DAT342

Lessons from migrating Oracle databases to

Amazon Aurora
Abhinav Singh Bijoy Nair
Database Engineer – DMS Software Development
Amazon Web Services Manager – DMS
Amazon Web Services

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Agenda
Introduction to AWS Database Migration Service (AWS DMS) and AWS
Schema Conversion Tool

AWS DMS and AWS SCT product highlights

Amazon use case – Migrating Oracle databases to Amazon Aurora

(PostgreSQL/MySQL)

Q&A
What to expect from this session
• AWS resources with regards to migration & replication

• How Amazon approached move-off Oracle initiative

• Top migration challenges faced by Amazon.com

What not to expect

• Schema conversion issues from Oracle

• How stuff works in PostgreSQL/MySQL

• Oracle vs. PostgreSQL vs. MySQL feature comparison

Related breakouts
DAT214 - Automation framework to migrate relational databases
DAT343 - Lessons from migrating SQL Server databases to Amazon Aurora
DAT345 - Assessing and categorizing your database migrations
DAT360 - Analytical use cases with AWS DMS
DAT362-R1- Dive deep into AWS SCT and AWS DMS (Jobvite use case)
DAT363 - Migrating your data warehouses to Amazon Redshift
DAT366 - Running Oracle on Amazon RDS and migrating to Aurora PostgreSQL
DAT367 - Running SQL Server on Amazon RDS and migrating to Aurora PostgreSQL
DAT377 - Getting started with AWS DMS and AWS SCT
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
AWS migration tooling
Our goal: Enable customers the freedom to choose the best data
platform for their needs #DBFreedom

AWS Schema Conversion Tool (AWS SCT) converts your

commercial database and data warehouse schemas to open-
source engines or AWS native services, such as Amazon Aurora
and Amazon Redshift

AWS Database Migration Service (AWS DMS) easily and

securely migrates and/or replicate your databases and data
warehouses to AWS
Database migration process
Step 1: Convert or copy your schema

AWS SCT

Source DB or DW Destination DB or DW
Native tool

Step 2: Move your data

AWS DMS

Source DB or DW Destination DB or DW

AWS SCT
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
 AWS Schema Conversion Tool
The AWS Schema Conversion Tool helps automate many
database schema and code conversion tasks when
migrating from source to target database engines
Features
Create assessment reports for homogeneous/heterogeneous migrations

Convert database schema

Convert data warehouse schema

Convert embedded application code

Code browser that highlights places where

manual edits are required

Secure connections to your databases with SSL

Source DB AWS SCT Target DB

Service substitutions/ETL modernization to AWS Glue

Migrate data to data warehouses using AWS SCT data extractors

Optimize schemas in Amazon Redshift

AWS SCT product highlights

Assess Plan Convert schema

and code

Optimize Migrate
data warehouses
Database migration assessment

Connect AWS SCT to

source and target
databases
Run assessment report
Read executive summary
Follow detailed
instructions
Supported source and targets
* Supported via AWS SCT data extractors

Relational NoSQL Data lake Data warehouse

Sources
Amazon Aurora Amazon S3 AWS Snowball

Amazon Aurora Amazon DynamoDB Amazon Elasticsearch Amazon Kinesis

Targets Service (Amazon ES) Data Streams
Amazon Redshift

Amazon DocumentDB (with Amazon Simple Storage

MongoDB compatibility) Service (Amazon S3)
AWS DMS product highlights

Secure Assess Validate AWS Snowball

integration

Monitor Stream data Low cost Multiple

options
The migration process
AWS Cloud

AWS
DMS
On-premises
data center VPN

Internet

• Start a replication instance • Let AWS DMS load data, and

• Connect to source and target keep them in sync
databases • Switch applications over to
• Select tables, schemas, or Application users the target once in sync at
databases your convenience
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Where we started…
Amazon’s Oracle footprint in 2017

• 7,406 Oracle database instances

• 20,000 CPU cores

• 100s of PBs of data

• >15% annual data growth since 2014

• 10s of Amazon data centers globally

• Thousands of sq. ft. of real estate

• Millions of $USD in operation costs

• Many careers based on Oracle

Where we started…
Challenges faced with Oracle Databases

• Complex and costly to scale for growing service

throughput
1

• Expensive and punitive Oracle licenses

• Complicated, expensive, and error-prone database
administration
• 100s of hours/month in administrative tasks
• Complicated and inefficient hardware provisioning
• 400 hours/year forecasting and planning database
hardware capacity
Overall benefits for Amazon.com, Inc.
Simplified cost allocation
New career paths for Oracle database engineers

40% Reduction in latency, with 2x the number of transactions

70%
Reduction in infrastructure costs
Eliminated capex infrastructure costs
Transformed organization with cost model where each team now manages its
own compute and storage infrastructure based on predicted usage and growth

70% Reduction in database administration overhead

Now focus on tasks like query optimization and performance monitoring
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon database freedom journey
Project Rollingstone
1. Eliminate Amazon’s dependency on relational databases for critical services Amazon
DynamoDB
2. Eliminate use of Oracle

Other considerations:
• Stringent timeline
• Ensure no performance degradation of service
• Ensure BAU tasks are not impacted
Amazon
Aurora
Amazon database freedom journey
1. Current application and database assessment

2. For relational migrations, MySQL vs. PostgreSQL:

a) Application/database use case fit Amazon
DynamoDB
b) Migration complexity

3. Aspects are not considered as part of this evaluation:

a) Operational or nonfunctional considerations
b) Performance
c) Native integration with AWS services

4. Majorly migration happened across three major databases

Amazon
a) Amazon DynamoDB – Nonrelational Aurora
b) MySQL (Amazon RDS/MySQL)
c) PostgreSQL (Amazon RDS / Amazon Aurora)
Decision Matrix : MySQL vs. PostgreSQL
Decision evaluation approach:

1. Determine team’s appropriate expertise in which database

2. Use AWS Schema Conversion Tool for assessment of the differences

from Oracle to MySQL/PostgreSQL

3. In case the above steps don’t help, deep dive into database features
• ACID compliance with multi-version concurrency control
• OLTP use-case
• OLAP use-case
• Support for advanced data types such as JSON and XML
• Partitioning
• Built in scheduler
Migration architecture

2 Oracle RDS
AWS DMS

Legacy Oracle
DB RDS
AWS DMS
PostgreSQL

1 Legacy to PostgreSQL
App1 App2
2 Amazon RDS
PostgreSQL to Amazon
RDS Oracle
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Top AWS DMS challenges/learnings

ORA-01555 (Snapshot too old) error with large table migrations

Performance issues with Large Objects (LOBs)
Migration task performance issues with high TPS source systems
Datatype conversion issues
4 Byte character set issue
Transaction isolation
ORA-01555 (Snapshot too old)

• Oracle’s default read consistency mechanism (using undo segments) which ensures
that all rows from a select are intact from the time it was executed
• AWS DMS runs a “select” on the source for a table when the full load phase starts

ORA-01555: snapshot too old: rollback segment number %n with name "%segname"
too small

What to do in Oracle?
1. Reschedule long-running queries when the system has less DML load
2. Increasing the size of your rollback segment (undo) size

Example – One of the internal AWS teams Timber migrated >=30 TB data store and had
more than 2 tables of >=4 TB each
How do we handle ORA-01555 on the source for
large table migrations in AWS DMS?
Use parallel-load AWS DMS Read from a fully synced active
feature: data guard standby or
1. This feature basically chunks snapshot standby without
the table based on partitions impacting source:
or column ranges and loads 1. Sync standby from source
each chunk in parallel 2. Stop the MRP process and
2. Documentation note the SCN at which the
3. Start a single CDC task for sync is stopped
the entire table from a 3. Run full load from standby
custom CDC start time 4. Run CDC from primary/ADG
instance using the SCN
Truncation and performance issues with
Large Object (LOBs)
• MANDATORY: For AWS DMS to replicate LOBs, the table must have a primary key or a
unique index because LOB replication is a two-step process

1. AWS DMS first migrates the entire row without LOBs

2. AWS DMS then using the PK/UK from the first step, queries for LOB data on source
and then updates the target table with LOB data
3. Commit

PS: If you choose FULL LOB MODE in AWS DMS, then even the migration (full load) phase will
be done in the above manner and not in bulk mode
Truncation and performance issues with Large Object
(LOBs)
Error handling Performance

1. For Oracle and PostgreSQL sources, use: 1. Use parallel-load feature in AWS DMS
failTasksOnLobTruncation
2. For Oracle targets with LOBs greater than
2. Have custom Amazon CloudWatch 10 MB, disable directpathfullload in AWS
alarms on AWS DMS task logs to trap DMS
for LOB truncation messages
3. Use limited lob mode, if the LOB size is
3. Use validation within AWS DMS deterministic from the database and
application side

4. Use inline lob mode, if the LOB sizes are

not deterministic

5. Use per table lob mode if many tables

needed to be part of the same task
Important things to note while migrating LOBs
1. Need to plan migrations for tables that have no PKs and contain LOBs. Here is a query to identify
those tables:

• SELECT owner,table_name FROM dba_tables where owner=‘schema_name' and

table_name NOT IN (SELECT table_name FROM dba_constraints WHERE constraint_type
='P' and owner='schema_name ') and table_name in (select DISTINCT table_name
from dba_tab_cols where data_Type IN ('CLOB', 'LOB', 'BLOB') and owner
='schema_name ');

2. Find the max LOB size using Oracle system tables:

select 'select (max(length(' || COLUMN_NAME || '))/(1024)) as "Size in KB" from '

|| owner || '.' || TABLE_NAME ||';' "maxlobsizeqry" from dba_tab_cols where
owner=‘schema_name' and data_type in ('CLOB','BLOB','LOB');

3. Attention to stream buffer settings

Migration task performance issues with high TPS
source systems
Full load process for Oracle to PostgreSQL/MySQL Migration

1. Run a “Select *” on the source

2. Unload data into the replication instance memory and prepare to create comma-delimited CSV files
3. Use copy/load data infile command in PostgreSQL to copy data from the replication instance to the target instance

Change data capture process

Outgoing stream
Incoming stream

Sorter
Source Target
Transactional apply vs. Batch apply
Transactional apply Batch apply

• AWS DMS does this by default • Read changes from the source engine and build
a batch
• Based on timeout
• All transactions are applied in the same order
as it happened on the source • Based on batch size

• Designed to work with foreign key • Create an intermediate net changes table in the
constraints and triggers being active on the target with all changes from a batch
target after applying cached changes
• Apply a net changes algorithm to condense
changes and apply separate DMLs in one
transaction

• Foreign keys and triggers need to be disabled

EXAMPLE – Clicks team used batch apply for the higher TPS workload described earlier
Data type conversion issues
Two-step process:

1. Convert source data type to AWS DMS type data type

2. Convert AWS DMS type data type to target data type

Scenario – Multiple internal customers:

• Number in Oracle becomes a double in PostgreSQL despite it having 0 precision because of the way Oracle
stores it. E.g., an ID column with entry 2045 becomes 2045.00 on the target
• Number to Boolean migration

Recommended workarounds:

1. For small tables, use a trigger on the target to convert this into an INT or BIGINT
2. For larger tables, stop full load, change column definition, resume task into CDC to get rid of extra
unwanted precision

Solution:

AWS DMS data type transformations

4-byte UTF8 character set issues

• AWS DMS used to not support migrating 4-byte UTF8 characters

➢ This limitation was overcome with the launch of AWS DMS version 2.x

• During migration/replication phase, AWS DMS will encode supplementary characters to UTF-16 which causes issues
as encoding strings in surrogate range are invalid UTF8 chars and fails when written to the target

Workarounds:

• Run CSSCAN on Oracle or use the following query to find the problematic characters:

SELECT <Primary_Key_Column>||' '||<Problematic Column> FROM Table name WHERE REGEXP_LIKE(<Problematic

Column>, UNISTR('[\D800-\DFFF]'));

• Replace invalid characters on the source manually and run the migration

• Identify invalid characters and replace them during the migration through an extra connection attribute
Transaction isolation
• Isolation level in Oracle is read-committed for every transaction
➢ This was something that was not an issue with regards to migration to PostgreSQL, but an issue with
MySQL

• MySQL InnoDB engine default isolation level set to repeatable read. This change also requires autocommit to be
disabled

When using AWS DMS, in the initial migration (full load) phase with parallelLoadThreads parameter, rollbacks were
happening, causing data mismatch between source and target

Workarounds:
• AWS DMS allows configuration of session-level parameters using extra connection attribute:

initstmt=SET AUTOCOMMIT=1

Solution
• This was permanently fixed with AWS DMS version 3.x
Before you begin
• Identify key “technical blockers within your organization” and
unblock them

• Database skills are important. Build a core team involving DBAs

and SDEs.

• Build a community to share knowledge across the organization

• Establish a process to provide feedback to AWS

“Difficulties are just things to
overcome, after all. When things
are easy, I hate it.”

Sir Ernest Shackleton

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Learn databases with AWS Training and Certification
Resources created by the experts at AWS to help you build and validate database skills

25+ free digital training courses cover topics and services

related to databases, including:
• Amazon Aurora • Amazon ElastiCache
• Amazon Neptune • Amazon Redshift
• Amazon DocumentDB • Amazon RDS
• Amazon DynamoDB

Validate expertise with the new AWS Certified Database - Specialty beta
exam

Visit aws.training

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Thank you!

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.