Ideally. pre-allocating RAM for that buffer, this would either trigger the OOM killer by using too much RAM and locking the system, or otherwise run fine since the buffers are ready to use. This feature has the added benefit of not losing crucial samples when the allocating RAM during a capture. It also lets you know that a capture will complete without running into a memory issue.
I think this makes much sense.
I can hardly understand why there's no (at list option of) usage of HDD storage for sampling. Even now SSDs are quite fast.
So if there's a pre-allocated buffer in RAM (which is fast) application should simultaneously fill in (insert) the buffer with samples from USB (Logic probe) AND load out (delete) from buffer to semi-final (temporary) storage file.
This means that buffer in RAM should act as queue or FIFO.
I'm posting this as feature request.
Without that there's always (or often) a problem of unknown record time length.