All posts published on this site.
Everything published here, newest first.
DataFusion’s multi-level merge merges the sorted runs it can afford, spills the merged result back to disk, and repeats until one pass remains, re-spilling skewed runs whose widest …
DataFusion records cheap atomic metrics per operator partition, measures only the time spent inside poll(), and renders the annotated plan after the query finishes — so EXPLAIN …
DataFusion lists object-store prefixes, prunes partitions by path, fetches file metadata concurrently, and packs the surviving files into FileGroups.
DataFusion’s hash join inserts the build side in reverse order into a LIFO hash table, so the chain head is always the earliest matching row and the join output preserves the build …
DataFusion borrows the ambient Tokio runtime, funnels every spawn through a cancel-safe wrapper, and runs CPU work inline with cooperative yields instead of spawn_blocking.