© 2ndQuadrant 2016 Big Data & PostgreSQL Using TABLESAMPLE to Analyze Very Large Datasets By Umair Shahid
© 2ndQuadrant 2016
Big Data & PostgreSQLUsing TABLESAMPLE to Analyze Very Large Datasets
By Umair Shahid
© 2ndQuadrant 2016
Who am I?● Director, Products @
2ndQuadrant● Got “pushed” into PostgreSQL in
2004, ended up falling in love with it
● Not a hardcore techie, yet passionate about open source software
● Interested in Big Data, especially the newer PostgreSQL features supporting it
2011
2015
© 2ndQuadrant 2016
What is Big Data?● Volume
○ Size: Text files to HD videos○ Sources: Spreadsheets to sensors○ From lakes to oceans
● Velocity○ More sources imply more speed○ Faster connectivity implies more speed○ High-paced world requires faster turnaround
● Variety
© 2ndQuadrant 2016
What is the problem?Number of Rows Size on Disk (MB) Time Taken (ms)
1k 0.23 219.706
100k 24 1,302.135
1M 195 7,696.386
5M 951 40,691.603
10M 1,923 60,012.457
100M 19,456 801,493.319
© 2ndQuadrant 2016
Why is this significant?
● Data mining has typically been a painful process● Major contributor to the pain has been the time it
takes for queries to return● Many false steps before the required data is
identified● Waiting time is wasted time● Sampling, count based or time based, reduces
the wasted time significantly
© 2ndQuadrant 2016
What is TABLESAMPLE?
● Ability to read a random sample of data in a table
● Defined in SQL:2003 (5th revision of SQL)
● Implemented in PostgreSQL 9.5
© 2ndQuadrant 2016
Syntax
SELECT select_expression
FROM table_name
TABLESAMPLE sampling_method ( argument [, ...] )
[ REPEATABLE ( seed ) ]
...
© 2ndQuadrant 2016
sampling_method
● argument is percentage of rows● SYSTEM
○ Block level sampling○ Very fast○ Non-independent rows
● BERNOULLI○ Row level sampling○ Slower than SYSTEM○ Independent rows (uniformly random)
© 2ndQuadrant 2016
© 2ndQuadrant 2016
Demo sampling methods
© 2ndQuadrant 2016
REPEATABLE results● (Reminder: [ REPEATABLE ( seed ) ])● Optional argument● Used if random, yet repeatable results are
required● seed and argument need to be the same to
produce repeatable results● Any changes made to the table will result in a
different data set
© 2ndQuadrant 2016
Now it gets interesting … ● TABLESAMPLE allows for additional sampling methods
via extensions● tsm_system_time specifies max number of
milliseconds to spend reading a table● Implements the syntax:
SELECT select_expression
FROM table_name
TABLESAMPLE SYSTEM_TIME (argument)
© 2ndQuadrant 2016
Demo tsm_system_time
© 2ndQuadrant 2016
Enter Orange ...● Funded by AXLE
(http://axleproject.eu)● Same project funded
TABLESAMPLE● Available integrated
with PostgreSQL in 2UDA (http://2ndquadrant.com/2uda)
● Uses TABLESAMPLE to very quickly create visualizations for data
● Can quickly create predictive models
© 2ndQuadrant 2016
Demo OrangeYou can find a very helpful tutorial at
http://2ndquadrant.com/2uda
© 2ndQuadrant 2016
Other Big Data features in PostgreSQL● HSTORE● XML● JSON & JSONB● BRIN INDEXES● Parallel sequential scan● Parallel aggregates● FDWs● Horizontal Scalability
○ Check out Postgres-XL http://www.postgres-xl.org/
© 2ndQuadrant 2016
Features from the latest release
● 9.6 Beta3 announced last night
● Added support to parallel query for TABLESAMPLE
© 2ndQuadrant 2016
Moving Forward … ● Next meetup: Tentatively August 19, 2016● Please come forward and share your
PostgreSQL stories● Today’s refreshments are sponsored by
2ndQuadrant - THANK YOU!○ Need more sponsors
OR○ Need to start charging for these sessions
Special Thanks!!
© 2ndQuadrant 2016
Umair ShahidEmail: [email protected]: @pg_umair
2ndQuadrant is hiring - All geographies!
Thank you for your time!