Jon Bratseth

Maybe we should just add it to the normal parser.

That would be great!

To avoid tx replay on startup, let vespa-stop-services compl,ete before stopping the container. How many documents are in this schema, do you use HSNW indexes (how many), and are the...

Is it reindexing that is the problem, or is it replaying the tx log at startup? How many documents are in this schema, do you use HSNW indexes (how many),...

Are the fields that are HNSW indexed paged?

I really think your problem is that you don't do a clean shutdown, leading to tx log replay on startup. Building 2000 indexes will take time. > Will confirm on...

Refer to https://swtch.com/~rsc/regexp/regexp4.html

Maybe we should support inputting tensors in JSON format somehow?

This has been considered before and does make sense. We need to use some variant of document id's with groups and distribute the same groups to the same nodes across...

Yes, we do use something similar to consistent hashing in Vespa (the CRUSH algorithm), but here we need to distribute each group to a limited set of nodes to avoid...