Articles

I’m joining PlanetScale! I’m excited to tackle ambitious database and distributed-systems challenges while helping make powerful infrastructure approachable to developers.

By implementing document-level parallelism and pipelined bulk operations, I significantly improved ClusterSync’s performance, delivering up to 18× higher replication throughput.

Deploying database infrastructure on Kubernetes is only getting more important, especially if you already have your infra running on some k8s cluster. Learn how to deploy them properly with ease using Kubernetes operators.

See how we make Percona Kubernetes operators more secure with the Trivy security scanner.

Every sharded-data system (a distributed datastore) has (or aims for) three fundamental properties: data consistency, data availability, and that it is tolerant to a network partition (a subset of nodes not being able to reach the rest of the cluster). But due to simple laws of physics it is impossible to have all of these.

I’m pretty sure we all know what technical debt is, attributed to Ward Cunningham in 1992, it basically states that there is no free lunch. But I love it. It makes us ship things quickly, validate them then refine them later. But it comes with a condition…

The foundation of every algorithm in a distributed system is on what kind of underlining hardware system it will execute on. In practice, nodes (physical or virtual machines) and communication links between them (the physical network layer), can behave in a variety of ways. So for an algorithm to be correct, we need to assume what properties the underlining systems will have.

If time is money, then clocks are rather an important part of the whole picture. Let’s see how we handle time ticking in a computer system.

In a multithreaded program, races are rally bad. The reason for this is that a program is compiling correctly, tests are passing and it runs in production for months, and then, all of a sudden, the program returns a result that is everything but the expected one.