Arpit Bhayani
Description
Blogs on Database Internals, Engineering Explorations, and Career Growth by Arpit Bhayani
Feed Activity
Similar Feeds
Indie Retro News
www.indieretronews.com
The BEST and ONLY site for Indie Games and Retro Gaming News!
Ed Zitron's Where's Your Ed At
www.wheresyoured.at
The Words of Ed Zitron, a PR person and writer.
Latest Posts
Engineering Speed And Abstractions Create Longer Failure Paths, An Unacknowledged Tradeoff
The text argues that engineers are losing their understanding of complex systems. This is attributed to the pressure for faster delivery and the increasing use of higher-order abstractions, which, while...
Demystifying Kafka's Exactly-Once Semantics: Idempotence, Transactions, and Fencing
Kafka's exactly-once semantics ensures messages are delivered and persisted exactly once, even in the face of retries and failures. This deep dive explores the three layers of its implementation: producer...
AI makes pull requests less central in open-source contribution
Artificial intelligence is causing a shift in how contributions are made to open-source projects. The community's primary role will evolve from submitting code to defining project direction and providing valuable...
Cloudflare Implements Cache Transcoding to Optimize Cache Storage with Zstandard
Cloudflare employs Cache Transcoding, compressing text assets like HTML and CSS with Zstandard before caching them. This method significantly reduces storage and costs by accepting a small, repeated decode cost...
Consistent Hashing: Demystifying the Algorithm and Its Simple Implementation
Consistent Hashing is a fundamental algorithm in distributed systems, often misunderstood as a complex service. This article demystifies its core purpose: solving data ownership in a cluster with minimal changes...
Choose a Depth, Not a Duration, To Navigate Rabbit Holes
I think it is a good thing when you find almost everything interesting. Curiosity that keeps pulling you deeper into a topic is not a bug, but a feature.
How Google Manages Billions of Lines of Code in a Single Monorepo
Explore Google's unique monorepo strategy, which houses billions of lines of code in a single repository, and understand the custom tooling and cultural practices that enable this massive scale. This...
Google's Monorepo: Managing Billions of Lines of Code
Google famously operates with a single, massive monorepository, a decision that initially seemed counterintuitive but has proven highly effective for managing its vast codebase. This document explores the engineering principles,...
Bitcask Explained: A Log-Structured, Fast Key-Value Store
Explore Bitcask, an embedded, log-structured key-value store renowned for its speed and minimal disk I/O. This deep dive covers its architecture, data file format, read/write/delete operations, and the crucial merge/compaction...
Design Multi-Context Processing Tools as Capabilities, Not Endpoints
One anti-pattern I keep seeing: mapping MCP tools 1:1 onto a REST API.
Schema-Driven Data Modeling Is Essential for Deterministic Agent Memory
The more I dig into agent memory, the more I am convinced (strong sentiment because of recent bias) that the schema is the real deal here.
Effective PII Masking Requires Layered Data Classification Beyond Regex
A common failure mode we are seeing is systems leaking PII, and the root cause is almost always the same... People treat masking as a text-replacement problem, not a data-classification...
Retrieving Correct Code for Agentic SDLCs Presents Three Major Problems
Every org is building an agentic SDLC in some capacity, but retrieving the right code is super tricky. Here are three big issues you will run into.
Deckrun: A Modern Presentation Utility for Markdown and Self-contained HTML
Deckrun is an open-source presentation utility that allows users to create and deliver presentations using Markdown and self-contained HTML. It features a powerful editor with themes, font customization, and export...
Automated Verification Ensures End-to-End Correctness and Human Accountability
Testing and verification is where we all should and will spend most of our time; things that we dreaded are what we are going to do now :)
Continuous batching achieves high GPU utilization for large language models
A single GPU serving an LLM does not process one request at a time. It processes one step at a time, and a step can belong to any request that...
How S3 Enables Stateless Write-Ahead Logs Despite Latency Trade-offs
Everybody is building their write-ahead log on S3 now. Turbopuffer, Chroma, and WarpStream are all doing versions of the same trick. Here's what they are doing and how...
Robust System Plumbing is Key for Production-Ready AI Agents
AI agents will retry. They will always retry. Given how long-running agentic loops are, network drops, timeouts, machines rotate, and rate limits kicking in are just part of the deal.
Tailor AI Code Generation Tradeoffs Per File Based on Module Risk
If you are letting AI agents write large chunks of your codebase, here is something I would recommend - decide your tradeoff per file, not as a blanket policy.
Robust Plumbing and Harness are Essential for Production-Ready AI Agents
AI agents will retry. They will always retry. Given how long-running agentic loops are, network drops, timeouts, machines rotate, and rate limits kicking in are just part of the deal.