“Building Journaling Databases with PostgreSQL” Cybertec Geschwinde & Schoenig Hans-Juergen Schoenig

Common Problems * downtimes can be caused by: - hardware failures - network failures - software bugs - troubles caused by users * history and documentation “I don't remember what we have done “Who has changed this?” “What was the value before the change?”

What about the users? * data can be deleted accidentally by clicking on a wrong button * data is changed to something wrong => These problems cannot be solved by redudant hardware => restoring data can take a lot of time

Downtime vs. Uptime 99.9% uptime = 8 ½ hours downtime/year Assumption: * 100 people are affected by a problem * 1 hour downtime costs $50/person * average time for recovery 2 hours * 1 problem caused by a user per year => costs: $ per year or 200 hours Reducing downtimes by 50% saves $5.000

The Idea Of Journaling * Every version of a record is stored * Every version of a record has: - a database wide unique id - a timestamp - a flag telling if a record is still valid or not - a transaction id to see which records have been changed within the same transaction - a user who has modified the record

Benefits of Journaling You can... * undo transactions * restore a snapshot of the database within minutes and without having to use your backup * monitor the changes made to your data * benchmark your employees * build a history of your data

Journaling And Performance Journaling databases are slower than “normal” databases due to overhead and a higher consumption of storage. In most cases this will NOT cause real performance problems.

A Case Study The Cybertec Intranet: Journaling helps us to: - keep track of the changes - increase the security of our data - see what has happened when - prevent errors 1 / 2

How does it work? 1 / 3 Historical + Current Data Functions And Rules Current Data View SELECT, INSERT, UPDATE, DELETE COPY, Triggers

How does it work? 2 / 3 CREATE TABLE t_mother ( globidint8, transidint4DEFAULT getxid(), transtimetimestamp without time zone DEFAULT now(), realtimetimestamp without time zone, statuschar DEFAULT 't', usernametext DEFAULT current_user() ); Historical + Current Data

How does it work? 3 / 3 Current Data View: - a recent version of the data is provided - the application can perform all common operations - the data is provided by a view Functions And Rules: - data structure is managed by rules - Functions are needed for managing serials, integrity, etc. Functions and Rules + Current Data View

Time Warps 3 / 3 Restoring Old Data: - old snapshots can be restored using a function called recovery - no data is lost during a Time Warp - recovery is based on functions Functions and Rules + Current Data View

Is it difficult? oreilly=# SELECT recovery(' :49:26'); recovery t (1 row)

Any Questions?