# more spinach

> A technical blog and lab notebook by Dustin Wilson. The blog is a rich archive (dating back to 2021) which covers topics such as databases, distributed systems, and algorithmic tricks—reflecting sustained engagement with systems-level thinking and practical experimentation. ([homepage](https://www.morespinach.xyz))

### About
- Dustin Wilson is a software engineer from New York, New York, currently working on the Mimir team at Grafana Labs. He previously worked on DBaaS (Postgres, MySQL, Redis, and Kafka) at DigitalOcean and before that was "applying microeconomic theory to litigation and regulatory matters" at NERA. ([about page](https://www.morespinach.xyz/abt.html))  
- Notable technical explorations include:
  - **Reflections on A Month of Claude Code** — Analysis of AI tool usage in software engineering work
  - **Modeling Distributed Systems with Generating Functions** — A library using generating functions to model distribution execution times without simulation
  - **Approximate Voronoi Regions** — An algorithm for generating approximate voronoi regions with low memory footprint
  - **Writing a Concurrency Control Manager** — Go implementation of Optimistic Concurrency Control with custom transaction validation
  - **A Method for Encoding Permutations** — Exploring storage of permutations in fewer than n·log(n) bits
  - **A Reservoir-Based Buffering Algorithm** — Fixed-memory buffering using reservoir sampling for outage resilience
  - **1-Bit Quantization For Faster Search in Vector DBs** — Single-bit quantization method for very fast, low-memory vector database queries

## Highlights

### AI and Software Engineering

**Reflections on A Month of Claude Code** — Personal experience and insights on using AI tools in software engineering work, with plans to update regularly as the relationship with these tools evolves ([agents.html](https://www.morespinach.xyz/agents.html)).

### Distributed Systems & Algorithms

**Modeling Distributed Systems with Generating Functions** — Instead of relying on simulation, developed a library that uses generating functions to model the distribution of execution times in distributed systems ([systems.html](https://www.morespinach.xyz/systems.html)).

**Approximate Voronoi Regions** — A novel algorithm for generating approximate voronoi regions with very low memory footprint that returns the true region with high probability, though it can be quite slow ([approximate_voronoi_algorithm.html](https://www.morespinach.xyz/approximate_voronoi_algorithm.html)).

**Writing a Concurrency Control Manager** — Implementation of Optimistic Methods for Concurrency Control with a custom transaction validation method, resulting in a Go package that wraps arbitrary data stores with OCC semantics ([occ.html](https://www.morespinach.xyz/occ.html)).

### Data Structures & Encoding

**A Method for Encoding Permutations** — Exploration of storing permutations in fewer than n·log(n) bits ([encoding_permutations.html](https://www.morespinach.xyz/encoding_permutations.html)).

**A Reservoir-Based Buffering Algorithm** — A buffering algorithm using fixed memory that keeps a random, uniform sample of messages during outages of arbitrary length ([cascading-buffer.html](https://www.morespinach.xyz/cascading-buffer.html)).

### Vector Search & Quantization

**1-Bit Quantization For Faster Search in Vector DBs** — Single-bit quantization method developed for querying embeddings in vector databases very quickly with low memory usage ([1bit.html](https://www.morespinach.xyz/1bit.html)).

### Open Source Contributions

Dustin has contributed to [Grafana Mimir](https://github.com/grafana/mimir), an open-source, horizontally scalable, highly available, multi-tenant time series database for Prometheus metrics.  
Selected pull requests include:
- [#11967](https://github.com/grafana/mimir/pull/11967): Improves distributor efficiency by preventing unnecessary decompression of incoming push requests when memory limits are nearly reached.
- [#11303](https://github.com/grafana/mimir/pull/11303): Fixes a bug in Mimir caused when multiple pods share ownership of a tenant. Dustin models scenarios where pod ownership changes could cause metric inconsistencies, identifying edge cases involving timing of pod termination and metric writes.
- [#10947](https://github.com/grafana/mimir/pull/10947): Improves compaction performance in Mimir by enabling concurrent uploads of block segment files and the index during compaction.