Expand description
Fast path for inserting many rows at once. Skips the per-row VM dispatch loop, coerces and validates the input batch up front, and writes directly through the storage commit path so a load of a million rows pays a small constant overhead rather than a million instruction-dispatch costs.
Used by ingestion endpoints, replication, and the bulk-load admin tool. Validation here matches the constraints the per-row INSERT path applies; if the two diverge, bulk-insert can accept rows that single-row INSERT would reject (or vice versa), which silently produces inconsistent state.