Open
Description
Is your feature request related to a problem? Please describe.
I'd like to be able to random-access RecordingView
rows. Currently, RecordingView.select
returns a streaming pyarrow.RecordBatchReader
.
Describe the solution you'd like
Exposing a random access reader (e.g. pyarrow.RecordBatchFileWriter
). Might be related to #6498?
Additional context
We recently introduced rrd
support to rbyte
to enable ML model training on .rrd
files. During training, samples may be requested in random order. To enable random access for frame/image data, we currently store the selected columns in memory, which may become prohibitively expensive memory-wise. Ideally, we'd probably use a random access reader on a mmap'd recording or something along those lines.