Resilience and Reliability on AWS

Chapter 6. Postgres

Ok, if we want to play with bricks, the we’re going to have to build some. This section will show how we approach building flexible, default components that can be used again and again. We require predictable performance and reliable operations. We have to be able to rotate, clone, link, cluster, backup, restore—all in minutes.

Postgres is one of components that we work with every day. Postgres is an open source, scalable relational database system. (The name “Postgres” is actually a shortened form of PostgreSQL.) We’ll discuss the tools we use (SimpleDB, Route 53) and show how we run Postgres to provide high availability. All of the source code is available on pgRDS on github.

As long as Amazon RDS does not support Postgres, there will be people struggling to get it running. The official lamenting to be heard online is that EBS sucks. EBS is said to be slow and unreliable in performance, for example.

As described before, EBS is a new storage technology. It is popularly described as something in between RAID and local disk storage. Because it is different, people have had unrealistic expectations of it. They are disappointed if performance fluctuatees, and start to call EBS unreliable. And they call EBS slow if hardware RAID or SSD feels faster.

But, if we focus on the opportunities EBS brings (many small volumes with sophisticated snapshot capabilities) we can solve this problem as well. We’ll show that Postgres can be made to scale really well, combining the strengths ...