Hello, is there anything to keep in mind when updating qdrant? I have a running container with the data is mounted on a file share (which seems to work fine now). Thank you for your awesome support!

Hii @andrey.vasnetsov , I am using v0.11.5 in docker. It was working fine for sometime and suddenly got stopped without giving any error log. When I try to restart the docker it gives "loading collection: " log for around half (8) collections and automatically exited without any error logs. Could you help to figure out what's happening here ?

@Kacper Łukawski Thanks for posting this article. https://lukawskikacper.medium.com/how-to-implement-a-visual-search-at-no-time-5515270d27e3. I've implemented this for one of my projects, but the implementation that I'm looking for is more along the lines of finding an exact match. We have 250k book covers indexed and we'd like to return the closest identical match from an image taken with a mobile device. The problem we are running into is that slight differences in the image captured on the mobile device return drastically different results or the results that are returned don't vary enough in score to determine which one is the correct result. Do you have any suggestions? Would store more then 512 dims help return more accurate results?

Does `mmap` configuration help if we have collections on NFS ?

Hi, thank you for the great work! I'm wondering if you considered a use case for `recommend` method to use non-indexable vectors as positive/negative examples. Either from the same collection or another one. The RecSys use cases for this are user embeddings or items that shouldn't be included in recommendations. This method is very convenient because it prevents round-trip to load vectors from another collection or storage and in combination with atomic collection/alias switching solves synchronization issues when a model is retrained and the vectors are updated.

what would be the reasonable value for this 2 config? when u are thinking of memory usage as a concern with 1million vectors. ` max_segment_size_kb: null memmap_threshold_kb: null `

Hey guys, currently testing out qdrant and first impression is really good! I am trying out how I could optimize the population of a new collection. If I increase the concurrency the databases crashes somehow and data is corrupted. Then I need to clean the docker container and reinstert with a lower pace. What the best strategy to insert a bunch of vectors (e.g 20Mio)?

Even if I take the payload off it still is taking 1.5 seconds and running more tests I'm seeing that ElasticSearch is actually only taking 500ms on average.

i had updated the param using the update collection api , and then i checked the collection info and it's status was red so i restarted but still getting optimization error on restart, how do i solve this `[2022-12-12T08:04:35Z ERROR collection::update_handler] Optimization error: Service internal error: Can't remove segment data at ./storage/collections/product_text_embeddings__10203/segments/460f03b5-24fc-4f77-aa90-a3481e731262.deleted, error: Directory not empty (os error 39) thread 'collection-product_text_embeddings__10203-shard-0-optimizer-677' panicked at 'Optimization error: Service internal error: Can't remove segment data at ./storage/collections/product_text_embeddings__10203/segments/460f03b5-24fc-4f77-aa90-a3481e731262.deleted, error: Directory not empty (os error 39)', lib/collection/src/update_handler.rs:231:37 thread 'collection-product_text_embeddings__10203-shard-0-optimizer-677' panicked at 'called `Option::unwrap()` on a `None` value', lib/segment/src/segment_constructor/segment_builder.rs:81:85` how do i make sure my collection is ok

Hi Is there an algorithm for text similarity in qdrant?