Overview
Please make sure you know common Glossary and SpEL (especially the Data types section) before proceeding further.
This part of the documentation describes various ways of customizing Nussknacker - from adding own Components to adding listeners for various Designer actions. The main way of adding customizations to Nussknacker is ServiceLoader
Please make sure to put jars with custom code on right classpath
- Customizations of model (in particular
ComponentProviders
) can be loaded by adding libs/classes to dedicatedcomponents/common/extra
,components/lite/extra
orcomponents/flink/extra
directory. For advanced usages, you can configuremodelConfig.classPath
in Model config. - Code of Designer customizations should go to the main Designer classpath (e.g. put the jars in the
lib
folder)
Types
Types of expressions are based on Java types. Nussknacker provides own abstraction of type, which can contain more information about given type than pure Java class - e.g. object type (like in Typescript) is represented in runtime as Java Map
, but during compilation we know the structure of this map.
We also handle union types (again, similar to Typescript) and we have Unknown
type which is represented as Java Object
in runtime, but behaves a bit like Typescript any (please note that Unknown
should be avoided as default Security settings settings prohibit omitting typechecking with Unknown
.
TypingResult
is the main class (sealed trait) that represents type of expression in Nussknacker.
Typed
object has many methods for constructing TypingResult
Components and ComponentProviders
Components are main method of customizing Nussknacker. Components are created by configured ComponentProvider
instances.
There are following types of components:
SourceFactory
SinkFactory
CustomStreamTransformer
- types of transformations depend on type of EngineService
- mainly for defining stateless enrichments To read more see ComponentProvider API
Deployment of scenarios
The Designer uses DeploymentManager
interface to perform actions on scenarios (deploy / cancel / etc.). All providers that are available in distribution deployment are located in managers
directory and are added to the Designer classpath.
If you want to implement own DeploymentManager
, you should implement this interface, package it, add to classpath and configure scenario type to use it. More info you can find on
DeploymentManagerConfiguration page
Other SPIs for Nussknacker customization (documentation will follow soon...)
Model customization
- Flink specific
- CustomParameterValidator
- ObjectNaming
- ToJsonEncoder
- WithExceptionExtractor
- ModelConfigLoader
- ProcessMigrations
- DictServicesFactory
Designer customization
- ProcessChangeListenerFactory
- Security
- CountsReporterCreator
- AdditionalInfoProvider
- CustomProcessValidatorFactory
- AdditionalUIConfigProviderFactory
Modules architecture and conventions
The diagram below shows dependencies between modules. You can see two main groups on it :
- API modules
- Utils
API modules contains interfaces that are needed by our core modules (on both Designer and runtime side).
On the other hand Utils modules contain classes built on top of API which can be used in extensions but are not mandatory. API of Utils modules can be changed more often than API inside API modules
Both API modules and Utils modules have several modules with -components
part in name. They should be used to build own Components
nussknacker-scenario-api
contains classes needed to operate on scenarios: creating it via DSL, marshalling to JSON, etc.
nussknacker-deployment-manager-api
contains interfaces needed to create own DeploymentManager
that can be used to scenario execution.
nussknacker-extensions-*
contains other extensions API.
Your code should depend only on nussknacker-xxx-api
or nussknacker-xxx-components-utils
/nussknacker-xxx-extensions-utils
packages and not on implementation modules, like
nussknacker-interpreter
, nussknacker-flink-executor
, nussknacker-lite-runtime
or other internal
modules. They should only be needed in test
scope.
If you find you need to depend on those modules, please bear in mind that they contain implementation details and their API should not be considered stable.
Plug-ins packaging
The plug-in jar should be fatjar containing all libraries necessary for running your customization,
except for dependencies provided by execution engine. In particular, for custom component implementation,
following dependencies should be marked as provided
and not be part of customization jar:
- All Nussknacker modules with names ending in
-api
, e.g.nussknacker-components-api
,nussknacker-flink-components-api
,nussknacker-lite-components-api
nussknacker-utils
,nussknacker-components-utils
,nussknacker-helpers-utils
(are provided indefaultModel.jar
)nussknacker-flink-components-utils
(is provided inflinkExecutor.jar
)- Basic Flink dependencies:
flink-streaming-scala
,flink-runtime
,flink-statebackend-rocksdb
etc. for Flink components (are provided inflinkExecutor.jar
) nussknacker-kafka-utils
for Streaming components in Lite engine
Please remember that provided
dependency are not transitive, i.e. if you depend on e.g. nussknacker-flink-kafka-components-utils
you still have to declare dependency on nussknacker-flink-components-utils
explicitly
(see Maven documentation for further info).