Struct StreamingDownsampler 

pub struct StreamingDownsampler { /* private fields */ }

Boundary-safe downsampling over a sequence of time-ordered batches.

Rows must arrive ascending by timestamp — which is what the query engine’s merge guarantees — so a bucket is complete as soon as a row belonging to a later bucket is seen. Buckets are emitted in order and exactly once.

Example

// A fixed minute. `TimeBucket::parse("1d", Some("Europe/Berlin"))` for a
// calendar bucket — the same type SQL's `time_bucket()` uses.
let bucket = TimeBucket::fixed(Duration::from_secs(60));
let mut ds = StreamingDownsampler::new("value", bucket, AggFn::Avg)?;
let mut out = Vec::new();
for batch in batches() {
    out.extend(ds.push(&batch)?);
}
out.extend(ds.finish()?);

Implementations

impl StreamingDownsampler

pub fn new( value_column: &str, bucket: TimeBucket, function: AggFn, ) -> Result<Self>

Create a downsampler for value_column over bucket-wide buckets.

Errors

Never fails today; the signature is fallible because push is, and callers already handle a Result here.

pub fn push(&mut self, batch: &RecordBatch) -> Result<Vec<RecordBatch>>

Fold one batch in, returning any batches of completed buckets.

Errors

Returns an error if the timestamp or value column is missing or has an unsupported type, or if a row arrives out of timestamp order.

pub fn finish(self) -> Result<Vec<RecordBatch>>

Close the final bucket and return any remaining output.

Errors

Returns an error if the output batch cannot be built.