Pattern statement

A pattern is a construct that maps string tuples to tabular expressions.

Each pattern must declare a pattern name and optionally define a pattern mapping. Patterns that define a mapping return a tabular expression when invoked. Separate any two statements by a semicolon.

Empty patterns are patterns that are declared but don’t define a mapping. When invoked, they return error SEM0036 along with the details of the missing pattern definitions in the HTTP header.

Middle-tier applications that provide a Kusto Query Language (KQL) experience can use the returned details as part of their process to enrich KQL query results. For more information, see Working with middle-tier applications.

Syntax

• Declare an empty pattern:

  declare pattern PatternName ;

• Declare and define a pattern:

  declare pattern PatternName = (ArgName : ArgType [, … ]) [[ PathName : PathArgType ]]

  {

        ( ArgValue1_1 [, ArgValue2_1, … ] ) [ .[ PathValue_1 ] ] = { expression1 } ;

      [ ( ArgValue1_2 [, ArgValue2_2, … ] ) [ .[ PathValue_2 ] ] = { expression2 } ; … ]

  } ;

• Invoke a pattern:

  • PatternName ( ArgValue1 [, ArgValue2 …] ).PathValue
  • PatternName ( ArgValue1 [, ArgValue2 …] ).["PathValue"]

Parameters

Examples

The examples in this section show how to use the syntax to help you get started.

Define a simple pattern

This example defines a pattern that maps states to an expression that returns its capital/major city.

declare pattern country = (name:string)[state:string]
{
  ("USA").["New York"] = { print Capital = "Albany" };
  ("USA").["Washington"] = { print Capital = "Olympia" };
  ("Canada").["Alberta"] = { print Capital = "Edmonton" };
};
country("Canada").Alberta

Output

Define a scoped pattern

This example defines a pattern to scope data and metrics of application data. The pattern is invoked to return a union of the data.

declare pattern App = (applicationId:string)[scope:string]
{
    ('a1').['Data']    = { range x from 1 to 5 step 1 | project App = "App #1", Data    = x };
    ('a1').['Metrics'] = { range x from 1 to 5 step 1 | project App = "App #1", Metrics = rand() };
    ('a2').['Data']    = { range x from 1 to 5 step 1 | project App = "App #2", Data    = 10 - x };
    ('a3').['Metrics'] = { range x from 1 to 5 step 1 | project App = "App #3", Metrics = rand() };
};
union App('a2').Data, App('a1').Metrics

Output

Normalization

There are syntax variations for invoking patterns. For example, the following union returns a single pattern expression since all the invocations are of the same pattern.

declare pattern app = (applicationId:string)[eventType:string]
{
    ("ApplicationX").["StopEvents"] = { database("AppX").Events | where EventType == "StopEvent" };
    ("ApplicationX").["StartEvents"] = { database("AppX").Events | where EventType == "StartEvent" };
};
union
  app("ApplicationX").StartEvents,
  app('ApplicationX').StartEvents,
  app("ApplicationX").['StartEvents'],
  app("ApplicationX").["StartEvents"]

No wildcards

There’s no special treatment given to wildcards in a pattern. For example, the following query returns a single missing pattern invocation.

declare pattern app = (applicationId:string)[eventType:string]
{
    ("ApplicationX").["StopEvents"] = { database("AppX").Events | where EventType == "StopEvent" };
    ("ApplicationX").["StartEvents"] = { database("AppX").Events | where EventType == "StartEvent" };
};
union app("ApplicationX").["*"]
| count

Output semantic error

Work with middle-tier applications

A middle-tier application provides its users with the ability to use KQL and wants to enhance the experience by enriching the query results with augmented data from its internal service.

To this end, the application provides users with a pattern statement that returns tabular data that their users can use in their queries. The pattern’s arguments are the keys the application will use to retrieve the enrichment data.

When the user runs the query, the application doesn’t parse the query itself but instead uses the error returned by an empty pattern to retrieve the keys it requires. So it prepends the query with the empty pattern declaration, sends it to the cluster for processing, and then parses the returned HTTP header to retrieve the values of missing pattern arguments. The application uses these values to look up the enrichment data and builds a new declaration that defines the appropriate enrichment data mapping.

Finally, the application prepends the new definition to the query, resends it for processing, and returns the result it receives to the user.

Example

In the examples, a pattern is declared, defined, and then invoked.

Declare an empty pattern

In this example, a middle-tier application enriches queries with longitude/latitude locations. The application uses an internal service to map IP addresses to longitude/latitude locations, and provides a pattern called map_ip_to_longlat. When the query is run, it returns an error with missing pattern definitions:

map_ip_to_longlat("10.10.10.10")

Declare and define a pattern

The application does not parse this query and hence does not know which IP address (10.10.10.10) was passed to the pattern. So it prepends the user query with an empty map_ip_to_longlat pattern declaration and sends it for processing:

declare pattern map_ip_to_longlat;
map_ip_to_longlat("10.10.10.10")

The application receives the following error in response.

Invoke a pattern

The application inspects the error, determines that the error indicates a missing pattern reference, and retrieves the missing IP address (10.10.10.10). It uses the IP address to look up the enrichment data in its internal service and builds a new pattern defining the mapping of the IP address to the corresponding longitude and latitude data. The new pattern is prepended to the user’s query and run again.

This time the query succeeds because the enrichment data is now declared in the query, and the result is sent to the user.

declare pattern map_ip_to_longlat = (address:string)
{
  ("10.10.10.10") = { print Lat=37.405992, Long=-122.078515 };
};
map_ip_to_longlat("10.10.10.10")

Output