Expand description
The sample as a step of the view: drawn into memory a chunk at a time, while the table shows the rows kept so far.
draw reads the rows crate::sampling would, but hands each chunk that is
final when kept to SampleRows as it lands: a seeded run of one Parquet or IPC
file, a batch of the head or of every row, and a batch’s share of a Bernoulli
sample when the total is known. A reservoir (an unknown total) and an equal per
value sample keep evicting rows until the end, so their rows arrive as one chunk
when the read ends. Rows land in arrival order; SampleRows::in_source_order
puts them in the order the source holds them once the draw ends.
Nothing is written to disk: saving a sample is export’s job.
Structs§
- Drawn
- What a draw read, beside the rows it kept.
- Live
- What a draw keeps its chunks in and how it says so.
- Memory
Check - The memory a draw is checked against, before it starts and as it runs.
- Sample
Rows - The chunks a draw has kept, shared between its worker and the view that shows them. The data is held once: the view’s frame is built on these chunks.
Enums§
- Draw
Path - How a random sample of a stream was drawn: what makes the same seed draw the same rows again, so a view keeps it and a redraw takes it again.
- Limit
- Where the memory a draw may take is measured against.
Constants§
- MEMORY_
SETTING - The setting that caps a sample’s memory, as every message about it names it.
Functions§
- available_
memory - The memory available now, as the system (or the cgroup it runs in) says.
- draw
- Draw
samplefromlf, already cut to its scope, intolive’s chunks.
Type Aliases§
- Memory
Probe - Reads the memory available now, in bytes;
Nonewhen the system will not say.