Package io.mats3

Interface MatsFactory

All Superinterfaces:
MatsConfig.StartStoppable
All Known Implementing Classes:
MatsFactory.MatsFactoryWrapper
public interface MatsFactory extends MatsConfig.StartStoppable
The start point for all interaction with Mats - you need to get hold of an instance of this interface to be able to code and configure Mats endpoints, and to perform initiations like sending a message, perform a request and publish a message. Getting hold of a MatsFactory is an implementation specific feature: The JmsMatsFactory is the standard, providing static factory methods, and can be backed by an ActiveMQ- or Artemis-specific JMS ConnectionFactory.

An alternative to Java-based programmatic creation of Mats Endpoints is using Mats SpringConfig integration where you use annotations like @EnableMats, @MatsMapping and @MatsClassMapping. Employing Mats SpringConfig, you'll need to get an instance of MatsFactory into the Spring context. The module is called "mats-spring".

It is worth realizing that all of the methods staged(..., config), single(...) + w/configs; terminator(...) + w/configs are just convenience methods to the main staged(...). These specializations could just as well have resided in a utility class. They are included in the API since these relatively few methods seem to cover most scenarios.

(Exception to this are the subscriptionTerminator(...) + w/configs, as they have different semantics, read the JavaDoc).

Regarding order of the Reply Message, State and Incoming Message, which can be a bit annoying to remember when creating endpoints, and when writing process lambdas: They are always ordered like this: R, S, I, i.e. Reply, State, Incoming. This is to resemble a method signature having the implicit this (or self) reference as the first argument: ReturnType methodName(this, arguments). Thus, if you remember that Mats is created to enable you to write messaging oriented endpoints that look like they are methods, then it might stick! The process lambda thus has args (context, state, incomingMsg), unless it lacks state. Even if it lacks state, the context is always first, and the incoming message is always last.

  • For a MultiStage endpoint, you will have all of reply message type, state type, and incoming message type. The params of the staged(..)-method of MatsFactory is [EndpointId, Reply Class, State Class] - and each of the added stages specifies their own Incoming class and stage process lambda. The stage lambda params will then be: [Context, State, Incoming]. If you employ the lastStage(..), its process return lambda will have the reply type as its return type.
  • For a SingleStage endpoint, you will have incoming message, and reply message - there is no state, since that is an object that traverses between the stages in a multi-stage endpoint, and this endpoint is just a single stage. The params of the single(..)-method of MatsFactory is thus [EndpointId, Reply Class, Incoming Class, process lambda] - you specify the process lambda directly. The process single lambda params are: [Context, Incoming] - the reply type is the the return type of the lambda.
  • In a Terminator, you have an incoming message, and state (which the initiator set), but no Reply, as a terminator doesn't reply. The params of the terminator(..)-method of MatsFactory is thus [EndpointId, State Class, Incoming Class, process lambda] - also here specifying the process lambda directly. The lambda params of the process terminator lambda are: [Context, State, Incoming]

All these type arguments and lambdas and whatnot in the JavaDoc can seem a bit overwhelming at first, but when actually coding Mats Endpoints, it clicks into place and hopefully gives a smooth experience.

Note: It should be possible to use instances of MatsFactory as keys in a HashMap, i.e. their equals and hashCode should remain stable throughout the life of the MatsFactory. Depending on the implementation, instance equality may be sufficient. Note that the MatsFactory.MatsFactoryWrapper is implemented to forward equals and hashCode to wrappee.

  • Field Details

  • Method Details

    • getFactoryConfig

      MatsFactory.FactoryConfig getFactoryConfig()
      Returns:
      the MatsFactory.FactoryConfig on which to configure the factory, e.g. defaults for concurrency.

    • staged

      <R, S> MatsEndpoint<R,S> staged(String endpointId, Class<R> replyClass, Class<S> stateClass)
      Sets up a MatsEndpoint on which you will add stages. The first stage is the one that will receive the incoming (typically request) DTO, while any subsequent stage is invoked when the service that the previous stage sent a request to, replies.

      Unless the state object was sent along with the request or send, the first stage will get a newly constructed empty state instance, while the subsequent stages will get the state instance in the form it was left in the previous stage.

      Parameters:
      endpointId - the identification of this MatsEndpoint, which are the strings that should be provided to the MatsInitiator.MatsInitiate.to(String) or MatsInitiator.MatsInitiate.replyTo(String, Object) methods for this endpoint to get the message. Typical structure is "OrderService.placeOrder" for public endpoints, or "OrderService.private.validateOrder" for private (app-internal) endpoints.
      replyClass - the class that this endpoint shall return.
      stateClass - the class of the State DTO that will be sent along the stages.
      Returns:
      the MatsEndpoint on which to add stages.

    • staged

      <R, S> MatsEndpoint<R,S> staged(String endpointId, Class<R> replyClass, Class<S> stateClass, Consumer<? super MatsEndpoint.EndpointConfig<R,S>> endpointConfigLambda)
      Variation of staged(String, Class, Class) that can be configured "on the fly".

    • single

      <R, I> MatsEndpoint<R,Void> single(String endpointId, Class<R> replyClass, Class<I> incomingClass, MatsEndpoint.ProcessSingleLambda<R,I> processor)
      Sets up a MatsEndpoint that just contains one stage, useful for simple "request the full person data for this/these personId(s)" scenarios. This sole stage is supplied directly, using a specialization of the processor lambda which does not have state (as there is only one stage, there is no other stage to pass state to), but which can return the reply by simply returning it on exit from the lambda.

      Do note that this is just a convenience for the often-used scenario where for example a request will just be looked up in the backing data store, and replied directly, using only one stage, not needing any multi-stage processing.

      Parameters:
      endpointId - the identification of this MatsEndpoint, which are the strings that should be provided to the MatsInitiator.MatsInitiate.to(String) or MatsInitiator.MatsInitiate.replyTo(String, Object) methods for this endpoint to get the message. Typical structure is "OrderService.placeOrder" for public endpoints, or "OrderService.private.validateOrder" for private (app-internal) endpoints.
      replyClass - the class that this endpoint shall return.
      incomingClass - the class of the incoming (typically request) DTO.
      processor - the stage that will be invoked to process the incoming message.
      Returns:
      the MatsEndpoint, but you should not add any stages to it, as the sole stage is already added.

    • single

      <R, I> MatsEndpoint<R,Void> single(String endpointId, Class<R> replyClass, Class<I> incomingClass, Consumer<? super MatsEndpoint.EndpointConfig<R,Void>> endpointConfigLambda, Consumer<? super MatsStage.StageConfig<R,Void,I>> stageConfigLambda, MatsEndpoint.ProcessSingleLambda<R,I> processor)
      Variation of single(String, Class, Class, ProcessSingleLambda) that can be configured "on the fly".

    • terminator

      <S, I> MatsEndpoint<Void,S> terminator(String endpointId, Class<S> stateClass, Class<I> incomingClass, MatsEndpoint.ProcessTerminatorLambda<S,I> processor)
      Sets up a MatsEndpoint that contains a single stage that typically will be the reply-to endpointId for a request initiation, or that can be used to directly send a "fire-and-forget" style invocation to. The sole stage is supplied directly. This type of endpoint cannot reply, as it has no-one to reply to (hence "terminator").

      Do note that this is just a convenience for the often-used scenario where an initiation requests out to some service, and then the reply needs to be handled - and with that the process is finished. That last endpoint which handles the reply is what is referred to as a terminator, in that it has nowhere to reply to. Note that there is nothing hindering you in setting the replyTo endpointId in a request initiation to point to a single-stage or multi-stage endpoint - however, any replies from those endpoints will just go void.

      It is possible to initiate from within a terminator, and one interesting scenario here is to do a publish to a subscriptionTerminator. The idea is then that you do the actual processing via a request, and upon the reply processing in the terminator, you update the app database with the updated information (e.g. "order is processed"), and then you publish an "update caches" message to all the nodes of the app, so that they all have the new state of the order in their caches (or, in a push-based GUI logic, you might want to update all users' view of that order). Note that you (as in the processing node) will also get that published message on your instance of the SubscriptionTerminator.

      It is technically possible reply from within a terminator - but it hard to envision many wise usage scenarios for this, as the stack at a terminator would probably be empty.

      Parameters:
      endpointId - the identification of this MatsEndpoint, which are the strings that should be provided to the MatsInitiator.MatsInitiate.to(String) or MatsInitiator.MatsInitiate.replyTo(String, Object) methods for this endpoint to get the message. Typical structure is "OrderService.placeOrder" for public endpoints (which then is of a "fire-and-forget" style, since a terminator is not meant to reply), or "OrderService.terminator.validateOrder" for private (app-internal) terminators that is targeted by the replyTo(endpointId,..) invocation of an initiation.
      stateClass - the class of the State DTO that will may be provided by the request initiation (or that was sent along with the invocation).
      incomingClass - the class of the incoming (typically reply) DTO.
      processor - the stage that will be invoked to process the incoming message.
      Returns:
      the MatsEndpoint, but you should not add any stages to it, as the sole stage is already added.

    • terminator

      <S, I> MatsEndpoint<Void,S> terminator(String endpointId, Class<S> stateClass, Class<I> incomingClass, Consumer<? super MatsEndpoint.EndpointConfig<Void,S>> endpointConfigLambda, Consumer<? super MatsStage.StageConfig<Void,S,I>> stageConfigLambda, MatsEndpoint.ProcessTerminatorLambda<S,I> processor)
      Variation of terminator(String, Class, Class, ProcessTerminatorLambda) that can be configured "on the fly".

    • subscriptionTerminator

      <S, I> MatsEndpoint<Void,S> subscriptionTerminator(String endpointId, Class<S> stateClass, Class<I> incomingClass, MatsEndpoint.ProcessTerminatorLambda<S,I> processor)
      Special kind of terminator that, in JMS-style terms, subscribes to a topic instead of listening to a queue (i.e. it uses "pub-sub"-style messaging, instead of queue-based). You may only communicate with this type of endpoints by using the MatsInitiator.MatsInitiate.publish(Object) or MatsInitiator.MatsInitiate.replyToSubscription(String, Object) methods.

      Notice that the concurrency of a SubscriptionTerminator is always 1, as it makes no sense to have multiple processors for a subscription - all of the processors would just get an identical copy of each message. If you do need to handle massive amounts of messages, or your work handling is slow, you should instead of handling the work in the processor itself, rather accept the message as fast as possible and send the work out to be processed by some kind of thread pool. (The tool MatsFuturizer does this for its Future-completion handling).

      Parameters:
      endpointId - the identification of this MatsEndpoint, which are the strings that should be provided to the MatsInitiator.MatsInitiate.to(String) or MatsInitiator.MatsInitiate.replyTo(String, Object) methods for this endpoint to get the message.
      stateClass - the class of the State DTO that will may be provided by the request initiation (or that was sent along with the invocation).
      incomingClass - the class of the incoming (typically reply) DTO.
      processor - the stage that will be invoked to process the incoming message.
      Returns:
      the MatsEndpoint, but you should not add any stages to it, as the sole stage is already added.

    • subscriptionTerminator

      <S, I> MatsEndpoint<Void,S> subscriptionTerminator(String endpointId, Class<S> stateClass, Class<I> incomingClass, Consumer<? super MatsEndpoint.EndpointConfig<Void,S>> endpointConfigLambda, Consumer<? super MatsStage.StageConfig<Void,S,I>> stageConfigLambda, MatsEndpoint.ProcessTerminatorLambda<S,I> processor)
      Variation of subscriptionTerminator(String, Class, Class, ProcessTerminatorLambda) that can be configured "on the fly", but notice that the concurrency of a SubscriptionTerminator is always 1.

    • getEndpoints

      List<MatsEndpoint<?,?>> getEndpoints()
      Returns:
      all MatsEndpoints created on this MatsFactory.

    • getEndpoint

      Optional<MatsEndpoint<?,?>> getEndpoint(String endpointId)
      Parameters:
      endpointId - which MatsEndpoint to return, if present.
      Returns:
      the requested MatsEndpoint if present, Optional.empty() if not.

    • getDefaultInitiator

      MatsInitiator getDefaultInitiator()
      Gets or creates the default Initiator (whose name is 'default') from which to initiate new Mats processes, i.e. send a message from "outside of Mats" to a Mats endpoint - NOTICE: This is an active object that can carry backend resources, and it is Thread Safe: You are not supposed to create one instance per message you send!

      IMPORTANT NOTICE!! The MatsInitiator returned from this specific method is special when used within a Mats Stage's context (i.e. the Thread running a Mats Stage): Any initiations performed with it within a Mats Stage will have the same transactional demarcation as an initiation performed using ProcessContext.initiate(..). The idea here is that you thus can create methods that both can be used from the outside of a Mats Stage (thus resulting in an ordinary initiation), but if the same method is invoked within a Mats Stage, the initiation will partake in the same transactional demarcation as the rest of what happens within that Mats Stage. Note that you get a bit strange semantics wrt. the exceptions that MatsInitiator.initiate(..) and MatsInitiator.initiateUnchecked(..) raises: Outside of a Mats Stage, they can throw in the given situations those exceptions describe. However, within a Mats Stage, they will never throw those exceptions, since the actual initiation is not performed until the Mats Stage exits. But, if you want to make such a dual mode method that can be employed both outside and within a Mats Stage, you should thus code for the "Outside" mode, handling those Exceptions as you would in an ordinary initiation.

      If you would really NOT want this - i.e. you for some reason want the initiation performed within the stage to execute even though the Mats Stage fails - you may use getOrCreateInitiator(String), and you can even request the same underlying default initiator by just supplying that method with the argument "default". Please, however, make sure you understand the quite heavy consequence of this: If the Mats Stage throws, the retry-mechanism will kick in, running the Mats Stage one more time, and you will thus potentially send that message many times - one time per retry - since such an initiation with a NON-default MatsInitiator is specifically then not part of the Stage's transactional demarcation.

      Just to ensure that this point comes across: The returned MatsInitiator is Thread Safe, and meant for reuse: You are not supposed to create one instance of MatsInitiator per message you need to send, and then close it afterwards - rather either create one for the entire application, or e.g. for each component: The MatsInitiator can have underlying backend resources attached to it - which also means that it needs to be closed for a clean application shutdown (Note that all MatsInitiators are closed when MatsFactory.stop() is invoked).

      Returns:
      the default MatsInitiator, whose name is 'default', on which messages can be initiated.
      See Also:

    • getOrCreateInitiator

      MatsInitiator getOrCreateInitiator(String name)
      Gets or creates a new Initiator from which to initiate new Mats processes, i.e. send a message from "outside of Mats" to a Mats endpoint - NOTICE: This is an active object that can carry backend resources, and it is Thread Safe: You are not supposed to create one instance per message you send!

      A reason for wanting to make more than one MatsInitiator could be that each initiator might have its own connection to the underlying message broker. You also might want to name the initiators based on what part of the application uses it; The name of the initiator shows up in monitors and tooling.

      IMPORTANT NOTICE!! Please read the JavaDoc of getDefaultInitiator() for important information wrt. transactional demarcation when employing a NON-default MatsInitiator within a Mats Stage.

      Just to ensure that this point comes across: The returned MatsInitiator is Thread Safe, and meant for reuse: You are not supposed to create one instance of MatsInitiator per message you need to send, and then close it afterwards - rather either create one for the entire application, or e.g. for each component: The MatsInitiator can have underlying backend resources attached to it - which also means that it needs to be closed for a clean application shutdown (Note that all MatsInitiators are closed when MatsFactory.stop() is invoked).

      Returns:
      a MatsInitiator, on which messages can be initiated.

    • getInitiators

      List<MatsInitiator> getInitiators()
      Returns:
      all MatsInitiators created on this MatsFactory.

    • start

      void start()
      Starts all endpoints that has been created by this factory, by invoking MatsEndpoint.start() on them.

      Subsequently clears the holdEndpointsUntilFactoryIsStarted()-flag.

      Specified by:
      start in interface MatsConfig.StartStoppable

    • holdEndpointsUntilFactoryIsStarted

      void holdEndpointsUntilFactoryIsStarted()
      If this method is invoked before any endpoint is created, the endpoints will not start even though MatsEndpoint.finishSetup() is invoked on them, but will wait till start() is invoked on the factory. This feature should be employed in most setups where the MATS endpoints might use other services or components whose order of creation and initialization are difficult to fully control, e.g. typically in an IoC container like Spring. Set up the "internal machinery" of the system, including internal services, and only after all this is running, then fire up the endpoints. If this is not done, the endpoints might start consuming messages off of the MQ (there might already be messages waiting when the service boots), and thus invoke services/components that are not yet fully started.

      Note: To implement delayed start for a specific endpoint, simply hold off on invoking MatsEndpoint.finishSetup() until you are OK with it being started and hence starts consuming messages (e.g. when the needed cache service is finished populated); This semantics works both when holdEndpointsUntilFactoryIsStarted() has been invoked or not.

      See Also:

    • waitForReceiving

      boolean waitForReceiving(int timeoutMillis)
      Waits until all endpoints have fully entered the receive-loops, i.e. runs MatsEndpoint.waitForReceiving(int) on all the endpoints started from this factory.

      Note: If there are no Endpoints registered, this will immediately return true!

      Specified by:
      waitForReceiving in interface MatsConfig.StartStoppable
      Parameters:
      timeoutMillis - number of milliseconds before giving up the wait, returning false. 0 is indefinite wait, negative values are not allowed.
      Returns:
      true if the entity started within the timeout, false if it did not start.

    • stop

      boolean stop(int gracefulShutdownMillis)
      Stops all endpoints and initiators, by invoking MatsEndpoint.stop(int) on all the endpoints, and MatsInitiator.close() on all initiators that has been created by this factory. They can be started again individually, or all at once by invoking start()

      Should be invoked at application shutdown.

      Specified by:
      stop in interface MatsConfig.StartStoppable
      Parameters:
      gracefulShutdownMillis - number of milliseconds to let the stage processors wait after having asked for them to shut down, and interrupting them if they have not shut down yet.
      Returns:
      true if all Endpoints (incl. e.g. threads) and resources (e.g. JMS Connections) closed successfully.

    • close

      default void close()
      Convenience method, particularly for Spring's default destroy mechanism: Default implemented to invoke stop(30 seconds).

    • unwrapTo

      default <I> I unwrapTo(Class<I> iface)
      In a situation where you might be given a MatsFactory.MatsFactoryWrapper, but need to find a wrappee that implements a specific interface, this method allows you to just always call matsFactory.unwrapTo(iface) instead of first checking whether it is a proxy and only then cast and unwrap.
      Returns:
      default this if iface.isAssignableFrom(thisClass), otherwise throws IllegalArgumentException (overridden by wrappers).

    • unwrapFully

      default MatsFactory unwrapFully()
      In a situation where you might be given a MatsFactory.MatsFactoryWrapper, but need the actual implementation, this method allows you to just always call matsFactory.unwrapFully() instead of first checking whether it is a proxy and only then cast and unwrap.
      Returns:
      default this for implementations (overridden by wrappers).