From a top level perspective, Flowman works like a build tool like make or maven. Of course in contrast to classical build tools, the project specification in Flowman also contains the logic to be build (normally that is separated in source code files which get compiles or otherwise processed with additional tools).

Each target supports at least some build phases

Common Fields#

All Targets support the following common fields:

  • kind (mandatory) (type: string): The kind of the target

  • before (optional) (type: list:string): List of targets that can only be executed after this target

  • after (optional) (type: list:string): List of targets that need to be executed before this target

  • labels (optional) (type: map): Optional list of labels.

Execution Model#

Normally, targets are executed within the scope of a job. Each target listed in a job can be active during some phases of the lifecycle. For example when you execute the VERIFY lifecycle, the following phases are actually executed by Flowman:

  • VALIDATE - perform some optional pre-execution checks

  • CREATE - create and/or migrate target data models

  • BUILD - populate target data models with fresh data

  • VERIFY - perform some optional post-execution checks

Depending on the kind, each target may contribute some work to each execution phase. Moreover, Flowman also checks the state of each target. Only dirty targets will be rebuilt (except when you forcibly want to execute all targets with the --force command line switch).

A target is considered dirty for the current execution phase, when it needs some work to be performed. For example, a JDBC table is considered dirty in the CREATE phase, when it does not exist yet or when its current schema (i.e. columns and data types) does not match the desired schema. Then Flowman will take care of creating and/or migrating the JDBC table.

Typically, a target is considered dirty during the BUILD phase, when it does not contain any data (possibly only within the desired partition, see relation target for more information).

Target Types#

Flowman supports different target types, each used for a different kind of physical entity or build recipe.


For each target Flowman provides the following execution metric:

  • metric: “target_runtime”

  • labels:

    • category: “target”

    • kind: The kind of the target

    • namespace: The name of the namespace

    • project: The name of the project