Docs (openmetadata-docs/)
- How to set an absolute vs. relative threshold, per test class.
- The sampling semantics —
PERCENTAGE is sample-stable, ABSOLUTE is not; what the denominator is for each test; which tests bypass the sampler entirely (rule-library SQL).
- The dimensional roll-up, including the
Others bucket limitation (groups outside the top N are only checked in aggregate).
- The zero-bound degenerate case for
PERCENTAGE on statistical tests.
- Note that
tableRowInsertedCountToBeBetween has its own rangeType/rangeInterval window, independent of the profiler partition config — so nobody conflates the two.
E2E (playwright/)
- Configure a relative threshold on
columnValuesToBeNotNull, run the test, see it pass where the strict version failed.
- Enum sentences render in the form (no raw
>= visible).
Depends on all implementation sub-tasks.
Design spec: docs/plans/2026-09-11-dq-failure-thresholds.md. Parent epic: #16301.
Docs (
openmetadata-docs/)PERCENTAGEis sample-stable,ABSOLUTEis not; what the denominator is for each test; which tests bypass the sampler entirely (rule-library SQL).Othersbucket limitation (groups outside the top N are only checked in aggregate).PERCENTAGEon statistical tests.tableRowInsertedCountToBeBetweenhas its ownrangeType/rangeIntervalwindow, independent of the profiler partition config — so nobody conflates the two.E2E (
playwright/)columnValuesToBeNotNull, run the test, see it pass where the strict version failed.>=visible).Depends on all implementation sub-tasks.
Design spec:
docs/plans/2026-09-11-dq-failure-thresholds.md. Parent epic: #16301.