这几个kill、rebanlance、activate、deactivate方法放在一起分析,是因为他们都依赖于nimbus定义的通用的 状态转移方法:
transitionName********************************************这里只具体分析kill方法*********************************
一:初始化LocalCluster同拓扑提交一样,详见大神http://www.maoxiangyi.cn/index.php/archives/734;
二:LocalCluster完成构造之后,即可使用killTopology/ killTopologyWithOpts杀死拓扑。
方法的实现都会调用nimbus的对应方法。追踪nimbus的实现类,就能看到killTopology/ killTopologyWithOpts这两个方法,真正的实现类就是killTopologyWithOpts
@Override public void killTopology(String name) throws NotAliveException, TException { killTopologyWithOpts(name, new KillOptions()); }
跟踪killTopologyWithOpts方法,方法定义如下:
public void killTopologyWithOpts(String topologyName, KillOptions options)可以看到需要两个参数,分别是拓扑的名称和kill的操作。
@Override public void killTopologyWithOpts(String topologyName, KillOptions options) throws NotAliveException, TException { try { checkTopologyActive(data, topologyName, true); Integer wait_amt = null; //查看是否设置了杀死拓扑的等待时间,如果设置了就等待 if (options.is_set_wait_secs()) { wait_amt = options.get_wait_secs(); } // NimbusUtils.transitionName(data, topologyName, true, StatusType.kill, wait_amt); } catch (NotAliveException e) { String errMsg = "KillTopology Error, no this topology " + topologyName; LOG.error(errMsg, e); throw new NotAliveException(errMsg); } catch (Exception e) { String errMsg = "Failed to kill topology " + topologyName; LOG.error(errMsg, e); throw new TException(errMsg); } }
流程:1,先查看拓扑是否Active,
public void checkTopologyActive(NimbusData nimbus, String topologyName, boolean bActive) throws Exception { if (isTopologyActive(nimbus.getStormClusterState(), topologyName) != bActive) { if (bActive) { throw new NotAliveException(topologyName + " is not alive"); } else { throw new AlreadyAliveException(topologyName + " is already active"); } } }
跟下去
public boolean isTopologyActive(StormClusterState stormClusterState, String topologyName) throws Exception { boolean rtn = false; if (Cluster.get_topology_id(stormClusterState, topologyName) != null) { rtn = true; } return rtn; }
2:查看是否设置了kill操作是否设置了等待时间,如果设置了就等待后再执行
3:transitionName:这是在nimbus定义的通用的 状态转移方法,,他们会根据传入的转移事件做相应的处理。
public static <T> void transitionName(NimbusData data, String topologyName, boolean errorOnNoTransition, StatusType transition_status, T... args) throws Exception { //获得nimbus的信息,通过getStormClusterState查找对应的storm-id(实际上就是将storm-name 转换为storm-id的状态),如果找到了,就调用transition方法。 StormClusterState stormClusterState = data.getStormClusterState(); String topologyId = Cluster.get_topology_id(stormClusterState, topologyName); if (topologyId == null) { throw new NotAliveException(topologyName); } //开始执行kill操作,转变拓扑状态。
transition(data, topologyId, errorOnNoTransition, transition_status, args); }
</pre><p><span style="white-space:pre"></span></p><pre name="code" class="java">public static <T> void transitionName(NimbusData data, String topologyName, boolean errorOnNoTransition, StatusType transition_status, T... args) throws Exception { //通过nimbus的信息,获得storm客户端的状态。 StormClusterState stormClusterState = data.getStormClusterState(); //通过cluster的状态和拓扑的名称获得拓扑的ID。 String topologyId = Cluster.get_topology_id(stormClusterState, topologyName); if (topologyId == null) { throw new NotAliveException(topologyName); } //开始执行kill操作,转变拓扑状态 transition(data, topologyId, errorOnNoTransition, transition_status, args); }
4:transition:这是在nimbus定义的通用的 状态转移方法
transition方法会根据传入的转移事件,获取与当前Topology状态对应的状态转移函数,并且执行该函数取得转换后的新状态。
public static <T> void transition(NimbusData data, String topologyid, boolean errorOnNoTransition, StatusType transition_status, T... args) { try { //转变拓扑的状态 data.getStatusTransition().transition(topologyid, errorOnNoTransition, transition_status, args); } catch (Exception e) { // TODO Auto-generated catch block LOG.error("Failed to do status transition,", e); } }
*********
最后一步追踪trantion有一步topologyLocks还没看太明白%>_<%
后续更新