[jira] [Commented] (HDFS-7999) FsDatasetImpl#createTemporary sometimes holds the FSDatasetImpl lock for a very long time

["Colin Patrick McCabe (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HDFS-7999?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14387312#comment-14387312 ] 

Colin Patrick McCabe commented on HDFS-7999:
--------------------------------------------

[~xinwei], even if we made the heartbeat lockless, there are still many other problems associated with having {{FsDatasetImpl#createTemporary}} hold the FSDatasetImpl lock for a very long time.  Any thread that needs to read or write from the datanode will be blocked.

I might be missing something (I'm not that familiar with this code), but I'm having trouble understanding the patch.  Isn't it going to throw this exception:
{code}
1447	            throw new ReplicaAlreadyExistsException("Block " + b
1448	                + " already exists in state " + currentReplicaInfo.getState()
1449	                + " and thus cannot be created.");
{code}

whereas previously it would have called {{ReplicaInfo#stopWriter}}?  Is that behavior change really intended?

It seems like what we should have is a non-blocking version of {{stopWriter}}, as well as a loop that waits for the writer to stop.

Also what is the significance of checking if we can get the same result from {{VolumeMap#get}} twice in a row?  Why break out of the loop when that happens?  There also seem to be some potential null pointer dereferences.

> FsDatasetImpl#createTemporary sometimes holds the FSDatasetImpl lock for a very long time
> -----------------------------------------------------------------------------------------
>
>                 Key: HDFS-7999
>                 URL: https://issues.apache.org/jira/browse/HDFS-7999
>             Project: Hadoop HDFS
>          Issue Type: Bug
>    Affects Versions: 2.6.0
>            Reporter: zhouyingchao
>            Assignee: zhouyingchao
>         Attachments: HDFS-7999-001.patch
>
>
> I'm using 2.6.0 and noticed that sometime DN's heartbeat were delayed for very long time, say more than 100 seconds. I get the jstack twice and looks like they are all blocked (at getStorageReport) by dataset lock, and which is held by a thread that is calling createTemporary, which again is blocked to wait earlier incarnation writer to exit.
> The heartbeat thread stack:
>    java.lang.Thread.State: BLOCKED (on object monitor)
>         at org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsVolumeImpl.getDfsUsed(FsVolumeImpl.java:152)
>         - waiting to lock <0x00000007b01428c0> (a org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl)
>         at org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.getStorageReports(FsDatasetImpl.java:144)
>         - locked <0x00000007b0140ed0> (a java.lang.Object)
>         at org.apache.hadoop.hdfs.server.datanode.BPServiceActor.sendHeartBeat(BPServiceActor.java:575)
>         at org.apache.hadoop.hdfs.server.datanode.BPServiceActor.offerService(BPServiceActor.java:680)
>         at org.apache.hadoop.hdfs.server.datanode.BPServiceActor.run(BPServiceActor.java:850)
>         at java.lang.Thread.run(Thread.java:662)
> The DataXceiver thread holds the dataset lock:
> "DataXceiver for client at XXXXX" daemon prio=10 tid=0x00007f14041e6480 nid=0x52bc in Object.wait() [0x00007f11d78f7000]
> java.lang.Thread.State: TIMED_WAITING (on object monitor)
> at java.lang.Object.wait(Native Method)
> at java.lang.Thread.join(Thread.java:1194)
> locked <0x00000007a33b85d8> (a org.apache.hadoop.util.Daemon)
> at org.apache.hadoop.hdfs.server.datanode.ReplicaInPipeline.stopWriter(ReplicaInPipeline.java:183)
> at org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.createTemporary(FsDatasetImpl.java:1231)
> locked <0x00000007b01428c0> (a org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl)
> at org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetImpl.createTemporary(FsDatasetImpl.java:114)
> at org.apache.hadoop.hdfs.server.datanode.BlockReceiver.<init>(BlockReceiver.java:179)
> at org.apache.hadoop.hdfs.server.datanode.DataXceiver.writeBlock(DataXceiver.java:615)
> at org.apache.hadoop.hdfs.protocol.datatransfer.Receiver.opWriteBlock(Receiver.java:137)
> at org.apache.hadoop.hdfs.protocol.datatransfer.Receiver.processOp(Receiver.java:74)
> at org.apache.hadoop.hdfs.server.datanode.DataXceiver.run(DataXceiver.java:235)
> at java.lang.Thread.run(Thread.java:662)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)