-24T21:09:21.186000 00000000120000004759 1 BGBXRKJL
2016-02-13:08:34:19 NULL
I BDTEST.TEST_TAB_2 2016-10-25 01:09:16.000168 2016-10-24T21:09:22.827000 00000000120000004921 2
FNMCEPWE 2015-08-17:04:50:18 NULL

28. Replication to Big Data.
To Kafka.
© The Pythian Group Inc., 2017 28
OGG Trail
Manager
Replicat
Kafka
producer
Kafka
topic
Kafka
topic
Zookeeper
Kafka
Consumer
Kafka
Consumer
Kafka
Consumer
Broker
Stream

29. Replication to Big Data.
Some notes about source:
• JSON,Text,Avro(different types) and
XML formats.
• Some strange behaviour with Avro.
• Hive support through text format.
• Trimming out leading or trailing
whitespaces.
• Truncates only as DML.
• Proper path to Java classes.
• We are replicating a log of changes.
• Only committed transactions.
• “Passive” commit.
• Different levels of supplemental
logging may lead to different
captured data (compressed DML).
• DDL only with following DML.
• CTAS is not supported.
© The Pythian Group Inc., 2017 29
and BD destination:

30. Replication to Big Data.
Some notes about Kafka
handler:
• Different topologies for Kafka.
• Topic partitioning to table.
• Topics are created automatically.
• Operation vs transactional mode.
• Blocking vs Non-Blocking.
• Version 0.9+.
• Schema definition can go to a
different topic(Avro).
© The Pythian Group Inc., 2017 30
and others:

31. Kafka connect and Kafka stream.
Confluent team and their additions.
• Kafka connect.
• Frame to build connector.
• https://www.confluent.io/product
/connectors/
• Main purpose is copy data to
and from Kafka.
© The Pythian Group Inc., 2017 31
• Kafka Stream.
• Incorporated to Kafka and works
together.
• https://www.confluent.io/product/kaf
ka-streams/
• Input data are transformed to output
data.

32. Replication to Big Data.
To Flume.
© The Pythian Group Inc., 2017 32
OGG Trail
Manager
Replicat
Flume
source
Flume
channel
HDFSFlume sink
Flume agent

33. Replication to Big Data.
Some notes about Flume handler:
• Configuration file path.
• Failover and load balancing (from
1.6.x for Avro).
• Kerberos security (from 1.6.x for
Thrift).
• Version 1.4+. (1.6.x better)
• Avro and Thrift formats.
• Schema changes go to a different
Flume event.
© The Pythian Group Inc., 2017 33
and others:

34. Replication to Big Data.
To HBase. MongoDB. Cassandra.
© The Pythian Group Inc., 2017 34
OGG Trail
Manager
Replicat
Hbase client HDFS
HBASE

35. Replication to Big Data.
Primary key for a source table in HBase.
orcl> alter table ggtest.test_tab_2 add constraint pk_test_tab_2 primary key (pk_id);
Table altered.
orcl> insert into ggtest.test_tab_2 values(9,'PK_TEST',sysdate,null);
© The Pythian Group Inc., 2017 35
• Having supplemental logging for all columns:
•Row id as concatenation of values for all columns.
• Adding a primary key and supplemental logging for keys:
•Row id is primary key.

36. Replication to Big Data.
Two cases. All columns supplemental logging and PK supplemental logging.
© The Pythian Group Inc., 2017 36
hbase(main):012:0> scan 'BDTEST:TEST_TAB_2'
ROW COLUMN+CELL
7|IJWQRO7T|2013-07-07:08:13:52 column=cf:ACC_DATE, timestamp=1459275116849, value=2013-07-07:08:13:52
7|IJWQRO7T|2013-07-07:08:13:52 column=cf:PK_ID, timestamp=1459275116849, value=7
7|IJWQRO7T|2013-07-07:08:13:52 column=cf:RND_STR_1, timestamp=1459275116849, value=IJWQRO7T
8|TEST_INS1|2016-03-29:15:14:37|TEST_ALTER column=cf:ACC_DATE, timestamp=1459278884047, value=2016-03-29:15:14:37
8|TEST_INS1|2016-03-29:15:14:37|TEST_ALTER column=cf:PK_ID, timestamp=1459278884047, value=8
8|TEST_INS1|2016-03-29:15:14:37|TEST_ALTER column=cf:RND_STR_1, timestamp=1459278884047, value=TEST_INS1
8|TEST_INS1|2016-03-29:15:14:37|TEST_ALTER column=cf:TEST_COL, timestamp=1459278884047, value=TEST_ALTER
9 column=cf:ACC_DATE, timestamp=1462473865704, value=2016-05-05:14:44:19
9 column=cf:PK_ID, timestamp=1462473865704, value=9
9 column=cf:RND_STR_1, timestamp=1462473865704, value=PK_TEST
9 column=cf:TEST_COL, timestamp=1462473865704, value=NULL

37. Replication to Big Data.
Some notes about HBase handler:
• Hbase-site.xml in the classpath.
• GROUPTRANSOPS batching for
performance.
• Kerberos security (from 1.6.x for
Thrift).
• Version 1.1+. (0.98 with compat
parameter)
• HBase row key mapping.
• Only single column family.
© The Pythian Group Inc., 2017 37
and others:

38. Replication to Big Data.
Some notes about Cassandra
handler:
• Does not allow the _id column to be
modified.
• Undo for bulk transactions (requires
full before image for deletes in OGG).
• Primary key updates are deletes and
inserts (watch compressed updates in
OGG).
• Does not automatically create
keyspaces (but creates tables)
• Primary keys are immutable.
• Inserts and Updates are different
than in traditional databases.
• Primary key updates are deletes and
inserts (watch compressed updates
in OGG).
© The Pythian Group Inc., 2017 38
and Mongodb:

39. Other tools.
What can be used as an alternative.
• DBvisit.
• Replicate connector to Kafka.
• Confluent version of Kafka.
• http://www.dbvisit.com/replicate
_connector_for_kafka
• Has similar functionality to
Oracle GoldenGate.
© The Pythian Group Inc., 2017 39
• Dell Shareplex.
• Replication to Kafka.
• Supports most of datatypes.
• http://documents.software.dell.com/
SharePlex/8.6.5/release-notes/syst
em-requirements
• https://documents.software.dell.com
/shareplex/8.6.4/administration-guid
e/configure-replication-to-open-targ
et-targets/configure-replication-to-a-
kafka-target

40. Back again
Using data from Big Data in Oracle

41. From Big Data to Oracle.
© The Pythian Group Inc., 2017 41
Tool Filtering Data
Conversion
Query Offload
query
Parallel
Sqoop Oracle Oracle Yes
ODBC gateway Hadoop Oracle Yes Yes
Oracle loader
for HDFS
Oracle Hadoop Yes
Oracle SQL
Connector for
HDFS
Oracle Oracle Yes Yes
BD SQL Hadoop Hadoop Yes Yes Yes
ODI KM KM Yes Yes
Gluent Hadoop Hadoop Yes Yes Yes

42. Oracle Data Integrator.
ODI vs traditional ETL.
Intermediate staging
and transformation
Source I
Source II
Target
Extract Transform Load
Load engine
Source I
Source II Target
Extract TransformLoad
Transform
© The Pythian Group Inc., 2017 42

43. Oracle Data Integrator.
How it looks.
Weblogic
Staging
schema
Oracle DB I
HDFS
Work
repository
Oracle DB II
CSV text
ODI Agent
Target
schema
Master
repository
ODI Studio
© The Pythian Group Inc., 2017 43

44. Oracle Data Integrator.
ODI workflow.
Data model
Project I
Designer
Project II
Data model
Data model
Data model
Logical
schema I
Logical view
Logical
schema II
Prod Batch
Context
Prod
Stream
PROD
Physical
PROD
Test
PROD
Stream
PROD
Stream
Test env
Test env
© The Pythian Group Inc., 2017 44

45. Oracle Data Integrator.
ODI logical mapping
• Logical mapping is
separated from
physical
implementation.
© The Pythian Group Inc., 2017 45

46. Oracle Data Integrator.
ODI physical mapping.
• Different physical
mapping for the
same logical map
• Different
knowledge
modules.
© The Pythian Group Inc., 2017 46

47. ODI
Oracle Data Integrator.
• Knowledge modules for most
platforms.
• Uses filters, mappings, joins and
constraints.
• Batch or event(stream) oriented
integration.
• ELT:
• Extract.
• Load.
• Transform.
• Logical and physical models separation.
• Knowledge modules:
• RKM - reverse engineering.
• LKM - load.
• IKM - integration.
© The Pythian Group Inc., 2017 47

48. Oracle Big Data SQL
Architecture.
© The Pythian Group Inc., 2017 48
Oracle Database Hadoop
CDH or HDP
management server
Hadoop
BD SQL
agent/service
BD SQL
Exadata technology for Big Data

49. Oracle Big Data SQL.
What it can do for us.
• Smart scan:
• Storage indexes.
• Bloom filters.
• Works with:
• Apache Hive.
• HDFS
• Oracle NoSQL Database.
• Apache HBase
• Fully supports SQL syntax.
• Predicate push down.
© The Pythian Group Inc., 2017 49

50. Oracle Big Data SQL.
Views and packages to use.
• Smart scan:
• Storage
indexes.
• Bloom filters.
© The Pythian Group Inc., 2017 50
orcl> select cluster_id,database_name, owner, table_name from all_hive_tables where database_name='bdtest';
CLUSTER_ID DATABASE_NAME OWNER TABLE_NAME
bigdatalite bdtest oracle test_tab_1
bigdatalite bdtest oracle test_tab_2
PROCEDURE CREATE_EXTDDL_FOR_HIVE
Argument Name Type In/Out Default?
------------------------------ ----------------------- ------ --------
CLUSTER_ID VARCHAR2 IN
DB_NAME VARCHAR2 IN
HIVE_TABLE_NAME VARCHAR2 IN
HIVE_PARTITION PL/SQL BOOLEAN IN
TABLE_NAME VARCHAR2 IN
PERFORM_DDL PL/SQL BOOLEAN IN DEFAULT
TEXT_OF_DDL CLOB OUT

51. Oracle Big Data SQL.
Building the external table.
• Hive table can be queried in
Oracle now.
© The Pythian Group Inc., 2017 51
CREATE TABLE test_tab_2 ( tran_flag VARCHAR2(4000), tab_name VARCHAR2(4000), ………..)
ORGANIZATION EXTERNAL (TYPE ORACLE_HIVE DEFAULT DIRECTORY DEFAULT_DIR ACCESS PARAMETERS
( com.oracle.bigdata.cluster=bigdatalite
com.oracle.bigdata.tablename=bdtest.test_tab_2) ) PARALLEL 2 REJECT LIMIT UNLIMITED
TRAN_FLAG ID RND_STR USE_DATE
---------- ---------- ---------- --------------------
I 1 BGBXRKJL 2016-02-13:08:34:19
I 2 FNMCEPWE 2015-08-17:04:50:18
I 1 BGBXRKJL 2016-02-13:08:34:19

52. Oracle Big Data SQL.
What it can do for us.
• External table FULL access.
© 2016 Pythian. Confidential 52

53. Oracle Big Data SQL.
Options for BD SQL.
• Oracle Copy to Hadoop utility.
•Using Data Pump format.
•ORACLE_DATAPUMP access driver.
•CTAS to external table.
•hadoop fs -put …
•Hive “INPUTFORMAT
'oracle.hadoop.hive.datapump.DPInputFormat'”
•Oracle Shell for Hadoop Loaders
• ACCESS PARAMETERS :
• com.oracle.bigdata.overflow
• com.oracle.bigdata.tablename
• com.oracle.bigdata.colmap
• com.oracle.bigdata.erroropt
• Directly to HDFS :
• TYPE oracle_hdfs.
© The Pythian Group Inc., 2017 53

54. Other tools.
What can be used as an alternative.
• Gluent Data Platform.
• Can offload data to Hadoop.
• Allows query data of any data
source in Hadoop.
• Offload engine .
• Supports Cloudera, Hortonworks or
any Hadoop with Impala or or Hive
SQL engine installed.
https://gluent.com/
© The Pythian Group Inc., 2017 54

55. And here we have returned back.
© The Pythian Group Inc., 2017 55

56. QA
© The Pythian Group Inc., 2017 56

57. THANK YOU
Email: otochkin@pythian.com
https://pythian.com/blog
@sky_vst
© The Pythian Group Inc., 2017 57