Drift-Adapter: A Practical Approach to Near Zero-Downtime Embedding Model Upgrades in Vector Databases

Upgrading embedding models in production vector databases typically necessitates reencoding the entire corpus and rebuilding the Approximate Nearest Neighbor (ANN) index, leading to significant operational disruption and computational cost. This paper presents Drift-Adapter, a lightweight, learnable transformation layer designed to bridge embedding spaces between model versions. By mapping new queries into the legacy embedding space, Drift-Adapter enables the continued use of the existing ANN index, effectively deferring full re-computation. We systematically evaluate three adapter parameterizations: Orthogonal Procrustes, Low-Rank Affine, and a compact Residual MLP, trained on a small sample of paired old/new embeddings. Experiments on MTEB text corpora and a CLIP image model upgrade (1M items) show that Drift-Adapter recovers 95-99% of the retrieval recall (Recall@10, MRR) of a full re-embedding, adding less than 10 µs query latency. Compared to operational strategies like full re-indexing or dual-index serving, Drift-Adapter dramatically reduces recompute costs (by over 100×) and facilitates upgrades with near-zero operational interruption. We analyze robustness to varied model drift, training data size, scalability to billion-item systems, and the impact of design choices like diagonal scaling, demonstrating Drift-Adapter's viability as a pragmatic solution for agile model deployment.