Struct SelectExecutor 

pub struct SelectExecutor { /* private fields */ }

SELECT query executor for SSTable-based storage

Implementations

impl SelectExecutor

pub async fn execute(&self, plan: OptimizedQueryPlan) -> Result<QueryResult>

Execute an optimized query plan.

Instrumented as query.select.plan (issue #1035): this span covers the modern SELECT pipeline — SSTable scan, filtering, projection, aggregation, and WRITETIME/TTL metadata extraction — and is the parent under which the read-path spans (issue #1034) nest. On completion it emits [catalog::QUERY_ROWS_SCANNED] (rows the scan step examined) dimensioned by the honest access path, so the rows-scanned vs rows-returned gap is observable. The bounded access-path attribute is recorded on the span; the query text and key values never are.

impl SelectExecutor

pub fn new(schema: Arc<SchemaManager>, storage: Arc<StorageEngine>) -> Self

Create a new SELECT executor with a system (wall-clock) now source.

Uses the default materialized-result byte budget (crate::config::DEFAULT_MAX_RESULT_BYTES); call SelectExecutor::with_max_result_bytes to override it (the query engine wires it from crate::config::QueryConfig::max_result_bytes).

pub fn with_max_result_bytes(self, max_result_bytes: usize) -> Self

Override the byte ceiling on a materialized result set (issue #1582).

Builder-style: the query engine calls this with crate::config::QueryConfig::max_result_bytes so the config knob is load-bearing on the read path.

pub fn with_max_result_rows(self, max_result_rows: usize) -> Self

Override the row-count safety valve on a materialized result set (issue #1582).

Builder-style: the query engine calls this with crate::config::QueryConfig::max_result_rows so the config knob is load-bearing on the read path (not a hardcoded constant).

pub fn with_forced_read_path( self, forced_read_path: Option<ReadPathMode>, ) -> Self

Set the read-path forcing override (issue #1918).

Builder-style: the query engine calls this with crate::config::QueryConfig::forced_read_path so the config knob is load-bearing on the read path. Some(mode) takes precedence over the CQLITE_READ_PATH env var; None leaves routing to the classifier + env.

pub async fn execute_streaming( &self, plan: OptimizedQueryPlan, config: StreamingConfig, ) -> Result<QueryResultIterator>

Execute an optimized query plan with streaming results (Issue #280)

Instead of materializing all rows in memory, this method returns a QueryResultIterator that yields rows incrementally via a bounded channel. This enables memory-efficient processing of large result sets.

Memory Budget

With default StreamingConfig::buffer_size of 1024 rows and ~1KB avg row size:

• Channel buffer: ~1MB in flight
• Background task: minimal overhead
• Total streaming overhead: ~1-2MB (well within 128MB target)

Limitations

Currently supports:

• SSTableScan with predicates (streaming)
• Filter/Limit/Project (applied during scan)

LIMIT (and OFFSET, when present in the plan) is enforced by the streaming producer (execute_streaming_background): it skips OFFSET matches and stops scanning once count rows have been sent, so a LIMIT N query yields exactly N rows without materializing the rest (Issue #581).

For ORDER BY/GROUP BY/DISTINCT, falls back to full execution then streams results.

Trait Implementations

impl Debug for SelectExecutor

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.