PGCon2017 - 0626

PGCon 2017
The PostgreSQL Conference

Speakers
Kuntal Ghosh
Schedule
Day Talks - Day 2 - 2017-05-26
Room DMS 1120
Start time 14:00
Duration 00:45
Info
ID 1052
Event type Lecture
Track Applications
Language used for presentation English
Feedback

Towards Comprehensive Testing Tools

Redefining testing mechanisms!

The objective of this talk is to bring into notice the utility of two tools —Picasso and CODD, to aid the ongoing feature development and performance analysis in PG community. Particularly, this talk will discuss upon the ‘hows’ of these two tools with PostgreSQL in the process of designing, testing and benchmarking of newly developed database operators.

Considering the extensive feature development in PG community, we require comprehensive testing tools, that can find the best and worst cases for the developed feature along with its scalability. Specifically, effective design, testing and benchmarking of a new database operator depend on the ability to construct the alternative database scenarios with regard to the database contents. However, it may be infeasible to create and maintain databases of large sizes or generate all the queries that are useful for the new operator.

This talk will introduce two tools —Picasso database query optimizer visualizer and CODD (COnstructing Dataless Databases), developed at Database Systems Lab of Indian Institute of Science [1,2]. Picasso is a software that can be used for query optimizer analysis, debugging, and redesign aid by system developers. The tool automatically generates a variety of diagrams including plan, cost, cardinality etc. that characterise the behaviour of optimizer over a selectivity space. At the time of benchmarking of a newly developed feature, it can be used to find that perfect query where new operator is useful or analyse its' worst case scenarios.

Next, to check the scalability of the feature, CODD can be put to use, wherein a unified visual interface is provided to create a database with the desired metadata characteristics. Thus, this tool can be used to generate plans for desired scale factors without actually creating the database.

The objective of this talk is to bring into notice the utility of these two tools, to aid the ongoing feature development and performance analysis in PG community. Particularly, this talk will discuss upon the ‘hows’ of these two tools with PostgreSQL in the process of designing, testing and benchmarking of different database operators.