use of org.I0Itec.zkclient.exception.ZkException in project otter by alibaba.
the class TransformZooKeeperArbitrateEvent method await.
/**
* <pre>
* 算法:
* 1. 检查当前的Permit,阻塞等待其授权(解决Channel的pause状态处理)
* 2. 开始阻塞获取符合条件的processId
* 3. 检查当前的即时Permit状态 (在阻塞获取processId过程会出现一些error信号,process节点会被删除)
* 4. 获取Select传递的EventData数据,添加next node信息后直接返回
* </pre>
*
* @return
*/
public EtlEventData await(Long pipelineId) throws InterruptedException {
Assert.notNull(pipelineId);
PermitMonitor permitMonitor = ArbitrateFactory.getInstance(pipelineId, PermitMonitor.class);
// 阻塞等待授权
permitMonitor.waitForPermit();
TransformStageListener transformStageListener = ArbitrateFactory.getInstance(pipelineId, TransformStageListener.class);
// 符合条件的processId
Long processId = transformStageListener.waitForProcess();
ChannelStatus status = permitMonitor.getChannelPermit();
if (status.isStart()) {
// 即时查询一下当前的状态,状态随时可能会变
// 根据pipelineId+processId构造对应的path
String path = StagePathUtils.getExtractStage(pipelineId, processId);
try {
byte[] data = zookeeper.readData(path);
EtlEventData eventData = JsonUtils.unmarshalFromByte(data, EtlEventData.class);
// 下一个节点信息即为自己
eventData.setNextNid(ArbitrateConfigUtils.getCurrentNid());
// 只有这一条路返回
return eventData;
} catch (ZkNoNodeException e) {
logger.error("pipeline[{}] processId[{}] is invalid , retry again", pipelineId, processId);
// /出现节点不存在,说明出现了error情况,递归调用重新获取一次
return await(pipelineId);
} catch (ZkException e) {
throw new ArbitrateException("transform_await", e.getMessage(), e);
}
} else {
logger.info("pipelineId[{}] transform ignore processId[{}] by status[{}]", new Object[] { pipelineId, processId, status });
// 释放下processId,因为load是等待processId最小值完成Tranform才继续,如果这里不释放,会一直卡死等待
String path = StagePathUtils.getProcess(pipelineId, processId);
zookeeper.delete(path);
// 递归调用
return await(pipelineId);
}
}
use of org.I0Itec.zkclient.exception.ZkException in project otter by alibaba.
the class TransformZooKeeperArbitrateEvent method single.
/**
* <pre>
* 算法:
* 1. 创建对应的transformed节点,标志transform已完成
* </pre>
*
* @param pipelineId 同步流id
*/
public void single(EtlEventData data) {
Assert.notNull(data);
String path = StagePathUtils.getTransformStage(data.getPipelineId(), data.getProcessId());
data.setCurrNid(ArbitrateConfigUtils.getCurrentNid());
// 序列化
byte[] bytes = JsonUtils.marshalToByte(data, SerializerFeature.WriteClassName);
try {
zookeeper.create(path, bytes, CreateMode.PERSISTENT);
} catch (ZkNoNodeException e) {
// process节点不存在,出现了rollback/shutdown操作,直接忽略
logger.warn("pipelineId[{}] transform ignore processId[{}] single by data:{}", new Object[] { data.getPipelineId(), data.getProcessId(), data });
} catch (ZkNodeExistsException e) {
// process节点已存在,出现了ConnectionLoss retry操作
logger.warn("pipelineId[{}] transform ignore processId[{}] single by data:{}", new Object[] { data.getPipelineId(), data.getProcessId(), data });
} catch (ZkException e) {
throw new ArbitrateException("transform_single", e.getMessage(), e);
}
}
use of org.I0Itec.zkclient.exception.ZkException in project otter by alibaba.
the class SelectProcessListener method processChanged.
public void processChanged(List<Long> processIds) {
super.processChanged(processIds);
// add by ljh at 2012-09-13,解决zookeeper ConnectionLoss问题
for (Long processId : processIds) {
if (!replyProcessIds.contains(processId)) {
logger.warn("process is not in order, please check processId:{}", processId);
addReply(processId);
}
}
try {
String path = StagePathUtils.getProcessRoot(getPipelineId());
// 根据并行度创建任务
int size = ArbitrateConfigUtils.getParallelism(getPipelineId()) - processIds.size();
if (size > 0) {
// 创建一个节点
PermitMonitor permit = ArbitrateFactory.getInstance(getPipelineId(), PermitMonitor.class);
if (permit.isPermit() == false) {
// 如果非授权,则不做任何处理
return;
}
String mainStemPath = StagePathUtils.getMainStem(getPipelineId());
byte[] bytes = zookeeper.readData(mainStemPath, true);
if (bytes == null) {
return;
}
MainStemEventData eventData = JsonUtils.unmarshalFromByte(bytes, MainStemEventData.class);
if (eventData.getNid().equals(ArbitrateConfigUtils.getCurrentNid()) == false) {
// 如果非自己设置的mainStem,则不做任何处理
return;
}
synchronized (this) {
// 重新再取一次, dobble-check
List<String> currentProcesses = zookeeper.getChildren(path);
size = ArbitrateConfigUtils.getParallelism(getPipelineId()) - currentProcesses.size();
if (size > 0) {
// 创建一个节点
ProcessNodeEventData nodeData = new ProcessNodeEventData();
// 标记为未使用
nodeData.setStatus(ProcessNodeEventData.Status.UNUSED);
nodeData.setMode(ArbitrateMode.RPC);
nodeData.setNid(ArbitrateConfigUtils.getCurrentNid());
byte[] nodeBytes = JsonUtils.marshalToByte(nodeData);
String processPath = zookeeper.create(path + "/", nodeBytes, CreateMode.PERSISTENT_SEQUENTIAL);
// 创建为顺序的节点
String processNode = StringUtils.substringAfterLast(processPath, "/");
// 添加到当前的process列表
Long processId = StagePathUtils.getProcessId(processNode);
addReply(processId);
}
}
}
} catch (ZkException e) {
// 出现异常后进行一次recovery,读取一下当前最新值,解决出现ConnectionLoss时create成功问题
recovery(getPipelineId());
logger.error("add process error!", e);
}
}
use of org.I0Itec.zkclient.exception.ZkException in project otter by alibaba.
the class TerminZooKeeperArbitrateEvent method await.
/**
* <pre>
* 算法:
* 1. 开始阻塞获取符合条件的processId,获取对应的data数据直接返回
* </pre>
*/
public TerminEventData await(Long pipelineId) throws InterruptedException {
Assert.notNull(pipelineId);
TerminMonitor terminMonitor = ArbitrateFactory.getInstance(pipelineId, TerminMonitor.class);
// 符合条件的processId
Long processId = terminMonitor.waitForProcess();
if (logger.isDebugEnabled()) {
logger.debug("## await pipeline[{}] processId[{}] is termin", pipelineId, processId);
}
// 根据pipelineId+processId构造对应的path
String path = StagePathUtils.getTermin(pipelineId, processId);
try {
byte[] data = zookeeper.readData(path);
return JsonUtils.unmarshalFromByte(data, TerminEventData.class);
} catch (ZkNoNodeException e) {
logger.error("pipeline[{}] processId[{}] is process", pipelineId, processId);
// modify for 2012-09-08, 发生主备切换时,await会进入死循环,针对NoNode后直接从内存队列中移除
terminMonitor.ack(processId);
// 再取下一个节点
return await(pipelineId);
} catch (ZkException e) {
throw new ArbitrateException("Termin_await", e);
}
}
use of org.I0Itec.zkclient.exception.ZkException in project otter by alibaba.
the class TerminZooKeeperArbitrateEvent method ack.
/**
* <pre>
* 算法:
* 1. 客户端处理完成对应的termin事件后,反馈给仲裁器处理完成。仲裁器根据对应S.E.T.L的反馈情况,判断是否删除对应的termin信号
* </pre>
*/
public void ack(TerminEventData data) {
Assert.notNull(data);
// 目前只有select模块需要发送ack信号,这里一旦收到一个信号后就删除对应的termin节点,后续可扩展
// 删除termin节点
String path = StagePathUtils.getTermin(data.getPipelineId(), data.getProcessId());
try {
zookeeper.delete(path);
} catch (ZkNoNodeException e) {
// ignore,说明节点已经被删除
} catch (ZkException e) {
throw new ArbitrateException("Termin_ack", e);
}
TerminMonitor terminMonitor = ArbitrateFactory.getInstance(data.getPipelineId(), TerminMonitor.class);
terminMonitor.ack(data.getProcessId());
}
Aggregations