Major improvements to Metric Monitor

We've rebuilt the Metric Monitor to address many key points of customer feedback! Improvements includes:

• “Seasonal” pattern detection, with special attention to day-over-day and week-over-week patterns.
• Tighter thresholds that ‘hug’ the trend line
• ML thresholds available on all metrics
• Rapidly trained thresholds by backfilling up to 1 year of historical metrics (learn more). Thresholds often appear after the first run of the monitor + backfill is complete
• Requirement for minimum row count per bucket is removed

Note – to support these changes, the backend and data model needed significant refactoring. So much so that this a hard “v2” of the Metric Monitor. Existing Metric Monitors will continue to function without interruption, but only newly created monitors will get these improvements.

We encourage users to try it out! There are some slight differences in the configuration of a v2 Metric Monitor, compared to before. For example:

• Users are now asked how many days or hours of recent data should be ignored, since it might not be mature yet for anomaly detection.
• To support ingest of much more historical data, segmented metric monitors are now limited to a single alert condition, which can contain only 1 metric and 1 field.

Refer to our documentation on the Metric Monitor and available metrics for more details.