Struct TraceQueries 

pub struct TraceQueries { /* private fields */ }

High-performance query patterns for Delta Lake trace storage

Design principles:

1. Time-based filters FIRST (enables partition pruning)
2. Binary ID comparisons (no hex decoding overhead)
3. Dictionary-encoded column filters (service_name, span_kind)
4. Minimize data scanned via projection pushdown
5. Leverage pre-computed fields (depth, span_order, path, root_span_id)

Implementations

impl TraceQueries

pub fn new(ctx: Arc<SessionContext>) -> Self

pub async fn get_trace_spans( &self, trace_id_bytes: Option<&[u8]>, service_name: Option<&str>, start_time: Option<&DateTime<Utc>>, end_time: Option<&DateTime<Utc>>, limit: Option<usize>, ) -> Result<Vec<TraceSpanBatch>, TraceEngineError>

Get all spans for a trace_id (direct equivalent of Postgres scouter.get_trace_spans)

This is MUCH simpler than the Postgres version because:

• No recursive CTE needed (depth/path/root_span_id pre-computed)
• Already sorted by span_order
• Direct binary comparison on trace_id

Arguments

• trace_id_bytes - 16-byte trace ID (no hex encoding needed)
• service_name - Optional service name filter
• start_time_hint - Optional time range hint for partition pruning

Performance

• Without time hint: Scans all Delta Lake files (still fast due to min/max stats)
• With time hint: 100-1000x faster via partition pruning

pub async fn get_trace_tree_batches( &self, trace_id_bytes: &[u8], start_time_hint: Option<(&str, &str)>, ) -> Result<Vec<RecordBatch>, TraceEngineError>

Get trace tree structure (same as get_trace_spans but returns raw RecordBatches)

Use this when you need Arrow-native processing without conversion overhead