io.hekate.rpc.RpcService Maven / Gradle / Ivy
Show all versions of hekate-all Show documentation
/*
* Copyright 2020 The Hekate Project
*
* The Hekate Project licenses this file to you under the Apache License,
* version 2.0 (the "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at:
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* License for the specific language governing permissions and limitations
* under the License.
*/
package io.hekate.rpc;
import io.hekate.cluster.ClusterFilterSupport;
import io.hekate.cluster.ClusterNode;
import io.hekate.cluster.ClusterView;
import io.hekate.core.Hekate;
import io.hekate.core.HekateBootstrap;
import io.hekate.core.service.DefaultServiceFactory;
import io.hekate.core.service.Service;
import io.hekate.messaging.loadbalance.LoadBalancerContext;
import io.hekate.messaging.retry.GenericRetryConfigurer;
import io.hekate.partition.RendezvousHashMapper;
import java.util.Collection;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.CompletableFuture;
/**
* « start hereMain entry point to Remote Procedure Call (RPC) API.
*
*
* Overview
*
* This service provides support for remote calls of Java objects in a cluster of {@link Hekate} nodes. Each such object must declare one
* or more Java interfaces marked with the @{@link Rpc} annotation. Client nodes use this interface to build a proxy object that
* transparently executes all local methods calls on remote nodes.
*
*
*
* - Service Configuration
* - RPC Interface
* - RPC Server
* - RPC Client
* - Routing and Load Balancing
* - Retrying on Error
*
*
*
* Service Configuration
*
* {@link RpcService} can be configured and registered in {@link HekateBootstrap} with the help of {@link RpcServiceFactory} as shown in
* the example below:
*
*
*
*
* - Java
* - Spring XSD
* - Spring bean
*
*
* ${source: rpc/RpcServiceJavadocTest.java#configure}
*
*
* Note: This example requires Spring Framework integration
* (see HekateSpringBootstrap).
* ${source: rpc/rpc-xsd.xml#example}
*
*
* Note: This example requires Spring Framework integration
* (see HekateSpringBootstrap).
* ${source: rpc/rpc-bean.xml#example}
*
*
*
*
* RPC Interface
*
* Every object that is exposed for RPC access must implement at least one @{@link Rpc}-annotated interface.
* Below is the example of a simple RPC interface:
* ${source: rpc/RpcServiceJavadocTest.java#interface}
*
*
* Synchronous vs Asynchronous
*
* RPC interfaces can declare both synchronous and asynchronous methods. Synchronous methods are regular methods that return an object of
* arbitrary Java type. Invocations of such methods are always synchronous and blocks the client thread unless remote invocation is
* completed.
*
*
*
* Asynchronous methods must use a {@link CompletableFuture} as their result type. Such methods are always executed asynchronously by the
* {@link RpcService} without blocking the client thread.
*
*
* Aggregate and Broadcast
*
* RPC service provides support for broadcast/aggregate operations. Each such RPC operations gets submitted to multiple nodes at once and
* all RPC results from those nodes gets aggregated into a single result object.
*
*
* In order to enable such functionality, an RPC method must be annotated with @{@link RpcAggregate}. Such method must be declared
* with one of the following result types:
*
*
* - {@link List}
* - {@link Set}
* - {@link Map}
* - {@link Collection}
* - {@link CompletableFuture}{@code <}{@link List}|{@link Set}|{@link Collection}|{@link Map}{@code >}
*
*
*
* If no results are expected to be returned by an RPC method then such method must be annotated with @{@link RpcBroadcast} and must have a
* {@code void} return type (or {@link CompletableFuture}{@code } for asynchronous calls).
*
*
* Arguments Splitting
*
* RPC service provides support for Map/Reduce style of data processing. This can be achieved by placing @{@link RpcSplit} annotation on
* one of an @{@link RpcAggregate}-annotated method's parameter. Such parameter must be of one of the following types:
*
*
* - {@link List}
* - {@link Set}
* - {@link Map}
* - {@link Collection}
*
*
* If @{@link RpcSplit} annotation is present then the value of that parameter will be split into smaller chunks (sub-collections) based
* on the number of available cluster nodes. All chunks will be evenly distributed among the cluster nodes for parallel processing and once
* processing on all nodes is completed then results of each node will be aggregated the same way as during the regular {@link
* RpcAggregate}
* call.
*
*
* Versioning
*
* RPC interface can define client compatibility rules by using the interface versioning approach. Version can be specified via {@link
* Rpc#version()} and {@link Rpc#minClientVersion()} attributes. {@link Rpc#version()} defines the current version of the RPC interface
* while {@link Rpc#minClientVersion()} defines the minimum version of this interface that can be used on the client side. If client
* detects that its local version is less than the minimum required version of the server then such client will not route any RPC requests
* to such server.
*
*
* Consider the following scenario:
*
*
* - Same jar with an RPC interface {@link Rpc#version()}{@code =1} is deployed both on the client and on the server nodes
* - After some time a new {@link Rpc#version()}={@code 2} of this interface is implemented (possibly with some breaking changes of
* API)
* - New jar file with {@link Rpc#version()}{@code =2} is deployed on a new node
* - At this point, if {@link Rpc#minClientVersion()} is set to {@code 2} then old client with version {@code 1} will know that its API
* is not compatible with the server version {@code 2} and will not try to route any requests to such server.
* - Alternatively, if {@link Rpc#minClientVersion()} is set to {@code 1} (meaning that there were no breaking changes) then old client
* will still be able to route requests to the new server.
*
*
*
* RPC Server
*
* RPC server is a Java class that implements one or more @{@link Rpc}-annotated interfaces. Below is the example of such class:
* ${source: rpc/RpcServiceJavadocTest.java#impl}
*
*
* Server Registration
*
* Each RPC server must be registered within the {@link RpcServiceFactory} in order to be exposed for remote access. Configuration of each
* RPC server is represented by the {@link RpcServerConfig} class. Below is the example of RPC server registration:
* ${source: rpc/RpcServiceJavadocTest.java#server}
*
*
* Tagging
*
* If multiple servers implement the same RPC interface and must be deployed to the same {@link Hekate} node then each such server must
* have an additional qualifier that will help RPC clients to distinguish which exact RPC server they are communicating with.
*
*
*
* Such qualifiers are called "tags" and can be specified for each RPC server via {@link RpcServerConfig#setTags(Set)} method. If one or
* more tags are specified for an RPC server then RPC clients of that server must be constructed via
* {@link RpcService#clientFor(Class, String)} method. The {@code tag} parameter of this method must match with one of the RPC server's
* tags. If client doesn't specify a tag or if tag doesn't match any of the RPC server's tags then such client will not be able to discover
* and communicate with that server.
*
*
*
* RPC Client
*
* The client side of RPC communication is represented by a Java reflections proxy of an @{@link Rpc}-annotated interface. Such proxies can
* be constructed on an RPC client node via {@link RpcService#clientFor(Class)} method. This method returns an instance of {@link
* RpcClientBuilder} interface that provides support for dynamically configuring different aspects of a client proxy object (f.e. timeouts,
* load balancing, retry policies, etc). The client proxy object can be created by calling the {@link RpcClientBuilder#build()} method as
* in the example below:
* ${source: rpc/RpcServiceJavadocTest.java#client}
*
*
*
* Instead of configuring all of the RPC client's options dynamically, it is possible to pre-configure some of those options by
* registering an instance of the {@link RpcClientConfig} class for each such client individually
* (see {@link RpcServiceFactory#setClients(List)} method).
*
*
*
* If such configuration is registered then an instance of {@link RpcClientBuilder}б that is returned from the {@link #clientFor(Class)}
* method, will contain all of the pre-preconfigured options.
*
*
*
* For the complete list of pre-configurable options please see the documentation of {@link RpcClientConfig} class.
*
*
*
* Routing and Load Balancing
*
* Every RPC client proxy uses an instance of {@link RpcLoadBalancer} interface to perform routing of RPC unicast operations. Load balancer
* can be pre-configured via the {@link RpcClientConfig#setLoadBalancer(RpcLoadBalancer)} method or specified dynamically via the {@link
* RpcClientBuilder#withLoadBalancer(RpcLoadBalancer)} method. If load balancer is not specified then the RPC client will fall back to
* the {@link DefaultRpcLoadBalancer}.
*
*
* Note that load balancing does not get applied to RPC broadcast operations (i.e. @{@link RpcAggregate}-annotated methods). Such
* operations are submitted to all nodes within the RPC client's cluster topology.
* Please see the "Cluster topology filtering" section for details of how to control the RPC client's
* cluster topology.
*
*
* Consistent Routing
*
* Applications can provide an affinity key to the {@link RpcLoadBalancer} so that it could perform consistent routing based on some
* application-specific criteria. For example, if the {@link DefaultRpcLoadBalancer} is being used by the RPC client then it will
* utilize the the {@link RendezvousHashMapper} algorithm make sure that all RPC operations with the same affinity key will always be
* routed to the same cluster node (unless the cluster topology doesn't change). Custom implementations of the {@link RpcLoadBalancer}
* interface can use their own algorithms for consistent routing.
*
*
* Affinity key can be specified by annotating one of the RPC method's parameters with @{@link RpcAffinityKey}. If such such annotation is
* present then RPC client will transparently use the value of that parameter as an {@link LoadBalancerContext#affinityKey() affinity key}
* for the {@link RpcLoadBalancer}.
*
*
* Note that there can be only one @{@link RpcAffinityKey}-annotated parameter per RPC method and value of that parameter can't be {@code
* null}. Also it is important to make sure that type of that parameter provides consistent implementation of {@link Object#hashCode()} and
* {@link Object#equals(Object)} methods.
*
*
* Thread Affinity
*
* Besides providing a hint to the {@link RpcLoadBalancer}, specifying an {@link RpcAffinityKey} also instructs the RPC service to
* process all RPC operations of the same affinity key on the same thread. This applies both to the server side and to the client side of
* RPC interactions. Thus, if RPC method returns an instance of {@link CompletableFuture} then such future will be
* {@link CompletableFuture#complete(Object) completed} on a thread that is mapped to the value of {@link RpcAffinityKey}.
*
*
*
* Cluster Topology Filtering
*
* Routing of RPC operations among the cluster nodes is based on the RPC client's cluster view. By default, it includes all of the cluster
* nodes that have a compliant {@link RpcServerConfig RPC server}.
*
*
*
* It is possible to narrow down the list of client-visible nodes via the following methods of the {@link RpcClientBuilder} class:
*
*
* - {@link RpcClientBuilder#forRemotes() forRemotes()}
* - {@link RpcClientBuilder#forRole(String) forRole(String)}
* - {@link RpcClientBuilder#forProperty(String) forProperty(String)}
* - {@link RpcClientBuilder#forNode(ClusterNode) forNode(ClusterNode)}
* - {@link RpcClientBuilder#forOldest() forOldest()}
* - {@link RpcClientBuilder#forYoungest() forYoungest()}
* - ...and other methods from the {@link ClusterFilterSupport} base interface
*
*
*
* If filtering rules are specified for an RPC client then all RPC operations of that client will be distributed only among those nodes
* that do match the filtering criteria.
*
*
*
* Retrying on Error
*
* RPC service provides support for specifying a retry behavior in case of a remote invocation error.
*
*
*
* This can be done by marking a method of an RPC interface with the {@link RpcRetry} annotation. If such annotation is present on a method
* then all failed invocations of that method will be transparently retried based on the annotation attribute values.
*
*
*
* {@link RpcRetry}'s attributes provide support for specifying different parameters of retry behavior (like maximum attempts, delay
* between retries, etc). It is also possible to configure default values of those attributes by registering an instance
* of {@link GenericRetryConfigurer} interface in the RPC client's configuration via the
* {@link RpcClientConfig#setRetryPolicy(GenericRetryConfigurer)} or at the RPC client construction time via the
* {@link RpcClientBuilder#withRetryPolicy(GenericRetryConfigurer)} method.
*
*
* @see RpcServiceFactory
*/
@DefaultServiceFactory(RpcServiceFactory.class)
public interface RpcService extends Service {
/**
* Constructs a new RPC client proxy builder for the specified Java interface and tag.
*
*
* This method returns an instance of {@link RpcClientBuilder} that can be used to configure and {@link RpcClientBuilder#build() build}
* a Java proxy object for the specified RPC interface that will redirect all method invocations to remote cluster nodes.
*
*
*
* Note: Some of the builder's options can be preconfigured via {@link RpcClientConfig} (see its javadocs for more details).
*
*
* @param type RPC interface (must be an @{@link Rpc}-annotated Java interface).
* @param tag Tag (see {@link RpcServerConfig#setTags(Set)}).
* @param RPC interface.
*
* @return Builder.
*/
RpcClientBuilder clientFor(Class type, String tag);
/**
* Constructs a new RPC client proxy builder for the specified Java interface.
*
*
* This method returns an instance of {@link RpcClientBuilder} that can be used to configure and {@link RpcClientBuilder#build() build}
* a Java proxy object for the specified RPC interface that will redirect all method invocations to remote cluster nodes.
*
*
*
* Note: Some of the builder's options can be preconfigured via {@link RpcClientConfig} (see its javadocs for more details).
*
*
* @param type RPC interface (must be an @{@link Rpc}-annotated Java interface).
* @param RPC interface.
*
* @return Builder.
*/
RpcClientBuilder clientFor(Class type);
/**
* Returns the cluster view that includes only those nodes that have a {@link #servers() server} for the specified
* {@link Rpc}-annotated Java interface and tag.
*
* @param type {@link Rpc}-annotated Java interface.
* @param tag Tag (see {@link RpcServerConfig#setTags(Set)}).
*
* @return Cluster view that includes only those node that have a {@link #servers() server} for the specified {@link Rpc}-annotated
* Java interface and tag.
*/
ClusterView clusterOf(Class type, String tag);
/**
* Returns the cluster view that includes only those nodes that have a {@link #servers() server} for the specified
* {@link Rpc}-annotated Java interface.
*
* @param type {@link Rpc}-annotated Java interface.
*
* @return Cluster view that includes only those node that have a {@link #servers() server} for the specified {@link Rpc}-annotated
* Java interface.
*/
ClusterView clusterOf(Class type);
/**
* Returns an immutable list of all RPC servers registered on this node.
*
* @return Immutable list of RPC servers.
*
* @see RpcServiceFactory#setServers(List)
*/
List servers();
/**
* Returns the size of a thread pool for handling NIO-based socket connections
* (see {@link RpcServiceFactory#setNioThreads(int)}).
*
* @return Size of a thread pool for handling NIO-based socket connections.
*/
int nioThreads();
/**
* Returns the worker thread pool size (see {@link RpcServiceFactory#setWorkerThreads(int)}).
*
* @return Worker thread pool size.
*/
int workerThreads();
}