User-defined Metrics

Users can define custom metrics to monitor specific data quality concerns and receive alerts when thresholds are violated. Each monitored metric is validated against its corresponding monitor, which defines thresholds, scope, notification settings, and alerting rules. When a metric violates the defined policy, an alert is automatically generated.

Telmai provides two methods for defining custom metrics:

1. Metric Expressions - Simple aggregations with grouping

2. Raw SQL Queries - Complex queries with full SQL capabilities

---

Metric Expressions

Metric expressions enable you to specify simple aggregations and groupings using a straightforward syntax.

Syntax Rules

• Attribute names must be wrapped in backticks: `attribute_name`

• Maximum of 4 dimensions in group by clauses

• Supports standard aggregation functions

Available Aggregations

in
max
count
avg
sum
distinct
variance
median
stddev

Examples

Simple aggregation:

SUM(`salary`) / COUNT(*)

Aggregation with grouping:

SUM(`sales`) GROUP BY `region`, `country`

Average by multiple dimensions:

AVG(`order_value`) GROUP BY `region`, `customer_type`, `product_category`

---

Raw SQL Queries

For more complex monitoring scenarios, you can write custom SQL queries that return a single metric value along with optional dimensions.

Requirements

• First column: Must return a numeric value (this is the tracked metric)

• Subsequent columns: Used as dimensions for grouping and filtering

• The query runs against the specified data connector

• Not limited to a single table - you can join multiple tables

Supported Data Connectors

Raw SQL queries are available for the following connectors:

• BigQuery

• Amazon Athena

• Databricks

• Trino

• Snowflake

• Amazon Redshift

Syntax Rules

• Table names must be wrapped in backticks: `table_name`

• Use valid SQL syntax for your specific connector

• Ensure the first selected column returns a numeric value

• Additional columns become dimensions for the metric

Examples

Basic query with dimension:

SELECT emp_salary, emp_region 
FROM `employee_table` 
WHERE emp_age > 60

• Tracked metric: emp_salary

• Dimension: emp_region

Query with multiple dimensions:

SELECT AVG(order_value), customer_region, product_category
FROM `orders` 
WHERE order_date >= CURRENT_DATE - 30
GROUP BY customer_region, product_category

• Tracked metric: AVG(order_value)

• Dimensions: customer_region, product_category

Complex query with joins:

SELECT COUNT(*), o.region, c.customer_tier
FROM `orders` o
JOIN `customers` c ON o.customer_id = c.id
WHERE o.status = 'failed'
GROUP BY o.region, c.customer_tier

• Tracked metric: COUNT(*)

• Dimensions: region, customer_tier

---

Best Practices

Choosing Between Expressions and SQL

Use Metric Expressions when:

• You need simple aggregations on a single table

• Your logic fits within 4 dimensions

• You want a quick, straightforward configuration

Use Raw SQL when:

• You need complex joins across multiple tables

• Your logic requires advanced SQL features (CTEs, window functions, etc.)

• You need fine-grained control over the query

• You're working with connector-specific SQL dialects

Performance Considerations

• Keep queries efficient to avoid impacting your data warehouse

• Use appropriate filters to limit data scanned

• Consider query execution time when setting monitoring frequency

• Test queries directly in your data warehouse before adding them as monitors

---

Troubleshooting

Common Issues

Metric expression fails:

• Verify attribute names are wrapped in backticks

• Check that you're not exceeding 4 dimensions in GROUP BY

• Ensure aggregation function is supported

SQL query returns no data:

• Verify table names are correct and wrapped in backticks

• Check that your filter conditions return results

• Confirm you have permissions to query the tables

First column is not numeric:

• Ensure your first SELECT column returns a number

• Use CAST() or CONVERT() if needed to ensure numeric type

• Aggregations like COUNT(), SUM(), AVG() automatically return numeric values

Query timeout:

• Optimize your query to reduce execution time

• Add more restrictive WHERE clauses

• Consider pre-aggregating data in your warehouse