Building a Production-Ready Postgres Kubernetes Operator in Java with Quarkus and GraalVM

Jan 10, 2026

Alvaro Hernandez, a software engineer and creator of StackGres, dives into the fascinating world of Postgres on Kubernetes. He discusses how LLMs can generate efficient Java code and why Java outshines Python and TypeScript in this context. Alvaro introduces StackGres as a curated stack for PostgreSQL, highlights its powerful automation capabilities, and explores its supportive interfaces. The conversation also touches on the benefits of Quarkus and GraalVM for container efficiency, the challenges of infrastructure automation, and the evolution of Java's role in modern development.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ANECDOTE

Workshop Demo: LLMs Mimic Developer Style

In a workshop Alvaro surprised attendees by showing LLMs generating Java following BCE patterns with minimal prompts.
Attendees were amazed because the LLM produced code that matched the developer's style and structure.

INSIGHT

Sharding Automated Into Simple YAML

StackGres automates complex sharding by creating multiple fully-managed clusters per shard and a coordinator, reducing months of manual effort to a few YAML lines.
Automation scales complex topologies like 60-node sharded clusters in minutes.

INSIGHT

Java Is Viable For Kubernetes Operators

StackGres operator is implemented in Java/Quarkus rather than Go, proving Java can be productive for cloud-native control planes.
Operators only need HTTP client/server functionality, which Java handles well with Quarkus and Fabric8.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

An airhacks.fm conversation with Alvaro Hernandez (@ahachete) about:

discussion about LLMs generating Java code with BCE patterns and architectural rules, Java being 20-30% better for LLM code generation than python and typescript, embedding business knowledge in Java source code for LLM context, stackgres as a curated opinionated stack for running Postgres on kubernetes, Postgres requiring external tools for connection pooling and high availability and backup and monitoring, StackGres as a Helm package and Kubernetes operator, comparison with oxide hardware for on-premise cloud environments, experimenting with Incus for system containers and VMS, limitations of Ansible for infrastructure automation and code reuse, Kubernetes as an API-driven architecture abstracting compute and storage, Custom Resource Definitions (CRDs) for declarative Postgres cluster management, StackGres supporting sharding with automated multi-cluster deployment, 13 lines of YAML to create 60-node sharded clusters, three interfaces for StackGres including CRDs and web console and REST API, operator written in Java with quarkus unlike typical Go-based operators, Google study showing Java faster than Go, GraalVM native compilation for 80MB container images versus 400-500MB JVM images, fabric8 Kubernetes client for API communication, reconciliation cycle running every 10 seconds to maintain desired state, pod local controller as Quarkus sidecar for local Postgres operations, dynamic extension installation without rebuilding container images, grpc bi-directional communication between control plane and control nodes, inverse connection pattern where nodes initiate connections to control plane, comparison with Jini and JavaSpaces leasing concepts from Sun Microsystems, quarter million lines of Java code in the operator mostly POJOs predating records, PostgreSQL configuration validation with 300+ parameters, automated tuning applied by default in StackGres, potential for LLM-driven optimization with clone clusters for testing, Framework Computer laptop automation with Ubuntu auto-install and Ansible and Nix, five to ten minute full system reinstall including BIOS updates

Alvaro Hernandez on twitter: @ahachete