Generics are Hard

Jun 30

Jun 30 Generics are Hard - Trifecta of Stateful Apps

In modern distributed systems, there has been a modern push to make these systems more generic. The constraints we live in today with cloud native architecture is that cloud native architecture requires us to be idempotent and ephemeral at the same time. These two concepts are pitted against each other and for systems that need to maintain state of consensus is difficult to achieve all three. The trifecta of clustering, load balancing, and replication is needed to achieve consistency and resiliency which is application-centric and not generic.

Idempotency

Idempotency is the mathematical and computer science concept that no matter how many times you execute an equation, you get the same results given the same inputs.

Ephemerality

If something is ephemeral, that means that something is short-lived. Containerized workloads are ephemeral because they are designed to gracefully die e.g not run forever.

Clustering/Load Balancing/Replication

Stateful applications above and beyond persistence have three key characteristics. Stateful applications need to have consistent read/writes and this is usually achieved by leader election or consensus which is a pillar of clustering. Because nodes in a cluster have similar goals, they can also be available in case of increased load or failure. Replication is key between the cluster nodes sharing valuable state information. Lastely as an end-user or consumer process, load balancing between the cluster nodes is important which also can be used as a facility for replication.

I was pretty fortunate to be included in Tech Republic and The New Stack discussing orchestrator SDKs and the rise of Kubernetes Operators. Computer Science in my opinion is a little cyclical. We solve for specifics then go back to generics and then specifics, again.