Menu

1 / 2

RAG Re-Indexing Without Downtime: A Dual-Write Pattern for Embeddings

DEV Community·Gabriel Anhaia·7 months ago
#BtviwgCH
#rag#ai#doc_id#await#embedding
Reading 0:00
15s threshold

Book: RAG Pocket Guide: Retrieval, Chunking, and Reranking Patterns for Production

Also by me: Thinking in Go (2-book series) — Complete Guide to Go Programming + Hexagonal Architecture in Go

My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools

Me: xgabriel.com | GitHub

You've shipped a RAG service. It works. Then imagine OpenAI

ships a hypothetical text-embedding-next , your eval rig says

it picks up 4 points of recall on the corpus you actually care

about, and your VP wants the upgrade live by Friday.

The naive plan is: stop ingest, drop the old vectors, re-embed

the corpus, write the new vectors back, resume traffic. On a

small corpus this takes minutes. On 8 million chunks it takes

a weekend, and your support bot returns nothing useful while

the swap is mid-flight.

The right shape is a dual-write re-index . Writes land in

both the old and the new index from the moment you start.

Reads keep going to the old one.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.