jstorm源码之RotatingTransactionalState

一、作用
   构建一个Rotationg transaction的state类 用于完成partition的state管理及操作
二、源码分析
package storm.trident.topology.state;

import backtype.storm.utils.Utils;
import org.apache.zookeeper.KeeperException;

import java.util.HashSet;
import java.util.List;
import java.util.SortedMap;
import java.util.TreeMap;

public class RotatingTransactionalState {
    // state intitle接口
    public static interface StateInitializer {
        Object init(long txid, Object lastState);
    }   
    // 事务状态  通过zookeeper来进行管理
    private TransactionalState _state;
    //  子目录
    private String _subdir;
    //  记录transactional id及其内容
    private TreeMap<Long, Object> _curr = new TreeMap<Long, Object>();

    public RotatingTransactionalState(TransactionalState state, String subdir) {
        _state = state;
        _subdir = subdir;
        state.mkdir(subdir);
        sync();
    }

   // 获取最近的transaction state 因为使用的TreeMap 最后一个元素即为最近的transaction state
    public Object getLastState() {
        if(_curr.isEmpty()) return null;
        else return _curr.lastEntry().getValue();
    }
    //  若是transaction state发生改变时需要更新对应的zookeeper对应节点的内容
    public void overrideState(long txid, Object state) {
        _state.setData(txPath(txid), state);
        _curr.put(txid, state);
    }
    //  根据指定的txid删除对应的transaction state内容
    public void removeState(long txid) {
        if(_curr.containsKey(txid)) {
            _curr.remove(txid);
            _state.delete(txPath(txid));
        }
    }
   // 根据执行txId获取本地对应的transaction state
    public Object getState(long txid) {
        return _curr.get(txid);
    }
    // 根据指定的txId 和 对应的初始化器 获取对应的transaction state
    public Object getState(long txid, StateInitializer init) {
        if(!_curr.containsKey(txid)) {
             // 获取小于指定txId的内容
            SortedMap<Long, Object> prevMap = _curr.headMap(txid);
             // 获取大于等于指定的txId的内容
            SortedMap<Long, Object> afterMap = _curr.tailMap(txid);           
            // 判断preMap是否为空 不为空获取最后一个key即为prev state
            Long prev = null;
            if(!prevMap.isEmpty()) prev = prevMap.lastKey();           
            // 
            Object data;
            if(afterMap.isEmpty()) {
                Object prevData;
                if(prev!=null) {
                    prevData = _curr.get(prev);
                } else {
                    prevData = null;
                }
                data = init.init(txid, prevData);
            } else {
                // ??????????
                data = null;
            }
            // 添加到本地
            _curr.put(txid, data);
            // 在zookeeper上创建对应的节点
            _state.setData(txPath(txid), data);
        }
        // 将对应的transaction state内容返回
        return _curr.get(txid);
    }

    public Object getPreviousState(long txid) {
        SortedMap<Long, Object> prevMap = _curr.headMap(txid);
        if(prevMap.isEmpty()) return null;
        else return prevMap.get(prevMap.lastKey());
    }
   // 判断本地cache是否存在
    public boolean hasCache(long txid) {
        return _curr.containsKey(txid);
    }

    /**
     * Returns null if it was created, the value otherwise.
     */
    public Object getStateOrCreate(long txid, StateInitializer init) {
        if(_curr.containsKey(txid)) {
            return _curr.get(txid);
        } else {
            getState(txid, init);
            return null;
        }
    }
   // 删除指定txId对应的node内容 包括两个部分：本地 和 zookeeper
    public void cleanupBefore(long txid) {
        SortedMap<Long, Object> toDelete = _curr.headMap(txid);
        for(long tx: new HashSet<Long>(toDelete.keySet())) {
            _curr.remove(tx);
            try {
                _state.delete(txPath(tx));
            } catch(RuntimeException e) {
                // Ignore NoNodeExists exceptions because when sync() it may populate _curr with stale data since
                // zookeeper reads are eventually consistent.
                if(!Utils.exceptionCauseIsInstanceOf(KeeperException.NoNodeException.class, e)) {
                    throw e;
                }
            }
        }
    }
   // 同步获取指定subdir的所有子节点 并获取对应的内容 同时完成在本地进行保存
    private void sync() {
        List<String> txids = _state.list(_subdir);
        for(String txid_s: txids) {
            Object data = _state.getData(txPath(txid_s));
            _curr.put(Long.parseLong(txid_s), data);
        }
    }
   
    private String txPath(long tx) {
        return txPath("" + tx);
    }

    private String txPath(String tx) {
        return _subdir + "/" + tx;
    }   

}