use of com.alibaba.otter.shared.arbitrate.exception.ArbitrateException in project otter by alibaba.
the class SelectZooKeeperArbitrateEvent method single.
/**
* <pre>
* 算法:
* 1. 创建对应的selected节点,标志selected已完成
* </pre>
*
* @param pipelineId 同步流id
*/
public void single(EtlEventData data) {
Assert.notNull(data);
String path = StagePathUtils.getSelectStage(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[{}] select 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[{}] select ignore processId[{}] single by data:{}", new Object[] { data.getPipelineId(), data.getProcessId(), data });
} catch (ZkException e) {
throw new ArbitrateException("Select_single", e.getMessage(), e);
}
}
use of com.alibaba.otter.shared.arbitrate.exception.ArbitrateException in project otter by alibaba.
the class SelectZooKeeperArbitrateEvent method await.
// private TerminArbitrateEvent terminEvent;
/**
* <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,创建空的EventData对象,添加next node信息后直接返回
* </pre>
*/
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();
SelectStageListener selectStageListener = ArbitrateFactory.getInstance(pipelineId, SelectStageListener.class);
// 符合条件的processId
Long processId = selectStageListener.waitForProcess();
ChannelStatus status = permitMonitor.getChannelPermit();
if (status.isStart()) {
try {
EtlEventData eventData = new EtlEventData();
eventData.setPipelineId(pipelineId);
eventData.setProcessId(processId);
// 返回当前时间
eventData.setStartTime(new Date().getTime());
// 获取下一个处理节点信息
Node node = LoadBalanceFactory.getNextExtractNode(pipelineId);
if (node == null) {
// terminEvent.single(termin);
throw new ArbitrateException("Select_single", "no next node");
} else {
eventData.setNextNid(node.getId());
// 标记为已使用
markUsed(eventData);
// 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("Select_await", e.getMessage(), e);
}
} else {
logger.warn("pipelineId[{}] select ignore processId[{}] by status[{}]", new Object[] { pipelineId, processId, status });
// add by ljh 2013-02-01
// 遇到一个bug:
// a. 某台机器发起了一个RESTART指令,然后开始删除process列表
// b. 此时另一个台机器(select工作节点),并没有收到PAUSE的推送,导致还会再创建一个process节点
// c. 后续收到PAUSE指令后,丢弃了processId,就出现了unused的processId
// 这里删除了,要考虑一个问题,就是和restart指令在并行删除同一个processId时的并发考虑,目前来看没问题
String path = StagePathUtils.getProcess(pipelineId, processId);
// 忽略删除失败
zookeeper.delete(path);
// 递归调用
return await(pipelineId);
}
}
use of com.alibaba.otter.shared.arbitrate.exception.ArbitrateException 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());
}
use of com.alibaba.otter.shared.arbitrate.exception.ArbitrateException 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 com.alibaba.otter.shared.arbitrate.exception.ArbitrateException in project otter by alibaba.
the class TerminZooKeeperArbitrateEvent method exhaust.
/**
* 消耗掉所有的termin信号
*/
public void exhaust(Long pipelineId) {
Assert.notNull(pipelineId);
TerminMonitor terminMonitor = ArbitrateFactory.getInstance(pipelineId, TerminMonitor.class);
int size = terminMonitor.size();
try {
for (int i = 0; i < size; i++) {
Long processId;
processId = terminMonitor.waitForProcess();
TerminEventData data = new TerminEventData();
data.setPipelineId(pipelineId);
data.setProcessId(processId);
ack(data);
}
} catch (InterruptedException e) {
throw new ArbitrateException(e);
}
}
Aggregations