Andrey Vasnetsov

Hey @JosuaKrause, this is reasonable request, but it is very hard to implement in distributed setup. You would basically need to do a distributed join, which proved to have very...

Hey @TheQuantumFractal thanks a lot for the contribution! We will take it from here and finish the integration as a separate PR.

`on_disk_payload` is enabled by default. Apart from payload, there are other components of the storage, which takes RAM: vectors, indexes. Here is a bit more info on the subject: https://qdrant.tech/articles/memory-consumption/

@harshakhmk how do you measure memory consumption? If you are using htop or similar tool, it is likely that you are using memory + disk cache

Hi @vishalnayak108, it looks like your memory is consumed by python process, not qdrant.

Hey @marevol, in the context of ANN search, measuring latency without accuracy might be misleading. Could you please confirm that those measurements you provide also yield same level of accuracy?

Hey @hugh2slowmo, you are right, the current implementation of multiple vectors per point reserves space for all vectors even if some of them are not created. It is possible that...

> Given the significant impact on latency that we've observed, were there any major changes made to improve accuracy starting from version 1.6, which might explain this increase in latency?...

from the disk perspective, it doesn't matter if you have one collection or many, but not creating empty vectors will definitely help

This might give more context: https://qdrant.tech/articles/memory-consumption/ Htop is, indeed, not providing valuable info. We figured that the best metric to monitor is `RssAnon` memory usage. You can get it by...