We added many cool new features to our anomaly detection tests!
fail_on_zero - You can now configure your volume anomaly tests to fail if there are exactly 0 new records, in addition to the anomaly detection algorithm. Read more here
detection_delay - If your data loads with an expected delay, you can now delay the detection period. This will prevent false alerts on freshness and volume. Thanks Francois Forster for raising the issue and making this amazing contribution!
ignore_small_changes - If your data set is very static, and even a small change can cause a false anomaly detection, you can now use percentiles to ignore small changes. Read more about this here
This test helps to make sure your exposures are not broken by any changes in the upstream models and sources! If a column the exposure depends on changes or gets deleted, the test will fail and alert about the change to those who need to know.
Alerts linked to the report! ⚠️🔗📊
If you are hosting your Elementary report on a server or a bucket, just add --report-url link/to/report to your edr monitor command and each alert will have a link to the results of the test in your hosted report
Crazy performance improvements ⌚️
You should see the impact in runtimes of -
dbt tests hook - Reduced the duration of tests failed row sampling and failed rows count
Schema changes test - Code was optimized, significantly improving performance
Anomaly detection tests - Metrics backfill process was optimized, also reducing the time it takes them to run
