Convert data by using dataflow conversions

Article
08/06/2024

Important

Azure IoT Operations Preview – enabled by Azure Arc is currently in preview. You shouldn't use this preview software in production environments.

You'll need to deploy a new Azure IoT Operations installation when a generally available release is made available. You won't be able to upgrade a preview installation.

See the Supplemental Terms of Use for Microsoft Azure Previews for legal terms that apply to Azure features that are in beta, preview, or otherwise not yet released into general availability.

You can use dataflow conversions to transform data in Azure IoT Operations. The conversion element in a dataflow is used to compute values for output fields. You can use input fields, available operations, data types, and type conversions in dataflow conversions.

The dataflow conversion element is used to compute values for output fields:

- inputs:
  - *.Max   # - $1
  - *.Min   # - $2
  output: ColorProperties.*
  expression: ($1 + $2) / 2

There are several aspects to understand about conversions:

Reference to input fields: How to reference values from input fields in the conversion formula.
Available operations: Operations that can be utilized in conversions. For example, addition, subtraction, multiplication, and division.
Data types: Types of data that a formula can process and manipulate. For example, integer, floating point, and string.
Type conversions: How data types are converted between the input field values, the formula evaluation, and the output fields.

Input fields

In conversions, formulas can operate on static values like a number such as 25 or parameters derived from input fields. A mapping defines these input fields that the formula can access. Each field is referenced according to its order in the input list:

- inputs:
  - *.Max        # - $1
  - *.Min        # - $2
  - *.Mid.Avg    # - $3
  - *.Mid.Mean   # - $4
  output: ColorProperties.*
  expression: ($1, $2, $3, $4)

In this example, the conversion results in an array containing the values of [Max, Min, Mid.Avg, Mid.Mean]. The comments in the YAML file (# - $1, # - $2) are optional, but they help to clarify the connection between each field property and its role in the conversion formula.

Data types

Different serialization formats support various data types. For instance, JSON offers a few primitive types: string, number, Boolean, and null. Also included are arrays of these primitive types. In contrast, other serialization formats like Avro have a more complex type system, including integers with multiple bit field lengths and timestamps with different resolutions. Examples are milliseconds and microseconds.

When the mapper reads an input property, it converts it into an internal type. This conversion is necessary for holding the data in memory until it's written out into an output field. The conversion to an internal type happens regardless of whether the input and output serialization formats are the same.

The internal representation utilizes the following data types:

Type	Description
`bool`	Logical true/false.
`integer`	Stored as 128-bit signed integer.
`float`	Stored as 64-bit floating point number.
`string`	A UTF-8 string.
`bytes`	Binary data, a string of 8-bit unsigned values.
`datetime`	UTC or local time with nanosecond resolution.
`time`	Time of day with nanosecond resolution.
`duration`	A duration with nanosecond resolution.
`array`	An array of any types listed previously.
`map`	A vector of (key, value) pairs of any types listed previously.

Input record fields

When an input record field is read, its underlying type is converted into one of these internal type variants. The internal representation is versatile enough to handle most input types with minimal or no conversion. However, some input types require conversion or are unsupported. Some examples:

Avro UUID type: It's converted to a string because there's no specific UUID type in the internal representation.
Avro decimal type: It isn't supported by the mapper, so fields of this type can't be included in mappings.
Avro duration type: Conversion can vary. If the months field is set, it's unsupported. If only days and milliseconds are set, it's converted to the internal duration representation.

For some formats, surrogate types are used. For example, JSON doesn't have a datetime type and instead stores datetime values as strings formatted according to ISO8601. When the mapper reads such a field, the internal representation remains a string.

Output record fields

The mapper is designed to be flexible by converting internal types into output types to accommodate scenarios where data comes from a serialization format with a limited type system. The following examples show how conversions are handled:

Numeric types: These types can be converted to other representations, even if it means losing precision. For example, a 64-bit floating-point number (f64) can be converted into a 32-bit integer (i32).
Strings to numbers: If the incoming record contains a string like 123 and the output field is a 32-bit integer, the mapper converts and writes the value as a number.
Strings to other types:
- If the output field is datetime, the mapper attempts to parse the string as an ISO8601 formatted datetime.
- If the output field is binary/bytes, the mapper tries to deserialize the string from a base64-encoded string.
Boolean values:
- Converted to 0/1 if the output field is numerical.
- Converted to true/false if the output field is string.

Explicit type conversions

Although the automatic conversions operate as you might expect based on common implementation practices, there are instances where the right conversion can't be determined automatically and results in an unsupported error. To address these situations, several conversion functions are available to explicitly define how data should be transformed. These functions provide more control over how data is converted and help maintain data integrity even when automatic methods fall short.

Use a conversion formula with types

In mappings, an optional formula can specify how data from the input is processed before being written to the output field. If no formula is specified, the mapper copies the input field to the output by using the internal type and conversion rules.

If a formula is specified, the data types available for use in formulas are limited to:

Integers
Floating-point numbers
Strings
Booleans
Arrays of the preceding types
Missing value

Map and byte can't participate in formulas.

Types related to time (datetime, time, and duration) are converted into integer values that represent time in seconds. After formula evaluation, results are stored in the internal representation and not converted back. For example, datetime converted to seconds remains an integer. If the value will be used in datetime fields, an explicit conversion method must be applied. An example is converting the value into an ISO8601 string that's automatically converted to the datetime type of the output serialization format.

Use irregular types

Special considerations apply to types like arrays and missing value.

Arrays

Arrays can be processed by using aggregation functions to compute a single value from multiple elements. For example, by using the input record:

{
    "Measurements": [2.34, 12.3, 32.4]
}

With the mapping:

- inputs:
  - Measurements # - $1
  output: Measurement
  expression: min($1)

This configuration selects the smallest value from the Measurements array for the output field.

It's also possible to use functions that result in a new array:

- inputs:
  - Measurements # - $1
  output: Measurements
  expression: take($1, 10)  # taking at max 10 items

Arrays can also be created from multiple single values:

- inputs:
  - minimum  # - - $1
  - maximum  # - - $2
  - average  # - - $3
  - mean     # - - $4
  output: stats
  expression: ($1, $2, $3, $4)

This mapping creates an array that contains the minimum, maximum, average, and mean.

Missing value

Missing value is a special type used in scenarios, such as:

Handling missing fields in the input by providing an alternative value.
Conditionally removing a field based on its presence.

Example mapping that uses a missing value:

{
    "Employment": {      
      "Position": "Analyst",
      "BaseSalary": 75000,
      "WorkingHours": "Regular"
    }
}

The input record contains the BaseSalary field, but possibly that's optional. Let's say that if the field is missing, a value must be added from a contextualization dataset:

{
  "Position": "Analyst",
  "BaseSalary": 70000,
  "WorkingHours": "Regular"
}

A mapping can check if the field is present in the input record. If the field is found, the output receives that existing value. Otherwise, the output receives the value from the context dataset. For example:

- inputs:
  - BaseSalary  # - - - - - - - - - - $1
  - $context(position).BaseSalary #  - $2 
  output: BaseSalary
  expression: if($1 == (), $2, $1)

The conversion uses the if function that has three parameters:

The first parameter is a condition. In the example, it checks if the BaseSalary field of the input field (aliased as $1) is the missing value.
The second parameter is the result of the function if the condition in the first parameter is true. In this example, it's the BaseSalary field of the contextualization dataset (aliased as $2).
The third parameter is the value for the condition if the first parameter is false.

Available functions

Functions can be used in the conversion formula to perform various operations:

min to select a single item from an array
if to select between values
String manipulation (for example, uppercase())
Explicit conversion (for example, ISO8601_datetime)
Aggregation (for example, avg())

Available operations

Dataflows offer a wide range of out-of-the-box conversion functions that allow users to easily perform unit conversions without the need for complex calculations. These predefined functions cover common conversions such as temperature, pressure, length, weight, and volume. The following list shows the available conversion functions, along with their corresponding formulas and function names:

Conversion	Formula	Function name
Celsius to Fahrenheit	F = (C * 9/5) + 32	`cToF`
PSI to bar	Bar = PSI * 0.0689476	`psiToBar`
Inch to cm	Cm = inch * 2.54	`inToCm`
Foot to meter	Meter = foot * 0.3048	`ftToM`
Lbs to kg	Kg = lbs * 0.453592	`lbToKg`
Gallons to liters	Liters = gallons * 3.78541	`galToL`

In addition to these unidirectional conversions, we also support the reverse calculations:

Conversion	Formula	Function name
Fahrenheit to Celsius	C = (F - 32) * 5/9	`fToC`
Bar to PSI	PSI = bar / 0.0689476	`barToPsi`
Cm to inch	Inch = cm / 2.54	`cmToIn`
Meter to foot	Foot = meter / 0.3048	`mToFt`
Kg to lbs	Lbs = kg / 0.453592	`kgToLb`
Liters to gallons	Gallons = liters / 3.78541	`lToGal`

These functions are designed to simplify the conversion process. They allow users to input values in one unit and receive the corresponding value in another unit effortlessly.

We also provide a scaling function to scale the range of value to the user-defined range. For the example scale($1,0,10,0,100), the input value is scaled from the range 0 to 10 to the range 0 to 100.

Moreover, users have the flexibility to define their own conversion functions by using simple mathematical formulas. Our system supports basic operators such as addition (+), subtraction (-), multiplication (*), and division (/). These operators follow standard rules of precedence. For example, multiplication and division are performed before addition and subtraction. Precedence can be adjusted by using parentheses to ensure the correct order of operations. This capability empowers users to customize their unit conversions to meet specific needs or preferences, enhancing the overall utility and versatility of the system.

For more complex calculations, functions like sqrt (which finds the square root of a number) are also available.

Available arithmetic, comparison, and Boolean operators grouped by precedence

Operator	Description
^	Exponentiation: $1 ^ 3

Because Exponentiation has the highest precedence, it's executed first unless parentheses override this order:

$1 * 2 ^ 3 is interpreted as $1 * 8 because the 2 ^ 3 part is executed first, before multiplication.
($1 * 2) ^ 3 processes the multiplication before exponentiation.

Operator	Description
-	Negation
!	Logical not

Negation and Logical not have high precedence, so they always stick to their immediate neighbor, except when exponentiation is involved:

-$1 * 2 negates $1 first, and then multiplies.
-($1 * 2) multiplies, and then negates the result.

Operator	Description
*	Multiplication: $1 * 10
/	Division: $1 / 25 (Result is an integer if both arguments are integers, otherwise float)
%	Modulo: $1 % 25

Multiplication, Division, and Modulo, having the same precedence, are executed from left to right, unless the order is altered by parentheses.

Operator	Description
+	Addition for numeric values, concatenation for strings
-	Subtraction

Addition and Subtraction are considered weaker operations compared to the operations in the previous group:

$1 + 2 * 3 results in $1 + 6 because 2 * 3 is executed first because of the higher precedence of multiplication.
($1 + 2) * 3 prioritizes Addition before Multiplication.

Operator	Description
<	Less than
>	Greater than
<=	Less than or equal to
>=	Greater than or equal to
==	Equal to
!=	Not equal to

Comparisons operate on numeric, Boolean, and string values. Because they have lower precedence than arithmetic operators, no parentheses are needed to compare results effectively:

$1 * 2 <= $2 is equivalent to ($1 * 2) <= $2.

Operator	Description
\|\|	Logical OR
&&	Logical AND

Logical operators are used to chain conditions:

$1 > 100 && $2 > 200

Share via