This tutorial walks you through the creation, from scratch, of a complete Talend input component for Hazelcast using the Talend Component Kit (TCK) framework. Hazelcast is an in-memory distributed system that can store data, which makes it a good example of input component for distributed systems. This is enough for you to get started with this tutorial, but you can find more information about it here: hazelcast.org/. A TCK project is a simple Java project with specific configurations and dependencies. You can choose your preferred build tool from Maven or Gradle as TCK supports both. In this tutorial, Maven is used. The first step consists in generating the project structure using Talend Starter Toolkit . Go to starter-toolkit.talend.io/ and fill in the project information as shown in the screenshots below, then click Finish and Download as ZIP. image::tutorial_hazelcast_generateproject_1.png[] image::tutorial_hazelcast_generateproject_2.png[] Extract the ZIP file into your workspace and import it to your preferred IDE. This tutorial uses Intellij IDE, but you can use Eclipse or any other IDE that you are comfortable with. You can use the Starter Toolkit to define the full configuration of the component, but in this tutorial some parts are configured manually to explain key concepts of TCK. The generated pom.xml file of the project looks as follows: Change the name tag to a more relevant value, for example: Component Hazelcast. The component-api dependency provides the necessary API to develop the components. talend-component-maven-plugin provides build and validation tools for the component development. The Java compiler also needs a Talend specific configuration for the components to work correctly. The most important is the -parameters option that preserves the parameter names needed for introspection features that TCK relies on. Download the mvn dependencies declared in the pom.xml file: You should get a BUILD SUCCESS at this point: Create the project structure: Create the component Java packages. Packages are mandatory in the component model and you cannot use the default one (no package). It is recommended to create a unique package per component to be able to reuse it as dependency in other components, for example to guarantee isolation while writing unit tests. The project is now correctly set up. The next steps consist in registering the component family and setting up some properties. Registering every component family allows the component server to properly load the components and to ensure they are available in Talend Studio. The family registration happens via a package-info.java file that you have to create. Move to the src/main/java/org/talend/components/hazelcast package and create a package-info.java file: @Components: Declares the family name and the categories to which the component belongs. @Icon: Defines the component family icon. This icon is visible in the Studio metadata tree. Talend Component Kit supports internationalization (i18n) via Java properties files. Using these files, you can customize and translate the display name of properties such as the name of a component family or, as shown later in this tutorial, labels displayed in the component configuration. Go to src/main/resources/org/talend/components/hazelcast and create an i18n Messages.properties file as below: You can define the component family icon in the package-info.java file. The icon image must exist in the resources/icons folder. TCK supports both SVG and PNG formats for the icons. Create the icons folder and add an icon image for the Hazelcast family. This tutorial uses the Hazelcast icon from the official GitHub repository that you can get from: avatars3.githubusercontent.com/u/1453152?s=200&v=4 Download the image and rename it to Hazelcast_icon32.png. The name syntax is important and should match _icon.32.png. The component registration is now complete. The next step consists in defining the component configuration. All Input and Output (I/O) components follow a predefined model of configuration. The configuration requires two parts: Datastore: Defines all properties that let the component connect to the targeted system. Dataset: Defines the data to be read or written from/to the targeted system. Connecting to the Hazelcast cluster requires the IP address, group name and password of the targeted cluster. In the component, the datastore is represented by a simple POJO. Create a HazelcastDatastore.java class file in the src/main/java/org/talend/components/hazelcast folder. Define the i18n properties of the datastore. In the Messages.properties file let add the following lines: The Hazelcast datastore is now defined. Hazelcast includes different types of datastores. You can manipulate maps, lists, sets, caches, locks, queues, topics and so on. This tutorial focuses on maps but still applies to the other data structures. Reading/writing from a map requires the map name. Create the dataset class by creating a HazelcastDataset.java file in src/main/java/org/talend/components/hazelcast. The @Dataset annotation marks the class as a dataset. Note that it also references a datastore, as required by the components model. Just how it was done for the datastore, define the i18n properties of the dataset. To do that, add the following lines to the Messages.properties file. The component configuration is now ready. The next step consists in creating the Source that will read the data from the Hazelcast map. The Source is the class responsible for reading the data from the configured dataset. A source gets the configuration instance injected by TCK at runtime and uses it to connect to the targeted system and read the data. Create a new class as follows. The source also needs i18n properties to provide a readable display name. Add the following line to the Messages.properties file. At this point, it is already possible to see the result in the Talend Component Web Tester to check how the configuration looks like and validate the layout visually. To do that, execute the following command in the project folder. This command starts the Component Web Tester and deploys the component there. Access localhost:8080/. The source is set up. It is now time to start creating some Hazelcast specific code to connect to a cluster and read values for a map. Add the hazelcast-client Maven dependency to the pom.xml of the project, in the dependencies node. Add a Hazelcast instance to the @PostConstruct method. Declare a HazelcastInstance attribute in the source class. Any non-serializable attribute needs to be marked as transient to avoid serialization issues. Implement the post construct method. The component configuration is mapped to the Hazelcast client configuration to create a Hazelcast instance. This instance will be used later to get the map from its name and read the map data. Only the required configuration in the component is exposed to keep the code as simple as possible. Implement the code responsible for reading the data from the Hazelcast map through the Producer method. The Producer implements the following logic: Check if the map iterator is already initialized. If not, get the map from its name and initialize the map iterator. This is done in the @Producer method to ensure the map is initialized only if the next() method is called (lazy initialization). It also avoids the map initialization in the PostConstruct method as the Hazelcast map is not serializable. All the objects initialized in the PostConstruct method need to be serializable as the source can be serialized and sent to another worker in a distributed cluster. From the map, create an iterator on the map keys that will read from the map. Transform every key/value pair into a Talend Record with a "key, value" object on every call to next(). The RecordBuilderFactory class used above is a built-in service in TCK injected via the Source constructor. This service is a factory to create Talend Records. Now, the next() method will produce a Record every time it is called. The method will return "null" if there is no more data in the map. Implement the @PreDestroy annotated method, responsible for releasing all resources used by the Source. The method needs to shut the Hazelcast client instance down to release any connection between the component and the Hazelcast cluster. The Hazelcast Source is completed. The next section shows how to write a simple unit test to check that it works properly. TCK provides a set of APIs and tools that makes the testing straightforward. The test of the Hazelcast Source consists in creating an embedded Hazelcast instance with only one member and initializing it with some data, and then in creating a test Job to read the data from it using the implemented Source. Add the required Maven test dependencies to the project. Initialize a Hazelcast test instance and create a map with some test data. To do that, create the HazelcastSourceTest.java test class in the src/test/java folder. Create the folder if it does not exist. The above example creates a Hazelcast instance for the test and creates the MY-distributed-MAP map. The getMap creates the map if it does not already exist. Some keys and values uses in the test are added. Then, a simple test checks that the data is correctly initialized. Finally, the Hazelcast test instance is shut down. Run the test and check in the logs that a Hazelcast cluster of one member has been created and that the test has passed. To be able to test components, TCK provides the @WithComponents annotation which enables component testing. Add this annotation to the test. The annotation takes the component Java package as a value parameter. Create the test Job that configures the Hazelcast instance and link it to an output that collects the data produced by the Source. Execute the unit test and check that it passes, meaning that the Source is reading the data correctly from Hazelcast. The Source is now completed and tested. The next section shows how to implement the Partition Mapper for the Source. In this case, the Partition Mapper will split the work (data reading) between the available cluster members to distribute the workload. The Partition Mapper calculates the number of Sources that can be created and executed in parallel on the available workers of a distributed system. For Hazelcast, it corresponds to the cluster member count. To fully illustrate this concept, this section also shows how to enhance the test environment to add more Hazelcast cluster members and initialize it with more data. Instantiate more Hazelcast instances, as every Hazelcast instance corresponds to one member in a cluster. In the test, it is reflected as follows: The above code sample creates two Hazelcast instances, leading to the creation of two Hazelcast members. Having a cluster of two members (nodes) will allow to distribute the data. The above code also adds more data to the test map and updates the shutdown method and the test. Run the test on the multi-nodes cluster. The Source is a simple implementation that does not distribute the workload and reads the data in a classic way, without distributing the read action to different cluster members. Start implementing the Partition Mapper class by creating a HazelcastPartitionMapper.java class file. When coupling a Partition Mapper with a Source, the Partition Mapper becomes responsible for injecting parameters and creating source instances. This way, all the attribute initialization part moves from the Source to the Partition Mapper class. The configuration also sets an instance name to make it easy to find the client instance in the logs or while debugging. The Partition Mapper class is composed of the following: constructor: Handles configuration and service injections Assessor: This annotation indicates that the method is responsible for assessing the dataset size. The underlying runner uses the estimated dataset size to compute the optimal bundle size to distribute the workload efficiently. Split: This annotation indicates that the method is responsible for creating Partition Mapper instances based on the bundle size requested by the underlying runner and the size of the dataset. It creates as much partitions as possible to parallelize and distribute the workload efficiently on the available workers (known as members in the Hazelcast case). Emitter: This annotation indicates that the method is responsible for creating the Source instance with an adapted configuration allowing to handle the amount of records it will produce and the required services. I adapts the configuration to let the Source read only the requested bundle of data. The Assessor method computes the memory size of every member of the cluster. Implementing it requires submitting a calculation task to the members through a serializable task that is aware of the Hazelcast instance. Create the serializable task. The purpose of this class is to submit any task to the Hazelcast cluster. Use the created task to estimate the dataset size in the Assessor method. The Assessor method calculates the memory size that the map occupies for all members. In Hazelcast, distributing a task to all members can be achieved using an execution service initialized in the getExecutorService() method. The size of the map is requested on every available member. By summing up the results, the total size of the map in the distributed cluster is computed. The Split method calculates the heap size of the map on every member of the cluster. Then, it calculates how many members a source can handle. If a member contains less data than the requested bundle size, the method tries to combine it with another member. That combination can only happen if the combined data size is still less or equal to the requested bundle size. The following code illustrates the logic described above. The next step consists in adapting the source to take the Split into account. The following sample shows how to adapt the Source to the Split carried out previously. The next method reads the data from the members received from the Partition Mapper. A Big Data runner like Spark will get multiple Source instances. Every source instance will be responsible for reading data from a specific set of members already calculated by the Partition Mapper. The data is fetched only when the next method is called. This logic allows to stream the data from members without loading it all into the memory. Implement the method annotated with @Emitter in the HazelcastPartitionMapper class. The createSource() method creates the source instance and passes the required services and the selected Hazelcast members to the source instance. Run the test and check that it works as intended. The component implementation is now done. It is able to read data and to distribute the workload to available members in a Big Data execution environment. Refactor the component by introducing a service to make some pieces of code reusable and avoid code duplication. Refactor the Hazelcast instance creation into a service. Inject this service to the Partition Mapper to reuse it. Adapt the Source class to reuse the service. Run the test one last time to ensure everything still works as expected. Thank you for following this tutorial. Use the logic and approach presented here to create any input component for any system.

A Processor is a component that converts incoming data to a different model. A processor must have a method decorated with @ElementListener taking an incoming data and returning the processed data: Processors must be Serializable because they are distributed components. If you just need to access data on a map-based ruleset, you can use Record or JsonObject as parameter type. From there, Talend Component Kit wraps the data to allow you to access it as a map. The parameter type is not enforced. This means that if you know you will get a SuperCustomDto, then you can use it as parameter type. But for generic components that are reusable in any chain, it is highly encouraged to use Record until you have an evaluation language-based processor that has its own way to access components. For example: A processor also supports @BeforeGroup and @AfterGroup methods, which must not have any parameter and return void values. Any other result would be ignored. These methods are used by the runtime to mark a chunk of the data in a way which is estimated good for the execution flow size. Because the size is estimated, the size of a group can vary. It is even possible to have groups of size 1. It is recommended to batch records, for performance reasons: You can optimize the data batch processing by using the maxBatchSize parameter. This parameter is automatically implemented on the component when it is deployed to a Talend application. Only the logic needs to be implemented. You can however customize its value setting in your LocalConfiguration the property _maxBatchSize.value - for the family - or ${component simple class name}._maxBatchSize.value - for a particular component, otherwise its default will be 1000. If you replace value by active, you can also configure if this feature is enabled or not. This is useful when you don’t want to use it at all. Learn how to implement chunking/bulking in this document. In some cases, you may need to split the output of a processor in two or more connections. A common example is to have "main" and "reject" output connections where part of the incoming data are passed to a specific bucket and processed later. Talend Component Kit supports two types of output connections: Flow and Reject. Flow is the main and standard output connection. The Reject connection handles records rejected during the processing. A component can only have one reject connection, if any. Its name must be REJECT to be processed correctly in Talend applications. You can also define the different output connections of your component in the Starter. To define an output connection, you can use @Output as replacement of the returned value in the @ElementListener: Alternatively, you can pass a string that represents the new branch: Having multiple inputs is similar to having multiple outputs, except that an OutputEmitter wrapper is not needed: @Input takes the input name as parameter. If no name is set, it defaults to the "main (default)" input branch. It is recommended to use the default branch when possible and to avoid naming branches according to the component semantic. Batch processing refers to the way execution environments process batches of data handled by a component using a grouping mechanism. By default, the execution environment of a component automatically decides how to process groups of records and estimates an optimal group size depending on the system capacity. With this default behavior, the size of each group could sometimes be optimized for the system to handle the load more effectively or to match business requirements. For example, real-time or near real-time processing needs often imply processing smaller batches of data, but more often. On the other hand, a one-time processing without business contraints is more effectively handled with a batch size based on the system capacity. Final users of a component developed with the Talend Component Kit that integrates the batch processing logic described in this document can override this automatic size. To do that, a maxBatchSize option is available in the component settings and allows to set the maximum size of each group of data to process. A component processes batch data as follows: Case 1 - No maxBatchSize is specified in the component configuration. The execution environment estimates a group size of 4. Records are processed by groups of 4. Case 2 - The runtime estimates a group size of 4 but a maxBatchSize of 3 is specified in the component configuration. The system adapts the group size to 3. Records are processed by groups of 3. Batch processing relies on the sequence of three methods: @BeforeGroup, @ElementListener, @AfterGroup, that you can customize to your needs as a component Developer. The group size automatic estimation logic is automatically implemented when a component is deployed to a Talend application. Each group is processed as follows until there is no record left: The @BeforeGroup method resets a record buffer at the beginning of each group. The records of the group are assessed one by one and placed in the buffer as follows: The @ElementListener method tests if the buffer size is greater or equal to the defined maxBatchSize. If it is, the records are processed. If not, then the current record is buffered. The previous step happens for all records of the group. Then the @AfterGroup method tests if the buffer is empty. You can define the following logic in the processor configuration: You can also use the condensed syntax for this kind of processor: When writing tests for components, you can force the maxBatchSize parameter value by setting it with the following syntax: .$maxBatchSize=10. You can learn more about processors in this document. Defining a processor/output logic General component execution logic Implementing bulk processing Best practices For the case of output components (not emitting any data) using bulking you can pass the list of records to the after group method:

Processors and output components are the components in charge of reading, processing and transforming data in a Talend job, as well as passing it to its required destination. Before implementing the component logic and defining its layout and configurable fields, make sure you have specified its basic metadata, as detailed in this document. A Processor is a component that converts incoming data to a different model. A processor must have a method decorated with @ElementListener taking an incoming data and returning the processed data: Processors must be Serializable because they are distributed components. If you just need to access data on a map-based ruleset, you can use Record or JsonObject as parameter type. From there, Talend Component Kit wraps the data to allow you to access it as a map. The parameter type is not enforced. This means that if you know you will get a SuperCustomDto, then you can use it as parameter type. But for generic components that are reusable in any chain, it is highly encouraged to use Record until you have an evaluation language-based processor that has its own way to access components. For example: A processor also supports @BeforeGroup and @AfterGroup methods, which must not have any parameter and return void values. Any other result would be ignored. These methods are used by the runtime to mark a chunk of the data in a way which is estimated good for the execution flow size. Because the size is estimated, the size of a group can vary. It is even possible to have groups of size 1. It is recommended to batch records, for performance reasons: You can optimize the data batch processing by using the maxBatchSize parameter. This parameter is automatically implemented on the component when it is deployed to a Talend application. Only the logic needs to be implemented. You can however customize its value setting in your LocalConfiguration the property _maxBatchSize.value - for the family - or ${component simple class name}._maxBatchSize.value - for a particular component, otherwise its default will be 1000. If you replace value by active, you can also configure if this feature is enabled or not. This is useful when you don’t want to use it at all. Learn how to implement chunking/bulking in this document. In some cases, you may need to split the output of a processor in two or more connections. A common example is to have "main" and "reject" output connections where part of the incoming data are passed to a specific bucket and processed later. Talend Component Kit supports two types of output connections: Flow and Reject. Flow is the main and standard output connection. The Reject connection handles records rejected during the processing. A component can only have one reject connection, if any. Its name must be REJECT to be processed correctly in Talend applications. You can also define the different output connections of your component in the Starter. To define an output connection, you can use @Output as replacement of the returned value in the @ElementListener: Alternatively, you can pass a string that represents the new branch: Having multiple inputs is similar to having multiple outputs, except that an OutputEmitter wrapper is not needed: @Input takes the input name as parameter. If no name is set, it defaults to the "main (default)" input branch. It is recommended to use the default branch when possible and to avoid naming branches according to the component semantic. Batch processing refers to the way execution environments process batches of data handled by a component using a grouping mechanism. By default, the execution environment of a component automatically decides how to process groups of records and estimates an optimal group size depending on the system capacity. With this default behavior, the size of each group could sometimes be optimized for the system to handle the load more effectively or to match business requirements. For example, real-time or near real-time processing needs often imply processing smaller batches of data, but more often. On the other hand, a one-time processing without business contraints is more effectively handled with a batch size based on the system capacity. Final users of a component developed with the Talend Component Kit that integrates the batch processing logic described in this document can override this automatic size. To do that, a maxBatchSize option is available in the component settings and allows to set the maximum size of each group of data to process. A component processes batch data as follows: Case 1 - No maxBatchSize is specified in the component configuration. The execution environment estimates a group size of 4. Records are processed by groups of 4. Case 2 - The runtime estimates a group size of 4 but a maxBatchSize of 3 is specified in the component configuration. The system adapts the group size to 3. Records are processed by groups of 3. Batch processing relies on the sequence of three methods: @BeforeGroup, @ElementListener, @AfterGroup, that you can customize to your needs as a component Developer. The group size automatic estimation logic is automatically implemented when a component is deployed to a Talend application. Each group is processed as follows until there is no record left: The @BeforeGroup method resets a record buffer at the beginning of each group. The records of the group are assessed one by one and placed in the buffer as follows: The @ElementListener method tests if the buffer size is greater or equal to the defined maxBatchSize. If it is, the records are processed. If not, then the current record is buffered. The previous step happens for all records of the group. Then the @AfterGroup method tests if the buffer is empty. You can define the following logic in the processor configuration: You can also use the condensed syntax for this kind of processor: When writing tests for components, you can force the maxBatchSize parameter value by setting it with the following syntax: .$maxBatchSize=10. You can learn more about processors in this document. Defining a processor/output logic General component execution logic Implementing bulk processing Best practices For the case of output components (not emitting any data) using bulking you can pass the list of records to the after group method: An Output is a Processor that does not return any data. Conceptually, an output is a data listener. It matches the concept of processor. Being the last component of the execution chain or returning no data makes your processor an output component: Currently, Talend Component Kit does not allow you to define a Combiner. A combiner is the symmetric part of a partition mapper. It allows to aggregate results in a single partition.

OSS addict I'm involved in several ASF projects (http://home.apache.org/committer-index.html#rmannibucau) and Yupiik (https://www.yupiik.io/projects.html). Blog: https://rmannibucau.metawerx.net Software engineer @Talend. Components team member. Blog: undx.github.io Technical Writer Frontend Architect QA Automation @ Talend Nantes R&D Senior principal product security engineer at Qlik, security contributor at @apache. Blog: http://coheigea.blogspot.com/ Frontend Architect. This is my Talend account. You can check out @jsomsanith for my personal account Blog: timeline.antoinenicolas.com Committer and PMC member of Apache Beam and Apache Avro. Free education and Open Source enthusiast. distributed Systems practitioner (victim?) Blog: https://ismaelmejia.com/ Templar I'm a serial tooler. Blog: https://www.linkedin.com/in/zoltantakacsdev/ Focused on Big Data. Open Source contributor. I'm an Apache Beam and Flink committer and Beam PMC member. I'm also an Apache Software Foundation member Blog: https://echauchot.blogspot.com/ Principal Cloud Software Architect Born in 2008 I joined Talend with my master in 2022, I am now the official ci-runner for the connectors team Product Manager, Information Architect Principal Security Engineer Blog: www.talend.com 573PH4N3 D3M15 K3RM480N Drums, Java, Rock'n'Roll Senior Software engineer at @komodohealth. Ph.D. in large-scale storage systems. Previous competitive programmer.

A Partition Mapper (PartitionMapper) is a component able to split itself to make the execution more efficient. This concept is borrowed from big data and useful in this context only (BEAM executions). The idea is to divide the work before executing it in order to reduce the overall execution time. The process is the following: The size of the data you work on is estimated. This part can be heuristic and not very precise. From that size, the execution engine (runner for Beam) requests the mapper to split itself in N mappers with a subset of the overall work. The leaf (final) mapper is used as a Producer (actual reader) factory. This kind of component must be Serializable to be distributable. A partition mapper requires three methods marked with specific annotations: @Assessor for the evaluating method @Split for the dividing method @Emitter for the Producer factory The Assessor method returns the estimated size of the data related to the component (depending its configuration). It must return a Number and must not take any parameter. For example: The Split method returns a collection of partition mappers and can take optionally a @PartitionSize long value as parameter, which is the requested size of the dataset per sub partition mapper. For example: The Emitter method must not have any parameter and must return a producer. It uses the partition mapper configuration to instantiate and configure the producer. For example:

Input components are the components generally placed at the beginning of a Talend job. They are in charge of retrieving the data that will later be processed in the job. An input component is primarily made of three distinct logics: The execution logic of the component itself, defined through a partition mapper. The configurable part of the component, defined through the mapper configuration. The source logic defined through a producer. Before implementing the component logic and defining its layout and configurable fields, make sure you have specified its basic metadata, as detailed in this document. A Partition Mapper (PartitionMapper) is a component able to split itself to make the execution more efficient. This concept is borrowed from big data and useful in this context only (BEAM executions). The idea is to divide the work before executing it in order to reduce the overall execution time. The process is the following: The size of the data you work on is estimated. This part can be heuristic and not very precise. From that size, the execution engine (runner for Beam) requests the mapper to split itself in N mappers with a subset of the overall work. The leaf (final) mapper is used as a Producer (actual reader) factory. This kind of component must be Serializable to be distributable. A partition mapper requires three methods marked with specific annotations: @Assessor for the evaluating method @Split for the dividing method @Emitter for the Producer factory The Assessor method returns the estimated size of the data related to the component (depending its configuration). It must return a Number and must not take any parameter. For example: The Split method returns a collection of partition mappers and can take optionally a @PartitionSize long value as parameter, which is the requested size of the dataset per sub partition mapper. For example: The Emitter method must not have any parameter and must return a producer. It uses the partition mapper configuration to instantiate and configure the producer. For example: The Producer defines the source logic of an input component. It handles the interaction with a physical source and produces input data for the processing flow. A producer must have a @Producer method without any parameter. It is triggered by the @Emitter method of the partition mapper and can return any data. It is defined in the Source.java file:

Each type of component has its own execution logic. The same basic logic is applied to all components of the same type, and is then extended to implement each component specificities. The project generated from the starter already contains the basic logic for each component. Talend Component Kit framework relies on several primitive components. All components can use @PostConstruct and @PreDestroy annotations to initialize or release some underlying resource at the beginning and the end of a processing. In distributed environments, class constructor are called on cluster manager nodes. Methods annotated with @PostConstruct and @PreDestroy are called on worker nodes. Thus, partition plan computation and pipeline tasks are performed on different nodes. All the methods managed by the framework must be public. Private methods are ignored. The framework is designed to be as declarative as possible but also to stay extensible by not using fixed interfaces or method signatures. This allows to incrementally add new features of the underlying implementations.

From the version 7.0 of Talend Studio, Talend Component Kit becomes the recommended framework to use to develop components. This framework is being introduced to ensure that newly developed components can be deployed and executed both in on-premise/local and cloud/big data environments. From that new approach comes the need to provide a complete yet unique and compatible way of developing components. With the Component Kit, custom components are entirely implemented in Java. To help you get started with a new custom component development project, a Starter is available. Using it, you will be able to generate the skeleton of your project. By importing this skeleton in a development tool, you can then implement the components layout and execution logic in Java. With the previous Javajet framework, metadata, widgets and configurable parts of a custom component were specified in XML. With the Component Kit, they are now defined in the Configuration (for example, LoggerProcessorConfiguration) Java class of your development project. Note that most of this configuration is transparent if you specified the Configuration Model of your components right before generating the project from the Starter. Any undocumented feature or option is considered not supported by the Component Kit framework. You can find examples of output in Studio or Cloud environments in the Gallery. Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit or Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Javajet Component Kit Return variables availability (1.51+) deprecates After Variables. Javajet AVAILABILITY can be : AFTER : set after component finished. FLOW : changed on row level. Component Kit Return variables can be nested as below: Javajet Component Kit or Previously, the execution of a custom component was described through several Javajet files: _begin.javajet, containing the code required to initialize the component. _main.javajet, containing the code required to process each line of the incoming data. _end.javajet, containing the code required to end the processing and go to the following step of the execution. With the Component Kit, the entire execution flow of a component is described through its main Java class (for example, LoggerProcessor) and through services for reusable parts. Each type of component has its own execution logic. The same basic logic is applied to all components of the same type, and is then extended to implement each component specificities. The project generated from the starter already contains the basic logic for each component. Talend Component Kit framework relies on several primitive components. All components can use @PostConstruct and @PreDestroy annotations to initialize or release some underlying resource at the beginning and the end of a processing. In distributed environments, class constructor are called on cluster manager nodes. Methods annotated with @PostConstruct and @PreDestroy are called on worker nodes. Thus, partition plan computation and pipeline tasks are performed on different nodes. All the methods managed by the framework must be public. Private methods are ignored. The framework is designed to be as declarative as possible but also to stay extensible by not using fixed interfaces or method signatures. This allows to incrementally add new features of the underlying implementations. To ensure that the Cloud-compatible approach of the Component Kit framework is respected, some changes were introduced on the implementation side, including: The File mode is no longer supported. You can still work with URIs and remote storage systems to use files. The file collection must be handled at the component implementation level. The input and output connections between two components can only be of the Flow or Reject types. Other types of connections are not supported. Every Output component must have a corresponding Input component and use a dataset. All datasets must use a datastore. To get started with the Component Kit framework, you can go through the following documents: Learn the basics about Talend Component Kit Create and deploy your first Component Kit component Learn about the Starter Start implementing components Integrate a component to Talend Studio Check some examples of components built with Talend Component Kit

This tutorial is the continuation of Talend Input component for Hazelcast tutorial. We will not walk through the project creation again, So please start from there before taking this one. This tutorial shows how to create a complete working output component for Hazelcast As seen before, in Hazelcast there is multiple data source type. You can find queues, topics, cache, maps… In this tutorials we will stick with the Map dataset and all what we will see here is applicable to the other types. Let’s assume that our Hazelcast output component will be responsible of inserting data into a distributed Map. For that, we will need to know which attribute from the incoming data is to be used as a key in the map. The value will be the hole record encoded into a json format. Bu that in mind, we can design our output configuration as: the same Datastore and Dataset from the input component and an additional configuration that will define the key attribute. Let’s create our Output configuration class. Let’s add the i18n properties of our configuration into the Messages.properties file The skeleton of the output component looks as follows: @Version annotation indicates the version of the component. It is used to migrate the component configuration if needed. @Icon annotation indicates the icon of the component. Here, the icon is a custom icon that needs to be bundled in the component JAR under resources/icons. @Processor annotation indicates that this class is the processor (output) and defines the name of the component. constructor of the processor is responsible for injecting the component configuration and services. Configuration parameters are annotated with @Option. The other parameters are considered as services and are injected by the component framework. Services can be local (class annotated with @Service) or provided by the component framework. The method annotated with @PostConstruct is executed once by instance and can be used for initialization. The method annotated with @PreDestroy is used to clean resources at the end of the execution of the output. Data is passed to the method annotated with @ElementListener. That method is responsible for handling the data output. You can define all the related logic in this method. If you need to bulk write the updates accordingly to groups, see Processors and batch processing. Now, we will need to add the display name of the Output to the i18n resources file Messages.properties Let’s implement all of those methods We will create the outpu contructor to inject the component configuration and some additional local and built in services. Built in services are services provided by TCK. Here we find: configuration is the component configuration class hazelcastService is the service that we have implemented in the input component tutorial. it will be responsible of creating a hazelcast client instance. jsonb is a built in service provided by tck to handle json object serialization and deserialization. We will use it to convert the incoming record to json format before inseting them into the map. Nothing to do in the post construct method. but we could for example initialize a hazle cast instance there. but we will do it in a lazy way on the first call in the @ElementListener method Shut down the Hazelcast client instance and thus free the Hazelcast map reference. We get the key attribute from the incoming record and then convert the hole record to a json string. Then we insert the key/value into the hazelcast map. Let’s create a unit test for our output component. The idea will be to create a job that will insert the data using this output implementation. So, let’s create out test class. Here we start by creating a hazelcast test instance, and we initialize the map. we also shutdown the instance after all the test are executed. Now let’s create our output test. Here we start preparing the emitter test component provided bt TCK that we use in our test job to generate random data for our output. Then, we use the output component to fill the hazelcast map. By the end we test that the map contains the exact amount of data inserted by the job. Run the test and check that it’s working. Congratulation you just finished your output component.

The framework provides built-in services that you can inject by type in components and actions. Type Description org.talend.sdk.component.api.service.cache.LocalCache Provides a small abstraction to cache data that does not need to be recomputed very often. Commonly used by actions for UI interactions. org.talend.sdk.component.api.service.dependency.Resolver Allows to resolve a dependency from its Maven coordinates. It can either try to resolve a local file or (better) creates for you a preinitialized classloader. javax.json.bind.Jsonb A JSON-B instance. If your model is static and you don’t want to handle the serialization manually using JSON-P, you can inject that instance. javax.json.spi.JsonProvider A JSON-P instance. Prefer other JSON-P instances if you don’t exactly know why you use this one. javax.json.JsonBuilderFactory A JSON-P instance. It is recommended to use this one instead of a custom one to optimize memory usage and speed. javax.json.JsonWriterFactory A JSON-P instance. It is recommended to use this one instead of a custom one to optimize memory usage and speed. javax.json.JsonReaderFactory A JSON-P instance. It is recommended to use this one instead of a custom one to optimize memory usage and speed. javax.json.stream.JsonParserFactory A JSON-P instance. It is recommended to use this one instead of a custom one to optimize memory usage and speed. javax.json.stream.JsonGeneratorFactory A JSON-P instance. It is recommended to use this one instead of a custom one to optimize memory usage and speed. org.talend.sdk.component.api.service.dependency.Resolver Allows to resolve files from Maven coordinates (like dependencies.txt for component). Note that it assumes that the files are available in the component Maven repository. org.talend.sdk.component.api.service.injector.Injector Utility to inject services in fields marked with @Service. org.talend.sdk.component.api.service.factory.ObjectFactory Allows to instantiate an object from its class name and properties. org.talend.sdk.component.api.service.record.RecordBuilderFactory Allows to instantiate a record. org.talend.sdk.component.api.service.record.RecordPointerFactory Allows to instantiate a RecordPointer which enables to extract a data from a Record based on jsonpointer specification. org.talend.sdk.component.api.service.record.RecordService Some utilities to create records from another one. It is typically what is used when you want to add an entry in a record and passthrough the other ones. It also provides a nice RecordVisitor API for advanced cases. org.talend.sdk.component.api.service.configuration.LocalConfiguration Represents the local configuration that can be used during the design. It is not recommended to use it for the runtime because the local configuration is usually different and the instances are distinct. You can also use the local cache as an interceptor with @Cached Every interface that extends HttpClient and that contains methods annotated with @Request Lets you define an HTTP client in a declarative manner using an annotated interface. See the Using HttpClient for more details. All these injected services are serializable, which is important for big data environments. If you create the instances yourself, you cannot benefit from these features, nor from the memory optimization done by the runtime. Prefer reusing the framework instances over custom ones. The local configuration uses system properties and the environment (replacing dots per underscores) to look up the values. You can also put a TALEND-INF/local-configuration.properties file with default values. This allows to use the local_configuration: syntax in @Ui annotation. Here is an example to read the default value of a property from the configuration: Ensure your key is unique across all components to avoid global overrides on the JVM. In practice, it is strongly recommended to always use the family as a prefix. Also note that you can use @Configuration("prefix") to inject a mapping of the LocalConfiguration in a component. It uses the same rules as for any configuration object. If you prefer to inject you configuration in a service, ensure to wrap it in a Supplier to always have an up to date version. If you want to ignore the local-configuration.properties, you can set the system property: talend.component.configuration.${componentPluginId}.ignoreLocalConfiguration=true. Here a sample @Configuration model: Here is how to use it from a service: And finally, here is how to use it in a component: it is recommended to convert this configuration in a runtime model in components to avoid to transport more than desired during the job distribution. You can access the API reference in the Javadocs. The HttpClient usage is described in this section by using the REST API example below. Assuming that it requires a basic authentication header: GET /api/records/{id} - POST /api/records JSON payload to be created: {"id":"some id", "data":"some data"} To create an HTTP client that is able to consume the REST API above, you need to define an interface that extends HttpClient. The HttpClient interface lets you set the base for the HTTP address that the client will hit. The base is the part of the address that needs to be added to the request path to hit the API. It is now possible, and recommended, to use @Base annotation. Every method annotated with @Request in the interface defines an HTTP request. Every request can have a @Codec parameter that allows to encode or decode the request/response payloads. You can ignore the encoding/decoding for String and Void payloads. The interface should extend HttpClient. In the codec classes (that implement Encoder/Decoder), you can inject any of your service annotated with @Service or @Internationalized into the constructor. Internationalization services can be useful to have internationalized messages for errors handling. The interface can be injected into component classes or services to consume the defined API. By default, /+json are mapped to JSON-P and /+xml to JAX-B if the model has a @XmlRootElement annotation. For advanced cases, you can customize the Connection by directly using @UseConfigurer on the method. It calls your custom instance of Configurer. Note that you can use @ConfigurerOption in the method signature to pass some Configurer configurations. For example, if you have the following Configurer: You can then set it on a method to automatically add the basic header with this kind of API usage: The framework provides in the component-api an OAuth1.Configurer which can be used as an example of configurer implementation. It expects a single OAuth1.Configuration parameter to be passed to the request as a @ConfigurationOption. Here is a sample showing how it can be used: By default, the client loads in memory the payload. In case of big payloads, it can consume too much memory. For these cases, you can get the payload as an InputStream: You can use the Response wrapper, or not.

This part is limited to specific kinds of Beam PTransform: PTransform> for inputs. PTransform, PDone> for outputs. Outputs must use a single (composite or not) DoFn in their apply method. To illustrate the input wrapping, this procedure uses the following input as a starting point (based on existing Beam inputs): To wrap the Read in a framework component, create a transform delegating to that Read with at least a @PartitionMapper annotation and using @Option constructor injections to configure the component. Also make sure to follow the best practices and to specify @Icon and @Version. To illustrate the output wrapping, this procedure uses the following output as a starting point (based on existing Beam outputs): You can wrap this output exactly the same way you wrap an input, but using @Processor instead of: Note that the org.talend.sdk.component.runtime.beam.transform.DelegatingTransform class fully delegates the "expansion" to another transform. Therefore, you can extend it and implement the configuration mapping: In terms of classloading, when you write an I/O, the Beam SDK Java core stack is assumed as provided in Talend Component Kit runtime. This way, you don’t need to include it in the compile scope, it would be ignored anyway. If you need a JSonCoder, you can use the org.talend.sdk.component.runtime.beam.factory.service.PluginCoderFactory service, which gives you access to the JSON-P and JSON-B coders. There is also an Avro coder, which uses the FileContainer. It ensures it is self-contained for IndexedRecord and it does not require—as the default Apache Beam AvroCoder—to set the schema when creating a pipeline. It consumes more space and therefore is slightly slower, but it is fine for DoFn, since it does not rely on serialization in most cases. See org.talend.sdk.component.runtime.beam.transform.avro.IndexedRecordCoder. If your PCollection is made of JsonObject records, and you want to convert them to IndexedRecord, you can use the following PTransforms: converts an IndexedRecord to a JsonObject. converts a JsonObject to an IndexedRecord. converts a JsonObject to an IndexedRecord with AVRO schema inference. There are two main provided coder for Record: it will unwrap the record as an Avro IndexedRecord and serialize it with its schema. This can indeed have a performance impact but, due to the structure of component, it will not impact the runtime performance in general - except with direct runner - because the runners will optimize the pipeline accurately. it will serialize the Avro IndexedRecord as well but it will ensure the schema is in the SchemaRegistry to be able to deserialize it when needed. This implementation is faster but the default implementation of the registry is "in memory" so will only work with a single worker node. You can extend it using Java SPI mecanism to use a custom distributed implementation. Sample input based on Beam Kafka: Because the Beam wrapper does not respect the standard Talend Component Kit programming model ( for example, there is no @Emitter), you need to set the false property in your pom.xml file (or equivalent for Gradle) to skip the component programming model validations of the framework.