© 2014 EDB All rights reserved. 1

High Availability with PPAS Jihoon Kim, EnterpriseDB Korea

2014-11-12

V03 – Aug 2013!

© 2014 EDB All rights reserved. 2

High Availability Overview

© 2014 EDB All rights reserved. 3

고가용성 요구 사항

요구사항 Solution Profile 장애 대처 능력 Any type, anywhere

복구 시간 Seconds to minutes

테스트 빈도 Whenever

데이터 유실 Zero or seconds

설치/설정 난이도 Simple

기능 통합 여부 Pre-integrated

ROI “Show me the money!”

© 2014 EDB All rights reserved. 4

고가용성 솔루션 요구 사항

High Availability

Disaster Recovery

Replication

Backup & Recovery

Clustering

Failover

Active-Active

Continuous Data Protection

Intelligent Volume Management

Minimal Downtime Planned Maintenance

Simple Integrated Scalable

© 2014 EDB All rights reserved. 5

PostgreSQL High Availability •  Failover Manager

•  Active/Passive Clustering

•  Near real-time Replication

•  Multi-Version Concurrency Control (MVCC)

•  Point-in-Time Recovery

•  Log Shipping for Standby (~ Oracle® Data Guard)

•  Online Operations

© 2014 EDB All rights reserved. 6

Streaming Replication

© 2014 EDB All rights reserved. 7

How Streaming Replication works?

User Transactions Queries, Updates, DDL

Com

mit Write

Ahead Logs

WAL Writer Mem

WAL

Buffer Receiver

Write Ahead Logs

Read-only Transaction Reporting

Master Database Standby Database

Continuous Recovery

Sender

© 2014 EDB All rights reserved. 8

Streaming Replication 구성 방법

Master

Slave

Sync or Async

Master

Slave 1

Async

Slave 2

Sync

Master

Slave 1 Slave 3 Slave 2

Slave 4 Slave 5

Type #1: Sync or Async

Type #2: Multiple Standby

Type #3: Cascade & Multiple Standby

© 2014 EDB All rights reserved. 9

Reader Farm : Scale out using Streaming Replication Streaming Replication 활용 예시

Master

Slave #1

Sync

Slave #2

Slave #3

Slave #9

Slave #10

Async

……

……

Load Balancer

Reader Farm

© 2014 EDB All rights reserved. 10

EDB Failover Manager

© 2014 EDB All rights reserved. 11

EDB Failover Manager

EDB

Failover Manager creates fault tolerant database clusters to minimize downtime when a master database fails by keeping data online in high availability configurations

FAILOVER MANAGER

© 2014 EDB All rights reserved. 12

EDB Failover Manager (EFM) •  A key component to

building a High Availability solution for Postgres databases

•  EFM monitors the health of databases participating in a Postgres HA configuration

•  EFM provides automatic or manual failover processes in the event of a failure

•  EFM is used in conjunction with Streaming Replication

•  A lightweight non-single-point-of-failure technology that is easy to install, configure and use

© 2014 EDB All rights reserved. 13

Backup and Recovery Tool

New

© 2014 EDB All rights reserved. 14

•  Simplifies and reduces errors with a system-wide catalog and command line tool that allows online backup and recovery across local and remote PostgreSQL and PPAS Servers.

What is Backup and Recovery Tool (BART)

BART

© 2014 EDB All rights reserved. 15

•  Backup Catalog to manage multiple PG or PPAS servers

•  SSH support for remote operations

•  Local and remote online physical backups •  Auto-compression and MD5 checksum verification •  Continuous WAL Archiving and PITR

•  Support for tablespace restoration on different paths

BART Feature Highlights

© 2014 EDB All rights reserved. 16

xDB Replication

© 2014 EDB All rights reserved. 17

•  GUI installation program •  Synchronize data between databases •  Distributed Publication / Subscription Architecture

•  Multiple Configuration Options: −  Snapshot or continuous synch modes −  Cascaded configurations −  Replicate whole databases or selected tables −  Replicate selected records with row filters −  Replication Scheduler

xDB SMR Feature Highlights

© 2014 EDB All rights reserved. 18

•  Fast and Easy Replication Management −  GUI Management Console −  Command Line interface −  Replication History Viewer −  Replicate between different Postgres versions

•  Heterogeneous Replication −  Oracle® to Postgres −  Microsoft® SQL Server to Postgres

xDB SMR Feature Highlights

© 2014 EDB All rights reserved. 19

•  Master-to-Master Replication for 2 or more nodes •  Trigger-based Asynchronous implementation

−  Delay from seconds to minutes depending on configuration and load

−  Latency with Logical Decoding in v6.0 expected to be very short

•  Automatic Conflict Detection −  Uniqueness −  Update −  Delete

xDB MMR Feature Highlights

© 2014 EDB All rights reserved. 20

•  Each geography updates a local master faster than to a single common master

•  Allows failover of one geography to another geography

xDB MMR: Ideal for Geo-dispersed servers

© 2014 EDB All rights reserved. 21

Summary

© 2014 EDB All rights reserved. 22

재난 대비 & 업무 활용

Simultaneous read & recovery

High ROI

Automated

Performance protection

Use in production & testing

Recovery mode only

Low ROI

Manual intensive

Disaster protection only

Rarely used

Invest in Improving Quality of Service

© 2014 EDB All rights reserved. 23