Originally from the User Slack
@Evaldas_Buinauskas: Hello! I would like to use scylla as an intermediate store in our data engineering pipeline where we need to keep search index up to date, very similar project to https://doordash.engineering/2021/07/14/open-source-search-indexing/
One thing that we need to store is clip image embeddings which are represented as a collection of floats. Should I store them as a list<float>
or as a blob
type? I’ll never need to mutate the list and would like to save on performance and reduce storage cost. I am also fine handling the float ↔ byte conversion at the code level.
@Botond_Dénes: If you don’t need to ever mutate this list after writing it, I recommend either frozen<list<float>>
or blob
. Frozen collections are much more lightweight at the storage level, they are essentially treated as a blob, with some added convenience for users.
@Evaldas_Buinauskas: Would I still be able to set it to null when I need to delete it?
@Botond_Dénes: Yes.
@Evaldas_Buinauskas: cool, let me try that, thanks!