package com.taobao.zeus.jobs.sub.tool;
import java.util.Collection;
import java.util.Date;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map.Entry;
import org.apache.commons.lang.StringUtils;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.hive.metastore.api.FieldSchema;
import org.apache.hadoop.hive.metastore.api.Partition;
import org.apache.hadoop.hive.metastore.api.Table;
import org.springframework.context.ApplicationContext;
import com.taobao.zeus.broadcast.alarm.ZeusAlarm;
import com.taobao.zeus.jobs.AbstractJob;
import com.taobao.zeus.jobs.JobContext;
import com.taobao.zeus.jobs.sub.conf.ConfUtil;
import com.taobao.zeus.model.processer.HiveProcesser;
import com.taobao.zeus.store.CliTableManager;
import com.taobao.zeus.store.HDFSManager;
import com.taobao.zeus.store.TableManager;
import com.taobao.zeus.util.ZeusDateTool;
public class HiveOutputCheckJob extends AbstractJob {
@SuppressWarnings("unused")
private HiveProcesser processer;
private List<String> tableNames;
private Integer percent;
private ZeusAlarm wangWangAlarm;
Integer exitCode = 0;
private Configuration conf;
private TableManager tableManager;
public HiveOutputCheckJob(final JobContext jobContext,
final HiveProcesser p, final ApplicationContext applicationContext)
throws Exception {
super(jobContext);
this.wangWangAlarm = (ZeusAlarm) applicationContext
.getBean("wangWangAlarm");
conf = ConfUtil.getDefaultCoreSite();
this.tableManager = new CliTableManager(conf);
this.processer = p;
this.tableNames = p.getOutputTables();
this.percent = p.getDriftPercent();
}
@Override
public Integer run() throws Exception {
if (jobContext.getCoreExitCode() != 0) {
jobContext.getJobHistory().getLog()
.appendZeus("Job 运行失败，不进行产出分区大小检测");
return exitCode;
}
jobContext.getJobHistory().getLog()
.appendZeus("HiveOutputCheck 开始进行产出分区大小检测");
for (String tableName : this.tableNames) {
log("开始进行检测表：" + tableName);
//FIXME hive表
Table t = tableManager.getTable("default",tableName);
// 表不存在
if (t == null) {
tableFailed("表 " + tableName + " 不存在。");
continue;
}
//FIXME hive
List<Partition> parts = tableManager.getPartitions("default", tableName, null);
// 没有分区
if (parts == null || parts.isEmpty()) {
tableFailed("表 " + tableName + " 没有获取到存在的分区。");
continue;
}
/********************
* 存在分区 ，继续检查
*/
int ptIndex = getPtIndex(t);
if (ptIndex != 0) {
log("表" + tableName + "的第一个分区字段不是pt。只支持第一个分区字段为pt的表。");
exitCode = -1;
continue;
}
String lastDay = new ZeusDateTool(new Date()).addDay(-1).format("yyyyMMdd");
// 获取最新分区
Partition lastPartition = getLastPartition(parts, lastDay, ptIndex);
if (lastPartition == null) {
tableFailed("表" + tableName + "最新分区pt=" + lastDay + "不存在");
continue;
}
// 可能有多级分区，按pt把分区大小聚合
HashMap<String, Long> sizeMap = getSizeMap(parts, ptIndex);
// 检查最新分区
Long lastSize = sizeMap.get(lastDay);
if (lastSize == null || lastSize <= 0) {
tableFailed("表" + tableName + "最新分区pt=" + lastDay
+ "目录不存在或大小为零。");
continue;
}
// 只有一个分区，不用继续检查
if (sizeMap.keySet().size() == 1) {
log("表" + tableName + "只有最新分区pt=" + lastDay + "。分区大小:"+lastSize+".不用检测浮动。");
continue;
}
// 有多个分区，检查浮动
// 去除最新分区
sizeMap.remove(lastDay);
// 去除不正常数据
long avgSize = getAvgSize(sizeMap);
// 正常数据范围
long upLimit = (long) (1.5 * avgSize);
long downLimit = (long) (0.5 * avgSize);
// 去除不正常数据,重新计算平均大小
for (Iterator<Entry<String, Long>> itr = sizeMap.entrySet()
.iterator(); itr.hasNext();) {
Entry<String, Long> entry = itr.next();
if (entry.getValue() == null || entry.getValue() < downLimit
|| entry.getValue() > upLimit) {
itr.remove();
}
}
if (sizeMap.size() != 0) {
avgSize = getAvgSize(sizeMap);
} else {
avgSize = 0;
}
// 浮动检查
log(" 表【" + tableName + "】 ");
log("有效的参考分区个数：" + sizeMap.size());
log("平均分区大小：" + avgSize);
log("设定分区大小浮动百分比：" + this.percent + "%");
log("最新分区大小：" + lastSize);
if (avgSize > 0) {
log("最新分区浮动百分比：" + (lastSize - avgSize) * 100 / (avgSize) + "%");
}
upLimit = (long) (avgSize * ((100.0 + this.percent) / 100));
downLimit = (long) (avgSize * ((100.0 - this.percent) / 100));
if (lastSize >= downLimit && lastSize <= upLimit) {
log(" 表【" + tableName + "】 " + "产出数据检测OK");
} else {
// 超出浮动范围
String jobId = jobContext.getJobHistory().getJobId();
StringBuffer sb = new StringBuffer("jobid=" + jobId + " 表【"
+ tableName + "】 " + " 产出分区大小超出浮动比例 " + this.percent
+ "%");
sb.append("\n平均分区大小为：" + avgSize);
sb.append("\n本次产出分区大小为：" + lastSize + "\n");
tableFailed(sb.toString());
}
}
return exitCode;
}
private long getAvgSize(final HashMap<String, Long> sizeMap) {
long avgTotal = 0;
for (Long size : sizeMap.values()) {
if (size != null) {
avgTotal += size / sizeMap.size();
}
}
return avgTotal;
}
private HashMap<String, Long> getSizeMap(final List<Partition> parts,
final int ptIndex) {
HashMap<String, Long> sizeMap = new HashMap<String, Long>(parts.size());
for (Partition p : parts) {
long size = HDFSManager.getPathSize(p.getSd().getLocation(), conf);
String pt = StringUtils.substring(p.getValues().get(ptIndex), 0, 8);
Long tempSize = sizeMap.get(pt);
if (tempSize == null || tempSize < 0) {
tempSize = 0L;
}
sizeMap.put(pt, tempSize + size);
}
return sizeMap;
}
@Override
public void cancel() {
canceled = true;
}
private void tableFailed(final String msg) throws Exception {
exitCode = -1;
log(msg + " 发出报警");
wangWangAlarm.alarm(jobContext.getJobHistory().getId(),
"Zeus任务产出分区检查报警", msg);
}
private int getPtIndex(final Table t) {
int ptIndex = 0;
boolean hasPt = false;
for (FieldSchema fs : t.getPartitionKeys()) {
if (fs.getName().equals("pt")) {
hasPt = true;
break;
}
ptIndex++;
}
if (!hasPt) {
return -1;
}
return ptIndex;
}
private Partition getLastPartition(final Collection<Partition> parts,
final String lastDay, final int ptIndex) {
for (Partition p : parts) {
if (StringUtils.substring(p.getValues().get(ptIndex), 0, 8).equals(
lastDay)) {
return p;
}
}
return null;
}
}