Flink CDC 3.0 表结构变更的处理流程

Flink CDC 3.0 表结构变更的处理流程

​ 表结构变更主要涉及到三个类SchemaOperatorDataSinkWriterOperatorSink端)和SchemaRegistry(协调器);SchemaOperator接收结构变更消息时会通知sink端和协调器,并等待结构变更操作在协调器执行完毕后在处理后续数据,具体流程参考如下。

前提条件

cdc版本:Flink-cdc 3.0

Flink版本:Flink 1.18

SchemaOperator

Source抓表结构变更事件推送到SchemaOperator时,SchemaOperator会向协调器(也就是SchemaRegistry)发起变更请求;如果是表结构变更,则向Sink发送flushEvent,让其(Sinkflush内存中数据(Sink是经过DataSinkWriterOperator包装),最后阻塞数据流;

SchemaOperator处理表结构变更事件

# SchemaOperator
@Override
public void processElement(StreamRecord<Event> streamRecord) {
    Event event = streamRecord.getValue();
    //如果是schame change事件
    if (event instanceof SchemaChangeEvent) {
        TableId tableId = ((SchemaChangeEvent) event).tableId();
        LOG.info(
                "Table {} received SchemaChangeEvent and start to be blocked.",
                tableId.toString());
        //处理schame change事件
        handleSchemaChangeEvent(tableId, (SchemaChangeEvent) event);
        return;
    }
    output.collect(streamRecord);
}

private void handleSchemaChangeEvent(TableId tableId, SchemaChangeEvent schemaChangeEvent) {
    // The request will need to send a FlushEvent or block until flushing finished
    //向协调节点(SchemaRegistry)发送表结构变更请求,是表结构变更会返回true 如果是建表则返回false
    SchemaChangeResponse response = requestSchemaChange(tableId, schemaChangeEvent);
    if (response.isShouldSendFlushEvent()) {
        LOG.info(
                "Sending the FlushEvent for table {} in subtask {}.",
                tableId,
                getRuntimeContext().getIndexOfThisSubtask());
        //向sink发送 flush事件和schame信息
        output.collect(new StreamRecord<>(new FlushEvent(tableId)));
        output.collect(new StreamRecord<>(schemaChangeEvent));
        // The request will block until flushing finished in each sink writer
        // 这个请求查询协调器,当前schame是否执行完毕,如果没有则阻塞等待,直到协调器完成schame change操作
        requestReleaseUpstream();
    }
}

Sink

Sinkflush掉变更前的数据,并上报给协调器(SchemaRegistry)缓存刷新完成

Sink端处理表结构变更事件,并上报给协调器

# DataSinkWriterOperator 创建sink时会使用DataSinkWriterOperator包装,用于处理FlushEventCreateTableEvent事件
@Override
public void processElement(StreamRecord<Event> element) throws Exception {
    Event event = element.getValue();

    // 处理FlushEvent事件
    if (event instanceof FlushEvent) {
        handleFlushEvent(((FlushEvent) event));
        return;
    }

    // CreateTableEvent marks the table as processed directly
    if (event instanceof CreateTableEvent) {
        processedTableIds.add(((CreateTableEvent) event).tableId());
        this.<OneInputStreamOperator<Event, CommittableMessage<CommT>>>getFlinkWriterOperator()
                .processElement(element);
        return;
    }

    // Check if the table is processed before emitting all other events, because we have to make
    // sure that sink have a view of the full schema before processing any change events,
    // including schema changes.
    ChangeEvent changeEvent = (ChangeEvent) event;
    if (!processedTableIds.contains(changeEvent.tableId())) {
        emitLatestSchema(changeEvent.tableId());
        processedTableIds.add(changeEvent.tableId());
    }
    processedTableIds.add(changeEvent.tableId());
    this.<OneInputStreamOperator<Event, CommittableMessage<CommT>>>getFlinkWriterOperator()
            .processElement(element);
}

# handleFlushEvent 向协调节点(SchemaRegistry)发送`FlushSuccess`请求
private void handleFlushEvent(FlushEvent event) throws Exception {
    copySinkWriter.flush(false);
    schemaEvolutionClient.notifyFlushSuccess(
            getRuntimeContext().getIndexOfThisSubtask(), event.getTableId());
}

协调器

​ 协调节点收到所有Sinkflush完成通知后,然后执行结构变更操作,最后通知完成给等待的requestReleaseUpstream请求。

协调节点处理FlushSuccess请求

public void flushSuccess(TableId tableId, int sinkSubtask) {
    flushedSinkWriters.add(sinkSubtask);
    //所有节点都处理完成
    if (flushedSinkWriters.equals(activeSinkWriters)) {
        LOG.info(
                "All sink subtask have flushed for table {}. Start to apply schema change.",
                tableId.toString());
        PendingSchemaChange waitFlushSuccess = pendingSchemaChanges.get(0);
        //执行表结构变更操作
        applySchemaChange(tableId, waitFlushSuccess.getChangeRequest().getSchemaChangeEvent());
        //通知等待的SchemaOperator,结构变更完成!
        waitFlushSuccess.getResponseFuture().complete(wrap(new ReleaseUpstreamResponse()));
        if (RECEIVED_RELEASE_REQUEST.equals(waitFlushSuccess.getStatus())) {
            startNextSchemaChangeRequest();
        }
    }
}

更多请参考github:参考地址

  • 6
    点赞
  • 10
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值