PgConf.Russia 2016

A backup is something that no Postgres deployments should go without as it gives the insurance to get back a deployment on its feet should a disaster strike.

In this talk we will discuss why backups are essential in any sane PostgreSQL deployments (this seems obvious) and what are the different options available to define and set up a good backup strategy. On top of that is discussed how the future of backups would need to be handled, particularly regarding differential backups that gain in popularity among users with large deployments.

In addition to providing a general purpose web platform, Heroku has a large, supporting Postgres service. Over the years, we've learned a lot about running Postgres at scale.
In this talk, we'll cover:

• why Postgres is attractive to run as a cloud service
• how to provision, manage, and monitor a Postgres fleet
• tradeoffs needed to make Postgres work in this environment
• automating failure recovery
• and more

Currently, PostgreSQL uses the interpreter to execute SQL-queries. This yields an overhead caused by indirect calls to handler functions and runtime checks, which could be avoided if the query were compiled into the native code "on-the-fly" (i.e. JIT-compiled): at a run time the specific table structure is known as well as data types used in the query. This is especially important for complex queries, which performance is CPU-bound. At the moment there are two major projects that implement JIT-compilation in PostgreSQL: a commercial database Vitesse DB and an open-source project PGStorm. The former uses LLVM JIT to achieve up to 8x speedup on selected TPC-H benchmarks, while the latter JIT-compiles the query using CUDA and executes it on GPU, which allows to speed up execution of specific query types by an order.

Our work is dedicated to adding support for SQL query JIT-compilation to PostgreSQL using LLVM compiler infrastructure. In the presentation we'll discuss how JIT-compilation can be used to speed up various stages of query execution in PostgreSQL, and the specifics of translating an SQL query into LLVM bitcode to achieve good performing native code. Also we'll present preliminary results for our JIT-compiler on TPC-H benchmark.

Postgres was initially designed to support access methods extendability. Well known citation about access method in Postgres claims: "It is imperative that a user be able to construct new access methods to provide efficient access to instances of nontraditional base types" Michael Stonebraker, Jeff Anton, Michael Hirohama. Extendability in POSTGRES, IEEE Data Eng. Bull. 10 (2) pp.16-23, 1987

Initially, heap was just one for access methods. So, extendability of access methods would also mean pluggable storage engines in modern terms. For now, only index access methods are defined in pg_am table of system catalog. Those index access methods also have well-defined interface. Therefore in order to meet initial design PostgreSQL need to support two features:

• Pluggable index access methods, i.e. ability to implement new index types by adding new tuples to pg_am;
• Pluggable storage engines, i.e. ability to implement completely different storages for tables without traditional heap.

Besides mechanical work like "CREATE ACCESS METHOD" command, extensible index access methods needs to be WAL-logged. For now, community doesn't want extensions to define their own WAL-records, because there is a chance to break both recovery and replication, which is not acceptable. Another approach is to define generic WAL-records, that specify a difference between pages in generalized way.

There are only few DBMS which support pluggable storage engines now. MySQL is the most common example here. However, dealing with different storage engines in MySQL is like dealing with different DBMS. This is not the way PostgreSQL should go from our view.

However, now PostgreSQL users realize benefits from other storages. Ideas of columnar storages and in-memory storages for PostgreSQL are very popular. Simultaneously, technical possibilities to implement them are growing. FDW and custom nodes are arrived. Generic WAL and extensible index access methods are pending for 9.6. Much work in the direction of pluggable storage engines is already done even if it had different aims.

It's time for PostgreSQL core developers to think about native support of pluggable storages without kludges. Finally, we should get "CREATE STORAGE ENGINE name ..." command as legal extendability mechanism.

In this talk we will show current state on pluggable index access method and design of pluggable storage engines.