Enforce data quality in a Lakeflow Declarative Pipeline (formerly Delta Live Tables) using expectations

Steps

1. Decorate a Python table/view function with @dp.expect("name", "constraint_sql") to log violations while still writing the rows — this is the default warn-and-keep behavior.
2. Use @dp.expect_or_drop("name", "constraint_sql") to silently drop violating rows before they're written to the target table.
3. Use @dp.expect_or_fail("name", "constraint_sql") to stop pipeline execution immediately on the first violation.
4. For multiple constraints at once, use the grouped forms expect_all / expect_all_or_drop / expect_all_or_fail, passing a dict of {name: constraint}.
5. In SQL pipelines, use CONSTRAINT <name> EXPECT (<expr>) ON VIOLATION {FAIL UPDATE | DROP ROW} inline on CREATE STREAMING TABLE / CREATE MATERIALIZED VIEW — omitting ON VIOLATION defaults to warn-and-keep.
6. Check the pipeline event log's ExpectationMetrics for per-constraint pass/fail row counts rather than inferring quality from before/after row counts.