PgConf.Russia 2019

This talk will represent a new platform of Distributed Control System for Nuclear Power Plant operation. Participants will learn about control system for very complicated automation objects. In a hard real time node more than 150 special subsystems are operating in order to control various technological processes of nuclear power plant (NPP), such as reactor control system for more than 1000 MW power unit with a turbine weighing more than 2000 tons. More than 100K of data gained from sensors are resulting in up to 500K of parameters representing 5 branches of physical processes: neutron kinetics, hydrodynamics, chemistry and radiochemistry, and physics of strength. Deviations may cause the whole system to become a huge DDoS source made of useful diagnostic information which is always much larger than the network and hardware are capable to manage. This may lead to normal operation failure. The talk will reveal the approaches to solve the issue.

You will learn about hardware and software architecture of such systems, about backup and replication, data redundancy and technological diversity. How to manage high loads, what is QoS, and what will happen in case of normal operation system failure, as for example was at Fukushima. But, hey, there should be a talk about coding! So, no SSD and HDDs, only InMemory, data structures from tens of millions of elements, and forget about processor cache as it does not work. Imagine your newest 4-generation Xeon has lost all the advantages and turned into a "pumpkin", so let's roll up your sleeves and examine timings, synchronicity, and try to make the most of your hardware, discovering the weakest link from processor, operating system and a network.

Shared_buffers = 25% – is it too much or not enough? Or it's the right value?

How can we ensure that this – pretty much outdated – recommendation suit well our needs?

It is time to start apply enterprise-level approach to tuning postgresql.conf. Not using various blind auto-tuners or advices from old articles and blog posts, but based on the following two aspects:

1. comprehensive database experiments, conducted in automated fashion, repeated multiple times in conditions as close to production as possible, and
2. deep understanding of DBMS and OS internals.

Using Nancy CLI (https://gitlab.com/postgres.ai/nancy) we will consider a concrete example: infamous shared_buffers, under various circumstances, in various projects. We will try to figure out, how to optimize this settings for given infrastructure, database, and workload.

This case study walks participants through a case where we decided to embark on a data recovery effort. This talk is applicable to all users, from novices to advanced PostgreSQL database administrators. Beginners will get an understanding of what data recovery is and is not, what expectations to have going into it, and how to work with contracted experts in order to ensure the best possible outcome, while more advanced users and experts will also get a fair bit out of the technical aspects of the case study.

While the talk will emphasize non-technical operational aspects of data recovery, it will also include discussions of the internals of PostgreSQL we had to work with, as well as how we went about approaching difficulties so that we could retrieve the data we hoped to.

A major responsibility of a database engine is to convert a declarative SQL query to an efficient execution plan, employing various methods to scan and join the relations. There is always a development effort to improve this area. What clever execution plans can PostgreSQL generate, what's new in version 11 and what is in development? To name a few things, the joins are optimized by removing unneeded outer and inner joins, and reducing joins from outer and semi to inner. There is work to enable merge joins on inequality and range overlap, and to improve join selectivity estimates with multi-column statistics. When it comes to scanning a single relation, covering indexes allow to use index-only scans more often. Incremental sort and more precise estimation of sorting costs help generate better paths when sorted output is required, e.g. when using GROUP BY and ORDER BY or performing merge joins. This talk aims to give an overview of such optimizations that already exist and that are being developed now.