升级spring-integration-hazelcast后启动报错：CP Subsystem is not enabled

问题背景 #

你之前使用spring-integration-hazelcast:1.0.0.RELEASE搭配hazelcast-all:3.12.4时应用运行正常，尝试升级到hazelcast-all:3.12.6和spring-integration-hazelcast:2.0.0.RELEASE后，应用启动失败，抛出如下异常：

cluster-manager_1 | com.hazelcast.core.HazelcastException: CP Subsystem is not enabled!
cluster-manager_1 | at com.hazelcast.cp.internal.RaftInvocationManager.completeExceptionallyIfCPSubsystemNotAvailable(RaftInvocationManager.java:113)
cluster-manager_1 | at com.hazelcast.cp.internal.RaftInvocationManager.invoke(RaftInvocationManager.java:186)
cluster-manager_1 | at com.hazelcast.cp.internal.RaftService.getGroupInfoForProxy(RaftService.java:752)
cluster-manager_1 | at com.hazelcast.cp.internal.RaftService.createRaftGroupForProxy(RaftService.java:704)
cluster-manager_1 | at com.hazelcast.cp.internal.datastructures.lock.RaftLockService.doCreateProxy(RaftLockService.java:193)
cluster-manager_1 | at com.hazelcast.cp.internal.datastructures.lock.RaftLockService.createProxy(RaftLockService.java:177)
cluster-manager_1 | at com.hazelcast.cp.internal.datastructures.lock.RaftLockService.createProxy(RaftLockService.java:43)
cluster-manager_1 | at com.hazelcast.cp.internal.CPSubsystemImpl.getLock(CPSubsystemImpl.java:76)
cluster-manager_1 | at org.springframework.integration.hazelcast.HazelcastLocalInstanceRegistrar.syncConfigurationMultiMap(HazelcastLocalInstanceRegistrar.java:99)
cluster-manager_1 | at org.springframework.integration.hazelcast.HazelcastLocalInstanceRegistrar.afterSingletonsInstantiated(HazelcastLocalInstanceRegistrar.java:86)
cluster-manager_1 | at org.springframework.beans.factory.support.DefaultListableBeanFactory.preInstantiateSingletons(DefaultListableBeanFactory.java:896)
cluster-manager_1 | at org.springframework.context.support.AbstractApplicationContext.finishBeanFactoryInitialization(AbstractApplicationContext.java:878)
cluster-manager_1 | at org.springframework.context.support.AbstractApplicationContext.refresh(AbstractApplicationContext.java:550)
cluster-manager_1 | at org.springframework.boot.web.servlet.context.ServletWebServerApplicationContext.refresh(ServletWebServerApplicationContext.java:141)
cluster-manager_1 | at org.springframework.boot.SpringApplication.refresh(SpringApplication.java:747)
cluster-manager_1 | at org.springframework.boot.SpringApplication.refreshContext(SpringApplication.java:397)
cluster-manager_1 | at org.springframework.boot.SpringApplication.run(SpringApplication.java:315)
cluster-manager_1 | at com.btnty.citm.clustermanager.springboot.ClusterManagerApplication.main(ClusterManagerApplication.java:49)
cluster-manager_1 | at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
cluster-manager_1 | at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
cluster-manager_1 | at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
cluster-manager_1 | at java.lang.reflect.Method.invoke(Method.java:498)
cluster-manager_1 | at org.springframework.boot.loader.MainMethodRunner.run(MainMethodRunner.java:48)
cluster-manager_1 | at org.springframework.boot.loader.Launcher.launch(Launcher.java:87)
cluster-manager_1 | at org.springframework.boot.loader.Launcher.launch(Launcher.java:51)
cluster-manager_1 | at org.springframework.boot.loader.JarLauncher.main(JarLauncher.java:52)

并且你已经验证：单独升级Hazelcast到3.12.6搭配旧版spring-integration-hazelcast:1.0.0.RELEASE可以正常运行，仅升级spring-integration-hazelcast到2.0.0.RELEASE就会触发该错误。

原因分析 #

这个问题的核心是spring-integration-hazelcast 2.0.0.RELEASE版本的底层实现发生了变化：

• 旧版本（1.0.0.RELEASE）使用的是Hazelcast普通的分布式锁（基于AP模型的ILock）；
• 新版本（2.0.0.RELEASE）则切换为使用Hazelcast CP Subsystem中的Raft Lock（FencedLock），这种锁提供了强一致性的分布式锁能力，但依赖于Hazelcast的CP子系统。

而在Hazelcast 3.x版本中，CP Subsystem默认是未启用的，所以当新版本的spring-integration-hazelcast尝试获取CP锁时，就会抛出CP Subsystem is not enabled!的异常。

解决方案 #

针对这个问题，你有两种可行的处理方式：

1. 启用Hazelcast的CP Subsystem

在你的Hazelcast配置中添加CP子系统的配置即可解决问题。

• 如果是XML配置（hazelcast.xml）：

<hazelcast>
    <!-- 启用CP子系统，生产环境建议设置为3/5等奇数节点数 -->
    <cp-subsystem>
        <cp-member-count>3</cp-member-count>
        <group-size>3</group-size>
    </cp-subsystem>
</hazelcast>

• 如果是Java代码配置：

Config hazelcastConfig = new Config();
CPSubsystemConfig cpSubsystemConfig = hazelcastConfig.getCPSubsystemConfig();
// 设置CP成员数量，单节点测试可设为1，生产环境建议3+
cpSubsystemConfig.setCPMemberCount(3);

注意：单节点测试时可以把cp-member-count设为1，但生产环境必须使用奇数个节点（3、5等），才能保证CP子系统的一致性和容错能力。

2. 回退spring-integration-hazelcast版本（临时方案）

如果暂时不需要强一致性的分布式锁，或者不想改动Hazelcast的配置，可以回退到spring-integration-hazelcast:1.0.0.RELEASE版本，搭配Hazelcast 3.12.6使用——你已经验证过这种组合是可以正常运行的。

原Gradle配置参考（回退版本后）：

implementation ("org.springframework.boot:spring-boot-starter-web:2.2.5.RELEASE")
implementation ("org.springframework.integration:spring-integration-hazelcast:1.0.0.RELEASE")
implementation "com.hazelcast:hazelcast-all:3.12.6"
compile group: 'org.springframework.integration', name: 'spring-integration-core', version: '5.2.1.RELEASE'

内容的提问来源于stack exchange，提问作者daravetu