Qdrant benchmark: post-upload latency spikes are optimizers, tuned configs yield up to 100x faster search

qdrant_engine · x · 2026-09-22

Qdrant's Clelia Bertelli explains why search latency spikes for minutes after bulk-loading millions of vectors: background index/merge/vacuum optimizers compete with queries for CPU, not a bug. Key points: fully optimized collections can search up to 100x faster than unoptimized ones, so don't disable optimizers; continuous indexing (even during upload) pays off long-term, though you can turn off indexing while uploading if tolerated. A benchmark harness used 1.76M Cohere-embedded MS MARCO passages on a single 32GB/14-core node, comparing 13 optimizer configurations during draining and settled phases, with uploads taking 70–316 seconds.

Original post →

More from coding & agent

coding & agent channel →