[bigdata] 使用Flume hdfs sink， hdfs文件未关闭的问题

现象： 执行mapreduce任务时失败

通过hadoop fsck -openforwrite命令查看发现有文件没有关闭。

[root@com ~]# hadoop fsck -openforwrite /data/rc/click/mpp/15-08-05/
DEPRECATED: Use of this script to execute hdfs command is deprecated.
Instead use the hdfs command for it.

Connecting to namenode via http://com.hunantv.hadoopnamenode:50070
FSCK started by root (auth:SIMPLE) from /10.100.1.46 for path /data/rc/click/mpp/15-08-05/ at Thu Aug 06 14:05:03 CST 2015
....................................................................................................
....................................................................................................
........./data/rc/click/mpp/15-08-05/FlumeData.1438758322864 42888 bytes, 1 block(s), OPENFORWRITE:
/data/rc/click/mpp/15-08-05/FlumeData.1438758322864: Under replicated BP-1672356070-10.100.1.36-1412072991411:blk_1120646538_47162789{blockUCState=UNDER_CONSTRUCTION, primaryNodeIndex=-1, replicas=[ReplicaUnderConstruction[[DISK]DS-f4fff5f3-f3fd-4054-a75c-1d7da53a73af:NORMAL|FINALIZED], ReplicaUnderConstruction[[DISK]DS-26f54bc5-5026-4e6a-94ec-8435224e4aa9:NORMAL|RWR], ReplicaUnderConstruction[[DISK]DS-4ab3fffc-6468-47df-8023-79f23a330371:NORMAL|FINALIZED]]}. Target Replicas is 3 but found 2 replica(s).
..........................................................................................
............................Status: HEALTHY
Total size: 99186583 B
Total dirs: 1
Total files: 328
Total symlinks: 0
Total blocks (validated): 328 (avg. block size 302398 B)
Minimally replicated blocks: 328 (100.0 %)
Over-replicated blocks: 0 (0.0 %)
Under-replicated blocks: 1 (0.30487806 %)
Mis-replicated blocks: 0 (0.0 %)
Default replication factor: 3
Average block replication: 2.996951
Corrupt blocks: 0
Missing replicas: 1 (0.101626016 %)
Number of data-nodes: 59
Number of racks: 6
FSCK ended at Thu Aug 06 14:05:03 CST 2015 in 36 milliseconds

The filesystem under path '/data/rc/click/mpp/15-08-05/' is HEALTHY

查看FLume日志

[root@10.100.1.117] out: 05 Aug 2015 11:15:19,322 INFO [SinkRunner-PollingRunner-DefaultSinkProcessor] (org.apache.flume.sink.hdfs.BucketWriter.open:234) - Creating hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp
[root@10.100.1.117] out: 05 Aug 2015 11:16:20,493 INFO [hdfs-sin_hdfs_201-roll-timer-0] (org.apache.flume.sink.hdfs.BucketWriter$5.call:429) - Closing idle bucketWriter hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp at 1438744580493
[root@10.100.1.117] out: 05 Aug 2015 11:16:20,497 INFO [hdfs-sin_hdfs_201-roll-timer-0] (org.apache.flume.sink.hdfs.BucketWriter.close:363) - Closing hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp
[root@10.100.1.117] out: 05 Aug 2015 11:16:30,501 WARN [hdfs-sin_hdfs_201-roll-timer-0] (org.apache.flume.sink.hdfs.BucketWriter.close:370) - failed to close() HDFSWriter for file (hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp). Exception follows.
[root@10.100.1.117] out: java.io.IOException: Callable timed out after 10000 ms on file: hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp
[root@10.100.1.117] out: 05 Aug 2015 11:16:30,503 INFO [hdfs-sin_hdfs_201-call-runner-7] (org.apache.flume.sink.hdfs.BucketWriter$8.call:629) - Renaming hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293.tmp to hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-08-05/FlumeData.1438744519293

关闭hdfs文件操作因为超时失败，

查看源码：

 public synchronized void close(boolean callCloseCallback)
    throws IOException, InterruptedException {
    checkAndThrowInterruptedException();
    try {
      flush();
    } catch (IOException e) {
      LOG.warn("pre-close flush failed", e);
    }
    boolean failedToClose = false;
    LOG.info("Closing {}", bucketPath);
    CallRunner<Void> closeCallRunner = createCloseCallRunner();
    if (isOpen) {
      try {
        callWithTimeout(closeCallRunner);
        sinkCounter.incrementConnectionClosedCount();
      } catch (IOException e) {
        LOG.warn(
          "failed to close() HDFSWriter for file (" + bucketPath +
            "). Exception follows.", e);
        sinkCounter.incrementConnectionFailedCount();
        failedToClose = true;
      }
      isOpen = false;
    } else {
      LOG.info("HDFSWriter is already closed: {}", bucketPath);
    }

    // NOTE: timed rolls go through this codepath as well as other roll types
    if (timedRollFuture != null && !timedRollFuture.isDone()) {
      timedRollFuture.cancel(false); // do not cancel myself if running!
      timedRollFuture = null;
    }

    if (idleFuture != null && !idleFuture.isDone()) {
      idleFuture.cancel(false); // do not cancel myself if running!
      idleFuture = null;
    }

    if (bucketPath != null && fileSystem != null) {
      // could block or throw IOException
      try {
        renameBucket(bucketPath, targetPath, fileSystem);
      } catch(Exception e) {
        LOG.warn(
          "failed to rename() file (" + bucketPath +
          "). Exception follows.", e);
        sinkCounter.incrementConnectionFailedCount();
        final Callable<Void> scheduledRename =
                createScheduledRenameCallable();
        timedRollerPool.schedule(scheduledRename, retryInterval,
                TimeUnit.SECONDS);
      }
    }
    if (callCloseCallback) {
      runCloseAction();
      closed = true;
    }
  }

默认超时为10000ms，失败后没有重试，代码中有 failedToClose变量， 但未用到，可能开发人员忘了处理了。。。

解决方法：

1. 配置调用操作超时时间，将其调大一点，如5分钟。Flume hdfs sink配置如下：

agent12.sinks.sin_hdfs_201.type=hdfs
agent12.sinks.sin_hdfs_201.channel=ch_hdfs_201
agent12.sinks.sin_hdfs_201.hdfs.path=hdfs://com.hunantv.hadoopnamenode:8020/data/logs/amobile/vod/15-%{month}-%{day}
agent12.sinks.sin_hdfs_201.hdfs.round=true
agent12.sinks.sin_hdfs_201.hdfs.roundValue=10
agent12.sinks.sin_hdfs_201.hdfs.roundUnit=minute
agent12.sinks.sin_hdfs_201.hdfs.fileType=DataStream
agent12.sinks.sin_hdfs_201.hdfs.writeFormat=Text
agent12.sinks.sin_hdfs_201.hdfs.rollInterval=0
agent12.sinks.sin_hdfs_201.hdfs.rollSize=209715200
agent12.sinks.sin_hdfs_201.hdfs.rollCount=0
agent12.sinks.sin_hdfs_201.hdfs.idleTimeout=300
agent12.sinks.sin_hdfs_201.hdfs.batchSize=100
agent12.sinks.sin_hdfs_201.hdfs.minBlockReplicas=1
agent12.sinks.sin_hdfs_201.hdfs.callTimeout=300000

　　

2. 修改源码，增加重试。如下：

 public synchronized void close(boolean callCloseCallback)
            throws IOException, InterruptedException {
        checkAndThrowInterruptedException();
        try {
            flush();
        } catch (IOException e) {
            LOG.warn("pre-close flush failed", e);
        }
        boolean failedToClose = false;
        LOG.info("Closing {}", bucketPath);
        CallRunner<Void> closeCallRunner = createCloseCallRunner();
        int tryTime = 1;
        while (isOpen && tryTime <= 5) {
            try {
                callWithTimeout(closeCallRunner);
                sinkCounter.incrementConnectionClosedCount();
            } catch (IOException e) {
                LOG.warn(
                        "failed to close() HDFSWriter for file (try times:" + tryTime + "): " + bucketPath +
                                ". Exception follows.", e);
                sinkCounter.incrementConnectionFailedCount();
                failedToClose = true;
            }
            if (failedToClose) {
                isOpen = true;
                tryTime++;
                Thread.sleep(this.callTimeout);
            } else {
                isOpen = false;
            }
        }
        //如果isopen失敗
        if (isOpen) {
            LOG.error("failed to close file: " + bucketPath + " after " + tryTime + " tries.");
        } else {
            LOG.info("HDFSWriter is already closed: {}", bucketPath);
        }

        // NOTE: timed rolls go through this codepath as well as other roll types
        if (timedRollFuture != null && !timedRollFuture.isDone())

        {
            timedRollFuture.cancel(false); // do not cancel myself if running!
            timedRollFuture = null;
        }

        if (idleFuture != null && !idleFuture.isDone())

        {
            idleFuture.cancel(false); // do not cancel myself if running!
            idleFuture = null;
        }

        if (bucketPath != null && fileSystem != null) {
            // could block or throw IOException
            try {
                renameBucket(bucketPath, targetPath, fileSystem);
            } catch (Exception e) {
                LOG.warn(
                        "failed to rename() file (" + bucketPath +
                                "). Exception follows.", e);
                sinkCounter.incrementConnectionFailedCount();
                final Callable<Void> scheduledRename =
                        createScheduledRenameCallable();
                timedRollerPool.schedule(scheduledRename, retryInterval,
                        TimeUnit.SECONDS);
            }
        }
        if (callCloseCallback)

        {
            runCloseAction();
            closed = true;
        }
    }