Data Quality Monitoring powered by DuckDB

---

Koality is a Python library for data quality monitoring (DQM) using DuckDB. It provides configurable checks that validate data in tables and can persist results to database tables for monitoring and alerting.

We would like to thank Norbert Maager who is the original inventor of Koality.

Warning

This library is a work in progress!

Breaking changes should be expected until a 1.0 release, so version pinning is recommended.

Documentation

For comprehensive documentation, visit the Koality Documentation.

Core Features

• Configurable Checks: Define data quality checks via simple YAML configuration files
• DuckDB-Powered: Fast, in-process analytics with DuckDB's in-memory engine
• External Database Support: Currently supports Google Cloud BigQuery via DuckDB extensions
• Multiple Check Types: Null ratios, regex matching, value sets, duplicates, counts, match rates, outlier detection, and more
• Flexible Filtering: Dynamic filtering system with column/value pairs for targeted checks
• Result Persistence: Store check results in database tables for historical tracking
• CLI Tool: Easy-to-use command-line interface for running checks
• Threshold Validation: Compare check results against configurable lower/upper bounds

Supported Databases

Database	Status
DuckDB (in-memory)	✅ Fully supported
Google Cloud BigQuery	✅ Fully supported

Koality uses DuckDB as its query engine. External databases are accessed through DuckDB extensions (e.g., the BigQuery extension for Google Cloud). External databases may need custom handling in execute_query!

Note on missing tables/datasets

Koality maps provider-specific "not found" errors (e.g., BigQuery Binder "Not found: Dataset ..." or DuckDB "does not exist") to a unified table_exists metric. When a check's query fails because the target table or dataset is missing, Koality records a table_exists failure for the affected table instead of a generic error to make missing-data diagnostics consistent across providers.

Available Checks

Check Type	Description
NullRatioCheck	Share of NULL values in a column
RegexMatchCheck	Share of values matching a regex pattern
ValuesInSetCheck	Share of values matching a predefined set
RollingValuesInSetCheck	Values in set over a rolling time window
DuplicateCheck	Number of duplicate values in a column
CountCheck	Row count or distinct value count
AverageCheck	Average of a column
MaxCheck	Maximum of a column
MinCheck	Minimum of a column
MatchRateCheck	Match rate between two tables after joining
RelCountChangeCheck	Relative count change vs. historical average
IqrOutlierCheck	Detect outliers using interquartile range
OccurrenceCheck	Check value occurrence frequency

Installation

pip install koality

Or add to your pyproject.toml:

[project]
dependencies = [
    "koality>=0.1.0",
]

Quick Start

1. Create a configuration file

# koality_config.yaml
name: My Data Quality Checks

# Database connection setup - executed before running checks
database_setup: |
  INSTALL bigquery;
  LOAD bigquery;
  ATTACH 'project=${PROJECT_ID}' AS bq (TYPE bigquery, READ_ONLY);

# Prefix for table references (use attached database name)
database_accessor: bq

defaults:
  result_table: bq.dqm.results
  log_path: dqm_failures.txt
  filters:
    partition_date:
      column: date
      value: yesterday
      type: date

check_bundles:
  - name: null_ratio_checks
    defaults:
      check_type: NullRatioCheck
      table: bq.dataset.orders
      lower_threshold: 0
      upper_threshold: 0.05
    checks:
      - check_column: customer_id
      - check_column: order_date
      - check_column: total_amount

For in-memory DuckDB (local testing or CSV/Parquet files):

database_setup: |
  CREATE TABLE orders AS SELECT * FROM 'data/orders.parquet';
  CREATE TABLE results (check_name VARCHAR, result DOUBLE, timestamp TIMESTAMP);
database_accessor: ""

2. Run checks via CLI

# Pass database setup variables via CLI
koality run --config_path koality_config.yaml -dsv PROJECT_ID=my-gcp-project

# Or via environment variable
DATABASE_SETUP_VARIABLES="PROJECT_ID=my-gcp-project" koality run --config_path koality_config.yaml

3. Review results

Results are persisted to your configured result table and failures are logged to the specified log path.

Configuration Hierarchy

Koality uses a hierarchical configuration system where more specific settings override general ones:

1. defaults: Base settings for all checks (result table, persistence, filters)
2. check_bundles.defaults: Bundle-level defaults (check type, table, thresholds)
3. checks: Individual check configurations (specific columns, custom thresholds)

Filter System

Apply dynamic filters to check specific data subsets using the structured filters syntax:

defaults:
  filters:
    partition_date:
      column: created_at
      value: yesterday
      type: date           # Required for rolling checks; auto-parses date values
    shop_id:
      column: shop_id
      value: SHOP01
      type: identifier     # Marks this as the identifier filter for result grouping
    revenue:
      column: total_revenue
      value: 1000
      operator: ">="       # Supports =, !=, >, >=, <, <=, IN, NOT IN, LIKE, NOT LIKE

Identifier filters and naming

Koality supports an identifier filter type which can be used to mark the field that identifies data partitions (e.g., shop, tenant). Use the global identifier_format setting in defaults to control how the identifier appears in result rows:

• identifier (default): result column IDENTIFIER contains column=value (e.g., shop_code=EC0601).
• filter_name: result column uses the filter name (e.g., SHOP_ID) and contains the value only.
• column_name: result column uses the database column name (e.g., SHOP_CODE) and contains the value only.

If an identifier-type filter is defined without a concrete column or value (for example in global defaults), it is treated as a naming-only hint and will not be turned into a WHERE clause; this is useful when you only want to control the result identifier column name (e.g., SHOP_ID) across checks.

Behavior for missing identifier values

When an identifier-type filter is present but its value is missing or explicitly null, Koality substitutes a configurable placeholder for logging and naming (defaults.identifier_placeholder, default: ALL) to avoid None appearing in metric messages. You can override the placeholder at bundle or check level by setting identifier_placeholder in the corresponding defaults.

Additional docs: see docs/identifier_placeholder.md for usage examples and configuration details.

Filter Properties

Property	Description
column	Database column name to filter on (optional in defaults, required after merge)
value	Filter value (optional in defaults, required after merge)
type	date, identifier, or other (default). Only one of each type allowed
operator	SQL operator: =, !=, >, >=, <, <=, IN, NOT IN, LIKE
parse_as_date	If true, parse value as date (for non-date-type filters)

Date Parsing

Koality automatically parses date values when type: date is set:

• Relative dates: today, yesterday, tomorrow
• ISO dates: 2024-01-15, 20240115
• With inline offset: yesterday-2 (2 days before yesterday), today+1 (tomorrow)

Contributing

Contributions are welcome! Please feel free to submit issues and pull requests on GitHub.

License

This project is licensed under the MIT License - see the LICENSE.md file for details.