The Data Validation Tool (DVT) is an open sourced Python CLI tool based on the Ibis framework that compares heterogeneous data source tables with multi-leveled validation functions.
Data validation is a critical step in a Data Warehouse, Database or Data Lake migration project, where structured or semi-structured data from both the source and the destination tables are compared to ensure they are matched and correct after each migration step (e.g. data and schema migration, SQL script translation, ETL migration, etc.). The Data Validation Tool provides an automated and repeatable solution to perform this task.
DVT supports the following validation types: * Table level * Table row count * Group by row count * Column aggregation * Filters and limits * Column level * Full column data type * Row level hash comparison (BigQuery tables only) * Raw SQL exploration * Run custom queries on different data sources
DVT supports the following connection types:
The Connections page provides details about how to create and list connections for the validation tool.
The Installation page describes the prerequisites and setup steps needed to install and use the data validation tool.
Before using this tool, you will need to create connections to the source and target tables. Once the connections are created, you can run validations on those tables. Validation results can be printed to stdout (default) or outputted to BigQuery. The validation tool also allows you to save or edit validation configurations in a YAML file. This is useful for running common validations or updating the configuration.
The Data Validation Tool expects to recieve a source and target connection for each validation which is run.
These connections can be supplied directly to the configuration, but more often you want to manage connections separately and reference them by name.
Connections can be stored locally or in a GCS directory.
To create connections please review the Connections page.
The data validation CLI is a main interface to use this tool.
The CLI has several different commands which can be used to create and re-run validations.
The validation tool first expects connections to be created before running validations. To create connections please review the Connections page.
Once you have your connections set up, you are ready to run the validations.
Below are the command syntax and options for running validations from the CLI. DVT supports column (including grouped column) and schema validations.
Below is the command syntax for column validations. To run a grouped column
validation, simply specify the --grouped-columns
flag. You can also take
grouped column validations a step further by providing the --primary-key
flag.
With this flag, if a mismatch was found, DVT will dive deeper into the slice
with the error and find the row (primary key value) with the inconsistency.
data-validation (--verbose or -v) validate column
--source-conn or -sc SOURCE_CONN
Source connection details
See: *Data Source Configurations* section for each data source
--target-conn or -tc TARGET_CONN
Target connection details
See: *Connections* section for each data source
--tables-list or -tbls SOURCE_SCHEMA.SOURCE_TABLE=TARGET_SCHEMA.TARGET_TABLE
Comma separated list of tables in the form schema.table=target_schema.target_table
Target schema name and table name are optional.
i.e 'bigquery-public-data.new_york_citibike.citibike_trips'
[--grouped-columns or -gc GROUPED_COLUMNS]
Comma separated list of columns for Group By i.e col_a,col_b
[--primary-keys or -pk PRIMARY_KEYS]
Comma separated list of columns to use as primary keys
(Note) Only use with grouped column validation
[--count COLUMNS] Comma separated list of columns for count or * for all columns
[--sum COLUMNS] Comma separated list of columns for sum or * for all numeric
[--min COLUMNS] Comma separated list of columns for min or * for all numeric
[--max COLUMNS] Comma separated list of columns for max or * for all numeric
[--avg COLUMNS] Comma separated list of columns for avg or * for all numeric
[--bq-result-handler or -bqrh PROJECT_ID.DATASET.TABLE]
BigQuery destination for validation results. Defaults to stdout.
See: *Validation Reports* section
[--service-account or -sa PATH_TO_SA_KEY]
Service account to use for BigQuery result handler output.
[--filters SOURCE_FILTER:TARGET_FILTER]
Colon spearated string values of source and target filters.
If target filter is not provided, the source filter will run on source and target tables.
See: *Filters* section
[--config-file or -c CONFIG_FILE]
YAML Config File Path to be used for storing validations.
[--threshold or -th THRESHOLD]
Float value. Maximum pct_difference allowed for validation to be considered a success. Defaults to 0.0
[--labels or -l KEY1=VALUE1,KEY2=VALUE2]
Comma-separated key value pair labels for the run.
[--format or -fmt] Format for stdout output. Supported formats are (text, csv, json, table).
Defaults to table.
The default aggregation type is a 'COUNT *'. If no aggregation flag (i.e count, sum , min, etc.) is provided, the default aggregation will run.
The Examples page provides many examples of how a tool can used to run powerful validations without writing any queries.
Below is the syntax for schema validations. These can be used to compare column types between source and target.
data-validation (--verbose or -v) validate schema
--source-conn or -sc SOURCE_CONN
Source connection details
See: *Data Source Configurations* section for each data source
--target-conn or -tc TARGET_CONN
Target connection details
See: *Connections* section for each data source
--tables-list or -tbls SOURCE_SCHEMA.SOURCE_TABLE=TARGET_SCHEMA.TARGET_TABLE
Comma separated list of tables in the form schema.table=target_schema.target_table
Target schema name and table name are optional.
i.e 'bigquery-public-data.new_york_citibike.citibike_trips'
[--bq-result-handler or -bqrh PROJECT_ID.DATASET.TABLE]
BigQuery destination for validation results. Defaults to stdout.
See: *Validation Reports* section
[--service-account or -sa PATH_TO_SA_KEY]
Service account to use for BigQuery result handler output.
[--config-file or -c CONFIG_FILE]
YAML Config File Path to be used for storing validations.
[--format or -fmt] Format for stdout output. Supported formats are (text, csv, json, table).
Defaults to table.
There are many occasions where you need to explore a data source while running validations. To avoid the need to open and install a new client, the CLI allows you to run custom queries.
data-validation query
--conn or -c CONN
The connection name to be queried
--query or -q QUERY
The raw query to run against the supplied connection
There may be ocassions we want to release a new CLI feature under a Beta flag. Any features under Beta may or may not make their way to production. However, if there is a Beta feature you wish to use than it can be accessed using the following.
data-validation beta --help
If you wish to use Data Validation as a Flask service, the following command will help. This same logic is also expected to be used for Cloud Run, Cloud Functions, and other deployment services.
data-validation beta deploy
You can customize the configuration for any given validation by providing use case specific CLI arguments or editing the saved YAML configuration file.
For example, the following command creates a YAML file for the validation of the
new_york_citibike
table: data-validation validate column -sc my_bq_conn -tc my_bq_conn -tbls bigquery-public-data.new_york_citibike.citibike_trips -c citibike.yaml
Here is the generated YAML file named citibike.yaml
:
result_handler: {}
source: my_bq_conn
target: my_bq_conn
validations:
- aggregates:
- field_alias: count
source_column: null
target_column: null
type: count
filters: []
labels: []
schema_name: bigquery-public-data.new_york_citibike
table_name: citibike_trips
target_schema_name: bigquery-public-data.new_york_citibike
target_table_name: citibike_trips
threshold: 0.0
type: Column
You can now edit the YAML file if, for example, the new_york_citibike
table is
stored in datasets that have different names in the source and target systems.
Once the file is updated and saved, the following command runs the new
validation:
data-validation run-config -c citibike.yaml
View the complete YAML file for a GroupedColumn validation on the examples page.
Aggregate fields contain the SQL fields that you want to produce an aggregate
for. Currently the functions COUNT()
, AVG()
, SUM()
, MIN()
and MAX()
are supported.
validations:
- aggregates:
- field_alias: count
source_column: null
target_column: null
type: count
- field_alias: count__tripduration
source_column: tripduration
target_column: tripduration
type: count
- field_alias: sum__tripduration
source_column: tripduration
target_column: tripduration
type: sum
- field_alias: bit_xor__hashed_column
source_column: hashed_column
target_column: hashed_column
type: bit_xor
Filters let you apply a WHERE statement to your validation query (ie. SELECT * FROM table WHERE created_at > 30 days ago AND region_id = 71;
). The filter is
written in the syntax of the given source.
Note that you are writing the query to execute, which does not have to match between source and target as long as the results can be expected to align. If the target filter is omitted, the source filter will run on both the source and target tables.
Grouped Columns contain the fields you want your aggregations to be broken out
by, e.g. SELECT last_updated::DATE, COUNT(*) FROM my.table
will produce a
resultset that breaks down the count of rows per calendar date.
Sometimes direct comparisons are not feasible between databases due to differences in how particular data types may be handled. These differences can be resolved by applying functions to columns in the source query itself. Examples might include trimming whitespace from a string, converting strings to a single case to compare case insensitivity, or rounding numeric types to a significant figure.
Once a calculated field is defined, it can be referenced by other calculated fields at any "depth" or higher. Depth controls how many subqueries are executed in the resulting query. For example, with the following yaml config...
- calculated_fields:
- field_alias: rtrim_col_a
source_calculated_columns: ['col_a']
target_calculated_columns: ['col_a']
type: rtrim
depth: 0 # generated off of a native column
- field_alias: ltrim_col_a
source_calculated_columns: ['col_b']
target_calculated_columns: ['col_b']
type: ltrim
depth: 0 # generated off of a native column
- field_alias: concat_col_a_col_b
source_calculated_columns: ['rtrim_col_a', 'ltrim_col_b']
target_calculated_columns: ['rtrim_col_a', 'ltrim_col_b']
type: concat
depth: 1 # calculated one query above
is equivalent to the following SQL query...
SELECT
CONCAT(rtrim_col_a, rtrim_col_b) AS concat_col_a_col_b
FROM (
SELECT
RTRIM(col_a) AS rtrim_col_a
, LTRIM(col_b) AS ltrim_col_b
FROM my.table
) as table_0
Calculated fields can be used by aggregate fields to produce validations on calculated or sanitized raw data, such as calculating the aggregate hash of a table. For example the following yaml config...
validations:
- aggregates:
- field_alias: xor__multi_statement_hash
source_column: multi_statement_hash
target_column: multi_statement_hash
type: bit_xor
calculated_fields:
- field_alias: multi_statement_hash
source_calculated_columns: [multi_statement_concat]
target_calculated_columns: [multi_statement_concat]
type: hash
depth: 2
- field_alias: multi_statement_concat
source_calculated_columns: [calc_length_col_a,
calc_ifnull_col_b,
calc_rstrip_col_c,
calc_upper_col_d]
target_calculated_columns: [calc_length_col_a,
calc_ifnull_col_b,
calc_rstrip_col_c,
calc_upper_col_d]
type: concat
depth: 1
- field_alias: calc_length_col_a
source_calculated_columns: [col_a]
target_calculated_columns: [col_a]
type: length
depth: 0
- field_alias: calc_ifnull_col_b
source_calculated_columns: [col_b]
target_calculated_columns: [col_b]
type: ifnull
depth: 0
- field_alias: calc_rstrip_col_c
source_calculated_columns: [col_c]
target_calculated_columns: [col_c]
type: rstrip
depth: 0
- field_alias: calc_upper_col_d
source_calculated_columns: [col_d]
target_calculated_columns: [col_d]
type: upper
depth: 0
is equivalent to the following SQL query...
SELECT
BIT_XOR(multi_statement_hash) AS xor__multi_statement_hash
FROM (
SELECT
FARM_FINGERPRINT(mult_statement_concat) AS multi_statement_hash
FROM (
SELECT
CONCAT(calc_length_col_a,
calc_ifnull_col_b,
calc_rstrip_col_c,
calc_upper_col_d) AS multi_statement_concat
FROM (
SELECT
CAST(LENGTH(col_a) AS STRING) AS calc_length_col_a
, IFNULL(col_b,
'DEFAULT_REPLACEMENT_STRING') AS calc_ifnull_col_b
, RTRIM(col_c) AS calc_rstrip_col_c
, UPPER(col_d) AS calc_upper_col_d
FROM my.table
) AS table_0
) AS table_1
) AS table_2
The output handlers tell the data validation tool where to store the results of each validation. The tool can write the results of a validation run to Google BigQuery or print to stdout (default).
View the schema of the results here.
data-validation validate column
-sc bq_conn
-tc bq_conn
-tbls bigquery-public-data.new_york_citibike.citibike_trips
-bqrh project_id.dataset.table
-sa [email protected]
Creating the list of matched tables can be a hassle. We have added a feature which may help you to match all of the tables together between source and target. The find-tables tool:
- Pulls all tables in the source (applying a supplied allowed-schemas filter)
- Pulls all tables from the target
- Uses Levenshtein distance to match tables
- Finally, it prints a JSON list of tables which can be a reference for the validation run config.
Note that our score cutoff default is a 0.8, which was manually tested to be an accurate value. If no matches occur, reduce this value.
data-validation find-tables --source-conn source --target-conn target \
--allowed-schemas pso_data_validator \
--score-cutoff 0.8
If you want to add an Ibis Data Source which exists, but was not yet supported in the Data Validation tool, it is a simple process.
-
In data_validation/data_validation.py
- Import the extened Client for the given source (ie. from ibis.sql.mysql.client import MySQLClient).
- Add the "": Client to the global CLIENT_LOOKUP dictionary.
-
In third_party/ibis/ibis_addon/operations.py
- Add the RawSQL operator to the data source registry (for custom filter support).
-
You are done, you can reference the data source via the config.
- Config: {"source_type": "", ...KV Values required in Client...}
#!/bin/bash
export COMPOSER_ENV=""
export LOCATION=""
echo "Creating Composer Env: $COMPOSER_ENV"
gcloud services enable composer.googleapis.com
gcloud composer environments create $COMPOSER_ENV --location=$LOCATION --python-version=3
echo "Updating Composer Env Reqs: $COMPOSER_ENV"
# Composer builds Pandas and BigQuery for you, these should be stripped out
cat requirements.txt | grep -v pandas | grep -v google-cloud-bigquery > temp_reqs.txt
gcloud composer environments update $COMPOSER_ENV --location=$LOCATION --update-pypi-packages-from-file=temp_reqs.txt
rm temp_reqs.txt
# Deploy Package to Composer (the hacky way)
echo "Rebuilding Data Validation Package in GCS"
export GCS_BUCKET_PATH=$(gcloud composer environments describe $COMPOSER_ENV --location=$LOCATION | grep dagGcsPrefix | awk '{print $2;}')
gsutil rm -r $GCS_BUCKET_PATH/data_validation
gsutil cp -r data_validation $GCS_BUCKET_PATH/data_validation
# Deploy Test DAG to Composer
echo "Pushing Data Validation Test Operator to GCS"
gsutil cp tests/test_data_validation_operators.py $GCS_BUCKET_PATH/
Contributions are welcome. See the contributing guide for details.