Skip to main content
The table stats monitor is an all-in-one monitor to detect if data is not flowing or flowing at a reduced rate. This can help you be the first to know about issues without having to maintain manual thresholds for freshness or volume
  • A source that’s entirely stopped sending data
  • A source that’s sending significantly fewer/more rows
  • A join that causes a spike in duplicate values
A table stats monitor highlighting a row count anomaly that’s been marked as expected title The table stats monitor tracks three metrics:
  • Row count: the number of rows in the table.
  • Delay: the time since data was last loaded.
  • Change delay: the time since the last load that added at least one row.
By combining the three, you’ll be more likely to detect different types of issues. For example, if an analytics engineer makes a join that creates a set of duplicate rows, the row count monitor can catch that. If an upstream source stops writing new data, the change delay monitor can catch that. Unlike manual freshness checks (e.g., dbt source freshness or Coalesce freshness tests), all table stats are self-learning and automatically adjust to the patterns of your data. This reduces the workload on your end on defining specific thresholds. Table stats monitors rely on your data warehouse metadata tables, such as the row_count from the Snowflake table information schema. As the monitor queries the information schema, you only incur minimal additional costs even when they are run frequently. The monitor works out of the box, which makes it easy to deploy at scale.

Setting up a table stats monitor

Table stats monitors are placed with deployment rules. A rule selects a set of tables and keeps a monitor on each of them, including tables that start to match later.
  1. Navigate to Deployment rules under Health, and click Add deployment rule.
  2. Under Settings, pick the metrics (Row count, Delay, Change delay), the sensitivity and the severity. You can set a different sensitivity per metric.
  3. Under Tables, select the tables with the Visual builder or write a ResolverQL query. For example, filter by type to cover all dbt sources, Coalesce sources or SQLMesh external models, or select a data product to cover the tables upstream of it.
  4. Check the tables the rule matches, give the rule a name (for example, important sources), and confirm.
All monitors show on the , where you can adjust their sensitivity. Coalesce Quality already stores the metadata of your tables, so the monitor learns from that history as soon as you set it up. For a new table, or a new workspace without history, expect about 10 days before the monitor knows the patterns in your data well enough to detect anomalies.

Schedule and coverage

A table stats monitor from a deployment rule runs every 30 minutes, with each metadata refresh. You cannot change this schedule. A view has no row count or last change time in the warehouse metadata, so a rule covers a view only if the rule has a freshness field pattern. The pattern is a regular expression that selects the timestamp column to read. It is case-sensitive unless it starts with (?i). If you give more than one pattern, the first pattern that matches a column of the view is used. A view with no matching column is skipped. Each view that the rule covers gets its own query, which runs every hour and adds to your warehouse cost. The freshness field patterns option is in alpha. If you do not see it, ask support to enable it for your workspace. A table that is truncated and reloaded on each run gives table stats that do not show the real data flow. Exclude such tables from the rule, and add a custom monitor with a daily schedule instead.