Do I need to update _node_content in qdrant?

[redbmk]

I'm needing to do a bulk update of a lot of data points in my collection. This should be really easy to do quickly with qdrant's set payload API using filters (e.g. update anything that matches a set of doc ids or some specific metadata, to upsert a couple data points).

However, I've noticed that llamaindex includes a _node_content metadata field that duplicates all of the metadata in the point. _node_content is just another metadata field as far as qdrant is concerned.

My question is, in order for llamaindex's qdrant retrievers to work properly, can I still use this endpoint, or do I need to download all the points, hydrate them with llama-index, update the metadata, then re-upload them to qdrant? that would be way less efficient, but i don't want to break anything.

I feel like I should be able to just delete the _node_content payload field - I don't get why it all needs to be duplicated in the first place - but just checking if it's actually used for something. I do see a metadata_dict_to_node function that seems to pull content from _node_content and ignore the rest of the payload, so I could see that being problematic, but I'm not seeing where this function is actually used.

I did find some similar, stale discussions/issues about _node_content but couldn't find anything definitive, including:

• [Question]: node_content is serialized within metadata and hence stored in the database #10967
• [Question]: update _node_content when metadata are updated #17313

[redbmk]

@dosu is this something you can help with?

[redbmk]

Maybe this bot only works on issues and not discussions