Auto-Mode and Partitioned Tables

If you use machine learning (ML)-based monitors on partitioned tables and enable the "auto" setting for the time field, Sifflet uses the partition key by default when it is a timestamp or date field.

In this case, during the first run, Sifflet uses the time partition key to train the model on historical data (the default is 365 days). In the following runs, Sifflet also uses the time partition key in the lookback period (the default is 1 day) to take the most recent data point into account incrementally, which optimizes the query.

📘

In auto mode, you no longer need to opt in to using the partition key to optimize the query.

🚧

This feature currently applies to two types of partitions for BigQuery: INGESTION_TIME and TIME_UNIT_COLUMN.