sharding-jdbc服務編排，觀後感

編排治理

• 官網文檔 http://shardingjdbc.io/docs/02-guide/orchestration/

概述

• 和dubbo的服務治理很相似啊，不過不帶UI，功能還在完善中

1. sharding-jdbc啓動時，將配置信息分庫分表，主從等信息持久化到zookeeper/etcd節點上
2. 註冊zookeeper/etcd 的監聽zookeeper的watch。（不理解的先了解zk的特色）。
3. 一旦節點改變，將新的節點信息下發到zk client，zk client從新創建配置信息

• 服務治理也就是這是原理，大同小異，配置信息改變，服務不須要重啓，可動態改變適應新配置

具體實現

• 工廠模式一路跟進，沒什麼特別的
• @1 開始註冊znode監聽器

DataSource dataSource = OrchestrationShardingDataSourceFactory.createDataSource(yamlFile);

public static DataSource createDataSource(......) throws SQLException {
        OrchestrationShardingDataSource result = new OrchestrationShardingDataSource(dataSourceMap, shardingRuleConfig, configMap, props, orchestrationConfig);
        result.init();
        return result;
    }

public void init() throws SQLException {
        orchestrationFacade.init(dataSourceMap, shardingRuleConfig, configMap, props, this);
    }

public void init(......) throws SQLException {
          ......
        //持久化配置到znode
        dataSourceService.persistDataSourcesNode();
        //@1 註冊zk監聽器
        listenerManager.initShardingListeners(shardingDataSource);
    }

public void initShardingListeners(final ShardingDataSource shardingDataSource) {
        ......
        dataSourceListenerManager.start(shardingDataSource);
    }

註冊監聽器

• 監聽數據源路徑，發現改變，從新建立數據源

@Override
    public void start(final ShardingDataSource shardingDataSource) {
        regCenter.watch(stateNode.getDataSourcesNodeFullPath(), new EventListener() {
            
            @Override
            public void onChange(final DataChangedEvent event) {
                if (DataChangedEvent.Type.UPDATED == event.getEventType() || DataChangedEvent.Type.DELETED == event.getEventType()) {
                    try {
                        shardingDataSource.renew(dataSourceService.getAvailableShardingRuleConfiguration().build(dataSourceService.getAvailableDataSources()), configService.loadShardingProperties());
                    } catch (final SQLException ex) {
                        throw new ShardingJdbcException(ex);
                    }
                }
            }
        });
    }

• renew具體更新上下文信息等操做

public void renew(final ShardingRule newShardingRule, final Properties newProps) throws SQLException {
        ShardingProperties newShardingProperties = new ShardingProperties(null == newProps ? new Properties() : newProps);
        int originalExecutorSize = shardingProperties.getValue(ShardingPropertiesConstant.EXECUTOR_SIZE);
        int newExecutorSize = newShardingProperties.getValue(ShardingPropertiesConstant.EXECUTOR_SIZE);
        if (originalExecutorSize != newExecutorSize) {
            executorEngine.close();
            executorEngine = new ExecutorEngine(newExecutorSize);
        }
        boolean newShowSQL = newShardingProperties.getValue(ShardingPropertiesConstant.SQL_SHOW);
        shardingProperties = newShardingProperties;
        shardingContext = new ShardingContext(newShardingRule, getDatabaseType(), executorEngine, newShowSQL);
    }

• watch裏面就是zk client發現改變，觸發事件。一些sharding-jdbc根據狀態作不一樣處理。比較直白

@Override
    public void watch(final String key, final EventListener eventListener) {
        final String path = key + "/";
        if (!caches.containsKey(path)) {
            addCacheData(key);
        }
        TreeCache cache = caches.get(path);
        cache.getListenable().addListener(new TreeCacheListener() {
            
            @Override
            public void childEvent(final CuratorFramework client, final TreeCacheEvent event) throws Exception {
                ChildData data = event.getData();
                if (null == data || null == data.getPath()) {
                    return;
                }
                eventListener.onChange(new DataChangedEvent(getEventType(event), data.getPath(), null == data.getData() ? null : new String(data.getData(), "UTF-8")));
            }
            
            private DataChangedEvent.Type getEventType(final TreeCacheEvent event) {
                switch (event.getType()) {
                    case NODE_UPDATED:
                        return DataChangedEvent.Type.UPDATED;
                    case NODE_REMOVED:
                        return DataChangedEvent.Type.DELETED;
                    default:
                        return DataChangedEvent.Type.IGNORED;
                }
            }
        });
    }

小結

• 服務治理思想更深入了，註冊中心，內容改變，全部註冊用戶都自動改變。仍是看代碼理解的透徹啊
• 中間應該可能作更多的事，限流啊，熔斷啊，監控啊什麼的
• 阿里的代碼咋就那麼難以閱讀呢，RocketMQ，Dubbo，再接再礪吧
• sharding-jdbc 路由sql重寫下一步