Struct SortedSearch 

pub struct SortedSearch {
    partition_by_sort_keys: Vec<PhysicalSortExpr>,
    ordered_partition_by_indices: Vec<usize>,
    input_schema: SchemaRef,
}

This object encapsulates the algorithm state for sorted searching when computing partitions.

Fields

partition_by_sort_keys: Vec<PhysicalSortExpr>

Stores partition by columns and their ordering information

ordered_partition_by_indices: Vec<usize>

Input ordering and partition by key ordering need not be the same, so this vector stores the mapping between them. For instance, if the input is ordered by a, b and the window expression contains a PARTITION BY b, a clause, this attribute stores [1, 0].

input_schema: SchemaRef

Implementations

impl SortedSearch

fn calculate_n_out_row( &mut self, window_agg_states: &[PartitionWindowAggStates], partition_buffers: &mut PartitionBatches, ) -> usize

Calculates how many rows we can output.

Trait Implementations

impl PartitionSearcher for SortedSearch

fn calculate_out_columns( &mut self, _input_buffer: &RecordBatch, window_agg_states: &[PartitionWindowAggStates], partition_buffers: &mut PartitionBatches, _window_expr: &[Arc<dyn WindowExpr>], ) -> Result<Option<Vec<ArrayRef>>>

This method constructs new output columns using the result of each window expression.

fn evaluate_partition_batches( &mut self, record_batch: &RecordBatch, _window_expr: &[Arc<dyn WindowExpr>], ) -> Result<Vec<(PartitionKey, RecordBatch)>>

fn mark_partition_end(&self, partition_buffers: &mut PartitionBatches)

Marks the partition as done if we are sure that corresponding partition cannot receive any more values.

fn input_schema(&self) -> &SchemaRef

fn is_mode_linear(&self) -> bool

Determine whether [InputOrderMode] is [InputOrderMode::Linear] or not.

fn prune(&mut self, _n_out: usize)

Prunes the state.

fn update_partition_batch( &mut self, input_buffer: &mut RecordBatch, record_batch: RecordBatch, window_expr: &[Arc<dyn WindowExpr>], partition_buffers: &mut PartitionBatches, ) -> Result<()>

Updates input_buffer and partition_buffers with the new record_batch.