PSeitz

https://github.com/quickwit-oss/quickwit/pull/4880 and https://github.com/quickwit-oss/quickwit/pull/4922 fixes issues which caused to emit a leaf request for each split instead for each index on multi index queries. (e.g. 1000 indices * 100 splits =...

Yes the leaf requests are index specific currently, they use e.g. storage and split_offsets. On the request there are currently index specific fields on the root level of the request:...

> I don't think split_offsets is really index specific. `split_offsets` and `search_request` contain the index_id but I assume it is never used by leaves, is it? > `split_offsets` are only...

multi index leaf requests https://github.com/quickwit-oss/quickwit/pull/4962

Yes, I think this should be mostly covered by having cost + seek_exact on the `DocSet` from https://github.com/quickwit-oss/tantivy/issues/2531 What it does not cover is to change the whole execution order,...

You could over-allocate (maximum compression is 250x) an unitialized buffer and decompress into it. As long as the memory is never touched it should be cheap, since pages in the...

I'm not sure this is possible, but I think the Reader passed to the `FrameDecoder` could block when the next block is not yet available yet. The data for a...