Skip to main content

dfplan_body_is_split_safe

Function dfplan_body_is_split_safe 

Source
pub fn dfplan_body_is_split_safe(body: &str) -> bool
Expand description

True when a dfplan body’s decoded plan may be split by map-task ranges (Phase 54 skew split) without changing results.

Splitting hands each split task a disjoint subset of the skewed upstream’s map outputs, so any operator that must observe the WHOLE partition before emitting (final-mode aggregation, sort, window, limit, distinct) would produce partial results per split. Safe plans are whitelisted structurally: shuffle reads, projections, filters, batch coalescing, and INNER hash joins (each row of the restricted side lands in exactly one split and joins against the other side read in full, so every match pair appears exactly once across splits; outer joins are excluded — unmatched-row padding would be emitted per split).