Data Quality Rules
Telmai Data Correctness Rules
One of the key metrics tracked by Telmai is the “Correctness” metric. It is defined as the percentage of records which are valid based on a corresponding rule or combination of rules. Correctness is - For example, for a table with 1M rows; if 50,000 records are invalid, correctness is then 95%.
The row validity checks (data expectations) are defined by users via Correctness Rules.
Correctness Rules
Correctness Rules are evaluated at row level. A single rule may validate one or more attribute values (plain or nested object) against a condition.
Rules are created following SQL-like syntax that uses variables and functions to perform the validation. The main key features for rules:
Eases the process of applying the same rule against multiple tables through templating
Rules DSL (domain specific language) allows for version control and management in a simple human-readable format
Optimized algorithms for unlimited number of rules for faster and more efficient processing
Building a Correctness Rule
Creating a correctness rule requires performing the following steps:
Define variables that will be used in the rule
Variables can be assigned to static lists of values, range, entered manually or uploaded, from files; etc.
Map variables to attributes used in the expression
Write rule expression
Define and Map Variables
Users can define variables to be used in their rule. Supported variable types:
Array of strings
Constant string value
Range (date-time or integer)
Reference to an attribute
Example:
We can define the following two variables (accepted_country_list, var_1):
accepted_country_list = [USA, Canada]var_1 = country_name
User can then write a rule like following
VALIDATE var_1
EXPECT value IN accepted_country_list
This rule checks if a value from the attribute country_name is in the list of accepted values. First it defines an expression var_1, which is a trivial expression, which doesn’t add any transformation to the original value. And after that it sets an expectation that the value of the expression should be in the list to which the previously defined variable is assigned.
Writing Rule Expression
A correctness rule consists of three parts:
WITH <transformation>
Variable transformation:
Optional section. It allows the user to create a transformation of one or more attributes to be validated and reused within the expression multiple times
Example:
WITH concat(country, '|', state) AS loc
This creates a new variable “loc” that can be used in validation
VALIDATE
<evaluated_expression>
Validated expression:
This could be simply validating an attribute as is, or adding transformation or case clause.
Example:
VALIDATE
case
when loc = 'US|CA' then country IN ('Alameda', 'Contra-Costa')
when loc = 'UAE|Dubai' then country IS NULL
else false //default output
End
EXPECT
<expectation>
Expected results
Expectation for the expression above. After the “EXPECT” keyword, the validation begins.
Example:
EXPECT is_true
this means the value of validated expression is expected to be true, and anything else is a violation)
How to get alerted for rule violation?
Once a rule is created, users can assign the rule to a policy and specify the threshold for alerting.
Variables transformation
This step allows defining variables from combining existing attributes to be used at a later step of validation. This can simplify writing the evaluated expression specially in case when multiple checks are applied.
To define a transformation, you can follow the following syntax:
WITH <transformation> AS <new_variable_name>
Telmai uses standard Spark SQL syntax for those transformations.
Evaluated expression
This is a Spark SQL expression. It allows users to define what will be evaluated, and exposes a wide range of SQL functions to allow for conditional checks.
Scenario 1: Validate attribute value as is, without transformations
VALIDATE my_column// Some codeThis validates “my_column” without any transformation
Scenario 2: Apply standard functions to attribute value
VALIDATE foo(my_column)This validates “my_column” after applying the “foo” function to it.
Scenario 3: Cases
VALIDATE case when my_colA <> 0 then my_colB else my_colC endThis validates the rule by using my_colB when my_colA is not 0; otherwise, it uses my_colC
Expectation
Expectation is defined through the context and the expected value.
EXPECT <context> [<operator>] <expected_value>Expectation context
The expectation context (<context>) is how the value gets evaluated. It can be any of the following:
Context
Description
COMPRESSED_PATTERN
Compressed pattern of the expression value
DATE_TIME_VALUE
Expression value as Datetime. Refer here for datetime patterns
EXPANDED_PATTERN
Expanded pattern of the expression value
FREQUENCY
Frequency of the expression value
LENGTH
String length of the expression value
NUMERIC_VALUE
Expression value as number
SPACE_COUNT
Count of spaces in the expression value
SPEC_CHAR_COUNT
Spec characters count of the expression value
VALUE
String representation of the expression value
Expectation operator
The expectation operator (<operator>) is an optional field. It’s the condition operator for the context and the expected value. It can be from the following list:
Operator
Description
IN
Value is in the list.
Expected value must be in the list.
IN_RANGE
Value in range including boundaries. Expected value must be in the range.
NOT
Boolean negation operator
NOT_IN
Value is not in the list
Expected value must not be in the list.
OUT_OF_RANGE
Value out of range
Expected value must be outside of the range.
Expected value
Expected value (<expected_value>) can be set using a static value or the boolean-functions belo
Static value
Example
List
["a", "b", "c"]
Range
(0,100)
String
"abcd"
Variable
var_1
Boolean function
Description
CONTAINS_PII
True if expression contains PII data
CONTAINS_PII_CREDIT_CARD
True if expression contains credit card number pattern
CONTAINS_PII_IP_ADDRESS
True if expression contains IP address
CONTAINS_PII_PHONE_NUMBER
True if expression contains phone number pattern
CONTAINS_PII_SSN
True if expression contains social security number pattern
IS_ALPHA
Value is only from alphabetic characters
IS_DATE
Value is UTC date pattern
IS_DATE_TIME
Value is UTC date-time pattern
IS_EMAIL
Value is email
IS_FALSE
Value is false (case insensitive string comparison, and not boolean)
IS_NUMBER
Value is numeric value
IS_TRIMMED_STRING
Value is a trimmed string
IS_TRUE
Value is true (case insensitive string comparison, and not boolean)
REGEX(pattern)
Value satisfy regex pattern
Examples:
VALIDATE Var_1 expect is_date
# Check if Var_1 is a valid date.
VALIDATE Var_1 expect contains_pii
# Check if Var_1 contains Personally Identifiable Information (PII)
VALIDATE Var_1 expect numeric_value in_range (1, 20)
# Ensure Var_1 is a number within range 1 to 20
VALIDATE Var_1 expect numeric_value out_of_range (50, 60)
# Ensure Var_1’s numeric value is outside a length range
VALIDATE Var_1 expect expanded_pattern IN ['D', 'DD']
# Validate Var_1 using an expanded pattern
VALIDATE Var_1 expect compressed_pattern in ['D', 'L']
# Validate Var_1 using a compressed pattern
VALIDATE Var_1 expect value IN ['USA', 'Germany', 'India']
# Check if Var_1 value is in the accepted country list
VALIDATE Var_1 expect value='Telm'
# Ensure Var_1 exactly matches the value Telm
VALIDATE concat(Var_1, '.', Var_2) expect value='Telm.ai'
# Validate full name format as Telm.ai, where Var_1=Telm and Var_2=ai
VALIDATE Var_1 expect length<>5
# Ensure Var_1 value is of length not equal to 5
VALIDATE Var_1 expect frequency=1
# Ensure Var_1 has exactly one occurrence
VALIDATE Var_1 expect regex('^[a-zA-Z]+$')
# Validate Var_1 contains only letters (no numbers/special characters) Combining Expectations
User is able to combine multiple expectations using AND/OR operator:
EXPECT <expectation_1> AND <expectation_2> OR <expectation_3>Examples:
VALIDATE case when Status='Active' and LastLogin < current_date - 30 then 'Inactive' else 'Valid' end expect value='Valid'
# Mark users as inactive if they haven't logged in for 30+ days, otherwise expect 'Valid'
with concat(Country1, ' | ', Country2) AS loc VALIDATE loc expect length<20
# Ensure concatenated location (Country1 | Country2) has a length less than 20
VALIDATE Password expect spec_char_count>=1 AND space_count=0 AND length>=8
# Ensure Password has at least 1 special character, no spaces, and is at least 8 characters long
VALIDATE Password expect (spec_char_count>=1 AND space_count=0) OR length>=20
# Ensure Password has at least 1 special character and no spaces, or is at least 20 characters long Rule Templates
Rule Templates is a predefined rule where the expectation is defined, then applied to 1 or more attributes. Users are recommended to use rule templates when the same check is applied on multiple tables. This centralizes the definition of the rule, and allows managing the rule from a single place.
To use a rule template, the user only maps attributes within their table to variables within the rules.
Creating a Rule
The following flow diagram describes the steps to create a new rule or rule template:
Define rule info:
Rule name
Rule description
[Optional] Template being used (this is only if creating a rule)
Set static variables:
These are variables that are validated against, and are not dynamic in nature; ex: list of allowed countries
Set input variables:
These are variables that are mapped to attributes in the table
In case this is a rule, user will need to map the attributes to variables
Write expression
Note; In case user is using an existing template, only input variables need to be mapped to variables in the table
Alerting Policies
Automatically, all created rules are monitored via an out-of-box policy “Correctness Rules Violation”. Users can add new policies and pick the desired rules and threshold to be alerted on
Last updated