Voltdb 源码阅读

入口文件：https://github.com/VoltDB/voltdb/blob/master/src/frontend/org/voltdb/VoltDB.java

主要就是Configuration类保存运行参数，主要的如 pathToDeployment（指向deployment.xml），m_startAction(create ，rejoin，，)，端口等

　　　　一个单体voltdb实例 https://github.com/VoltDB/voltdb/blob/master/src/frontend/org/voltdb/RealVoltDB.java
主要voltdb 实例类：https://github.com/VoltDB/voltdb/blob/master/src/frontend/org/voltdb/RealVoltDB.java

m_config.m_computationCoreBindings 是个 ArrayQueue<String>，ThreadFactory从这个里面取得元素创建一个线程到线程池{computation threadpool 个数为核心数/4，用的是ThreadPoolExecutor}，由MoreExecutors封装{任务执行完有回调通知}
http://guojuanjun.blog.51cto.com/277646/948298/
public final Queue<String> m_networkCoreBindings = new ArrayDeque<String>();

对应buildMeshCluster里的HostMessenger的VoltNetworkPool，cpu/4个线程，所以VoltNetworkPool含有VoltNetwork[cpu/4]。

public final Queue<String> m_computationCoreBindings = new ArrayDeque<String>();

对应private ListeningExecutorService m_computationService; 是个ExecuteService的线程池，cpu/4个线程。

public final Queue<String> m_executionCoreBindings = new ArrayDeque<String>();

VoltNetwork是一个实现了Runnable的类;

HostMessager还包含一个SocketJoiner类实例，顾名思义，看下面：

/**
* SocketJoiner runs all the time listening for new nodes in the cluster. Since it is a dedicated thread
* it is able to block while a new node joins without disrupting other activities.
*
* At startup socket joiner will connect to the rest of the cluster in the start method if it fails
* to bind to the leader address.
*
* If it binds to the leader address and becomes the leader the start method returns immediately and runPrimary
* is run from a separate thread. runPrimary will wait for the countdown latch for bootstrapping zk to count down
* before accepting new connections
*/

--host 指定的机器，运行runPrimary，其它机器执行connectToPrimary。第一个连上primary，第二个连上primary，被告知第一个的监听地址，于是再连接第一个；第三个连上 primary后，会被告知第一个和第二个host的监听地址，再连上第一个和第二个。如此进行。 知道最后所有节点都会两两之间存在连接。

AgreementSite代表一个zk node，zk 的一个非持久的内置实现。