FAQ
The problem we are having is that datanodes periodically stall for 10-15
minutes and drop off the active list and then come back.

What is going on is that a long operation set is holding the lock on on
FSDataset.volumes, and all of the other block service requests stall
behind this lock.

"DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10 tid=0x4d7ad400
nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
java.lang.Thread.State: RUNNABLE
at java.lang.String.lastIndexOf(String.java:1628)
at java.io.File.getName(File.java:399)
at
org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)
at
org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
at
org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)
at
org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)
- locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
at org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
at java.lang.Thread.run(Thread.java:619)

This is basically taking a stat on every hdfs block on the datanode,
which in our case is ~ 2million, and can take 10+ minutes (we may be
experiencing problems with our raid controller but have no visibility
into it) at the OS level the file system seems fine and operations
eventually finish.

It appears that a couple of different data structures are being locked
with the single object FSDataset$Volume.

Then this happens:
"org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry [0x459fe000..0x459ff0d0]
java.lang.Thread.State: BLOCKED (on object monitor)
at
org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)
- waiting to lock <0x551e8d48> (a
org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
- locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
at
org.apache.hadoop.dfs.DataNode$BlockReceiver.(DataNode.java:1187)
at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
at java.lang.Thread.run(Thread.java:619)

which locks the FSDataset while waiting on the volume object

and now all of the Datanode operations stall waiting on the FSDataset
object.
----------

Our particular installation doesn't use multiple directories for hdfs,
so a first simple hack for a local fix would be to modify getNextVolume
to just return the single volume and not be synchronized

A richer alternative would be to make the locking more fine grained on
FSDataset$FSVolumeSet.

Of course we are also trying to fix the file system performance and dfs
block loading that results in the block report taking a long time.

Any suggestions or warnings?

Thanks.

Search Discussions

  • Konstantin Shvachko at Jan 9, 2009 at 7:47 pm
    Hi Jason,

    2 million blocks per data-node is not going to work.
    There were discussions about it previously, please
    check the mail archives.

    This means you have a lot of very small files, which
    HDFS is not designed to support. A general recommendation
    is to group small files into large ones, introducing
    some kind of record structure delimiting those small files,
    and control it in on the application level.

    Thanks,
    --Konstantin


    Jason Venner wrote:
    The problem we are having is that datanodes periodically stall for 10-15
    minutes and drop off the active list and then come back.

    What is going on is that a long operation set is holding the lock on on
    FSDataset.volumes, and all of the other block service requests stall
    behind this lock.

    "DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10 tid=0x4d7ad400
    nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
    java.lang.Thread.State: RUNNABLE
    at java.lang.String.lastIndexOf(String.java:1628)
    at java.io.File.getName(File.java:399)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)

    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)

    - locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
    at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
    at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
    at java.lang.Thread.run(Thread.java:619)

    This is basically taking a stat on every hdfs block on the datanode,
    which in our case is ~ 2million, and can take 10+ minutes (we may be
    experiencing problems with our raid controller but have no visibility
    into it) at the OS level the file system seems fine and operations
    eventually finish.

    It appears that a couple of different data structures are being locked
    with the single object FSDataset$Volume.

    Then this happens:
    "org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
    tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry
    [0x459fe000..0x459ff0d0]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)

    - waiting to lock <0x551e8d48> (a
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
    - locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.<init>(DataNode.java:2322)
    at
    org.apache.hadoop.dfs.DataNode$DataXceiver.writeBlock(DataNode.java:1187)
    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    which locks the FSDataset while waiting on the volume object

    and now all of the Datanode operations stall waiting on the FSDataset
    object.
    ----------

    Our particular installation doesn't use multiple directories for hdfs,
    so a first simple hack for a local fix would be to modify getNextVolume
    to just return the single volume and not be synchronized

    A richer alternative would be to make the locking more fine grained on
    FSDataset$FSVolumeSet.

    Of course we are also trying to fix the file system performance and dfs
    block loading that results in the block report taking a long time.

    Any suggestions or warnings?

    Thanks.

  • Raghu Angadi at Jan 9, 2009 at 10:49 pm
    2M files is excessive. But there is no reason block reports should
    break. My preference is to make block reports handle this better. DNs
    dropping in and out of the cluster causes too many other problems.

    Raghu.

    Konstantin Shvachko wrote:
    Hi Jason,

    2 million blocks per data-node is not going to work.
    There were discussions about it previously, please
    check the mail archives.

    This means you have a lot of very small files, which
    HDFS is not designed to support. A general recommendation
    is to group small files into large ones, introducing
    some kind of record structure delimiting those small files,
    and control it in on the application level.

    Thanks,
    --Konstantin
  • Raghu Angadi at Jan 9, 2009 at 10:46 pm
    The scan required for each block report is well known issue and it can
    be fixed. It was discussed multiple times (e.g.
    https://issues.apache.org/jira/browse/HADOOP-3232?focusedCommentId=12587795#action_12587795
    ).

    Earlier, inline 'du' on datanodes used to cause the same problem and
    they they were moved to a separate thread (HADOOP-3232). block reports
    can do the same...

    Though 2M blocks on DN is very large, there is no reason block reports
    should break things. Once we fix block reports, something else might
    break.. but that is different issue.

    Raghu.

    Jason Venner wrote:
    The problem we are having is that datanodes periodically stall for 10-15
    minutes and drop off the active list and then come back.

    What is going on is that a long operation set is holding the lock on on
    FSDataset.volumes, and all of the other block service requests stall
    behind this lock.

    "DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10 tid=0x4d7ad400
    nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
    java.lang.Thread.State: RUNNABLE
    at java.lang.String.lastIndexOf(String.java:1628)
    at java.io.File.getName(File.java:399)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)

    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)

    - locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
    at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
    at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
    at java.lang.Thread.run(Thread.java:619)

    This is basically taking a stat on every hdfs block on the datanode,
    which in our case is ~ 2million, and can take 10+ minutes (we may be
    experiencing problems with our raid controller but have no visibility
    into it) at the OS level the file system seems fine and operations
    eventually finish.

    It appears that a couple of different data structures are being locked
    with the single object FSDataset$Volume.

    Then this happens:
    "org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
    tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry
    [0x459fe000..0x459ff0d0]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)

    - waiting to lock <0x551e8d48> (a
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
    - locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.<init>(DataNode.java:2322)
    at
    org.apache.hadoop.dfs.DataNode$DataXceiver.writeBlock(DataNode.java:1187)
    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    which locks the FSDataset while waiting on the volume object

    and now all of the Datanode operations stall waiting on the FSDataset
    object.
    ----------

    Our particular installation doesn't use multiple directories for hdfs,
    so a first simple hack for a local fix would be to modify getNextVolume
    to just return the single volume and not be synchronized

    A richer alternative would be to make the locking more fine grained on
    FSDataset$FSVolumeSet.

    Of course we are also trying to fix the file system performance and dfs
    block loading that results in the block report taking a long time.

    Any suggestions or warnings?

    Thanks.

  • Sagar Naik at Jan 10, 2009 at 5:59 pm
    Hi Raghu,


    The periodic "du" and block reports thread thrash the disk. (Block
    Reports takes abt on an avg 21 mins )

    and I think all the datanode threads are not able to do much and freeze

    "org.apache.hadoop.dfs.DataNode$DataXceiver@f2127a" daemon prio=10
    tid=0x41f06000 nid=0x7c7c waiting for monitor entry [0x43918000..0x43918f50]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at org.apache.hadoop.dfs.FSDataset.getFile(FSDataset.java:1158)
    - waiting to lock <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.FSDataset.validateBlockFile(FSDataset.java:1074)
    at org.apache.hadoop.dfs.FSDataset.isValidBlock(FSDataset.java:1066)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:894)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.(DataNode.java:1187)
    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    Locked ownable synchronizers:
    - None

    "org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
    tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry [0x459fe000..0x459ff0d0]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)
    - waiting to lock <0x551e8d48> (a
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
    - locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.(DataNode.java:1187)
    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    Locked ownable synchronizers:
    - None

    "DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10 tid=0x4d7ad400
    nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
    java.lang.Thread.State: RUNNABLE
    at java.lang.String.lastIndexOf(String.java:1628)
    at java.io.File.getName(File.java:399)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)
    - locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
    at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
    at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
    at java.lang.Thread.run(Thread.java:619)


    and the lock is *<0x54e550e0> *is held by another similar thread and tht
    thread is waiting on FSVolume blocked by getBlockReport()

    Infact, during this time, the datanode appears as dead node and clients
    keep on getting createBlockException with timeout

    We dont see this problem on other DNs with less number of blocks. So I
    think, 2Million files is an issue here

    Pl correct me if I missed on something

    -Sagar

    Raghu Angadi wrote:
    The scan required for each block report is well known issue and it can
    be fixed. It was discussed multiple times (e.g.
    https://issues.apache.org/jira/browse/HADOOP-3232?focusedCommentId=12587795#action_12587795
    ).

    Earlier, inline 'du' on datanodes used to cause the same problem and
    they they were moved to a separate thread (HADOOP-3232). block reports
    can do the same...

    Though 2M blocks on DN is very large, there is no reason block reports
    should break things. Once we fix block reports, something else might
    break.. but that is different issue.

    Raghu.

    Jason Venner wrote:
    The problem we are having is that datanodes periodically stall for
    10-15 minutes and drop off the active list and then come back.

    What is going on is that a long operation set is holding the lock on
    on FSDataset.volumes, and all of the other block service requests
    stall behind this lock.

    "DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10
    tid=0x4d7ad400 nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
    java.lang.Thread.State: RUNNABLE
    at java.lang.String.lastIndexOf(String.java:1628)
    at java.io.File.getName(File.java:399)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)

    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)

    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)

    - locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at
    org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
    at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
    at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
    at java.lang.Thread.run(Thread.java:619)

    This is basically taking a stat on every hdfs block on the datanode,
    which in our case is ~ 2million, and can take 10+ minutes (we may be
    experiencing problems with our raid controller but have no visibility
    into it) at the OS level the file system seems fine and operations
    eventually finish.

    It appears that a couple of different data structures are being
    locked with the single object FSDataset$Volume.

    Then this happens:
    "org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
    tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry
    [0x459fe000..0x459ff0d0]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)

    - waiting to lock <0x551e8d48> (a
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
    - locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.<init>(DataNode.java:2322)
    at
    org.apache.hadoop.dfs.DataNode$DataXceiver.writeBlock(DataNode.java:1187)

    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    which locks the FSDataset while waiting on the volume object

    and now all of the Datanode operations stall waiting on the FSDataset
    object.
    ----------

    Our particular installation doesn't use multiple directories for
    hdfs, so a first simple hack for a local fix would be to modify
    getNextVolume to just return the single volume and not be synchronized

    A richer alternative would be to make the locking more fine grained
    on FSDataset$FSVolumeSet.

    Of course we are also trying to fix the file system performance and
    dfs block loading that results in the block report taking a long time.

    Any suggestions or warnings?

    Thanks.

  • Raghu Angadi at Jan 12, 2009 at 7:45 pm

    Sagar Naik wrote:
    Hi Raghu,


    The periodic "du" and block reports thread thrash the disk. (Block
    Reports takes abt on an avg 21 mins )

    and I think all the datanode threads are not able to do much and freeze
    yes, that is the known problem we talked about in the earlier mails in
    this thread.

    When you have millions of blocks, one hour for du and block report
    intervals is too often for you. May be you could increase it to
    something like 6 or 12 hours.

    It still does not fix the block report problem since DataNode does the
    scan in-line.

    As I mentioned in earlier mails, we should really fix the block report
    problem. As simple fix would scan (very slowly, unlike DU) the
    directories in the background.

    Even after fixing block reports, you should be aware that excessive
    number of block does impact the performance. No system can guarantee
    performance when overloaded. What we want to do is to make Hadoop
    degrade gracefully.. rather than DNs getting killed.

    Raghu.
  • Jason Venner at Jan 12, 2009 at 7:51 pm
    There is no reason to do the block scans. All of the modern kernels will
    provide you notification when an file or directory is altered.

    This could be readily handled with a native application that writes
    structured data to a receiver in the Datanode, or via JNA/JNI for pure
    java or mixed solutions.



    Raghu Angadi wrote:
    Sagar Naik wrote:
    Hi Raghu,


    The periodic "du" and block reports thread thrash the disk. (Block
    Reports takes abt on an avg 21 mins )

    and I think all the datanode threads are not able to do much and freeze
    yes, that is the known problem we talked about in the earlier mails in
    this thread.

    When you have millions of blocks, one hour for du and block report
    intervals is too often for you. May be you could increase it to
    something like 6 or 12 hours.

    It still does not fix the block report problem since DataNode does the
    scan in-line.

    As I mentioned in earlier mails, we should really fix the block report
    problem. As simple fix would scan (very slowly, unlike DU) the
    directories in the background.

    Even after fixing block reports, you should be aware that excessive
    number of block does impact the performance. No system can guarantee
    performance when overloaded. What we want to do is to make Hadoop
    degrade gracefully.. rather than DNs getting killed.

    Raghu.
  • Raghu Angadi at Jan 12, 2009 at 8:10 pm

    Jason Venner wrote:
    There is no reason to do the block scans. All of the modern kernels will
    provide you notification when an file or directory is altered.

    This could be readily handled with a native application that writes
    structured data to a receiver in the Datanode, or via JNA/JNI for pure
    java or mixed solutions.
    Agreed. pure java would be best. I didn't mean to suggest scan as the
    main solution. I only stressed that the problem needs to be fixed.

    Unlike DU, since scan would be in our control and it could be throttled
    at our own pace so that it does not thrash the disks. It will have
    minimal effect on DN.

    Please open a jira for this issue, so the these solutions could be proposed.

    would such a notifier help with thrashing from DU? May be we should have
    alternatives to calling du shell command on the entire directory.

    Raghu.
  • Jason Venner at Jan 12, 2009 at 8:48 pm
    The thought is that the notifier would stat each file as it was notified
    about it, and thus would have the real time dusage information also.
    There would be no need for the current du task or the block task after
    startup (ie: do it one time to compute the current blocks and space).

    After startup the recipient of the notifications would keep up the block
    information and the du information.

    Raghu Angadi wrote:
    Jason Venner wrote:
    There is no reason to do the block scans. All of the modern kernels
    will provide you notification when an file or directory is altered.

    This could be readily handled with a native application that writes
    structured data to a receiver in the Datanode, or via JNA/JNI for
    pure java or mixed solutions.
    Agreed. pure java would be best. I didn't mean to suggest scan as the
    main solution. I only stressed that the problem needs to be fixed.

    Unlike DU, since scan would be in our control and it could be
    throttled at our own pace so that it does not thrash the disks. It
    will have minimal effect on DN.

    Please open a jira for this issue, so the these solutions could be
    proposed.

    would such a notifier help with thrashing from DU? May be we should
    have alternatives to calling du shell command on the entire directory.

    Raghu.
  • Jason Venner at Jan 12, 2009 at 9:03 pm
    Here is some simple code I wrote using JNA to handline linux INOTIFY.
    This code was my first and only attempt to use JNA.

    The JNA jars are available from https://jna.dev.java.net/


    Raghu Angadi wrote:
    Jason Venner wrote:
    There is no reason to do the block scans. All of the modern kernels
    will provide you notification when an file or directory is altered.

    This could be readily handled with a native application that writes
    structured data to a receiver in the Datanode, or via JNA/JNI for
    pure java or mixed solutions.
    Agreed. pure java would be best. I didn't mean to suggest scan as the
    main solution. I only stressed that the problem needs to be fixed.

    Unlike DU, since scan would be in our control and it could be
    throttled at our own pace so that it does not thrash the disks. It
    will have minimal effect on DN.

    Please open a jira for this issue, so the these solutions could be
    proposed.

    would such a notifier help with thrashing from DU? May be we should
    have alternatives to calling du shell command on the entire directory.

    Raghu.
  • Jason Venner at Jan 10, 2009 at 4:59 am
    I propose an alternate solution for this.
    If the block information was managed by having a inotify task (in
    linux/solaris), and the windows equivalent which I forget, the datanode
    could be informed each time a file in the dfs tree is created, updated,
    or deleted.

    With this information being delivered, it can maintain an accurate block
    map with only 1 full scan of the datanode blocks, at start time.

    With this algorithm the data nodes will be able to scale to a much
    larger number of blocks.

    The other thing is the way the sync blocks on the FSDataset.FSVolumeSet
    are held totally aggravates this bug in 0.18.1.

    I have implemented a pure java version of inotify, using JNA
    (https://jna.dev.java.net/) and there is a windows version also
    available, or some simple jni could be written.

    The jason@attributor.com address will be going away shortly, I will be
    switching to jason.hadoop@gmail.com in the next little bit.



    Jason Venner wrote:
    The problem we are having is that datanodes periodically stall for
    10-15 minutes and drop off the active list and then come back.

    What is going on is that a long operation set is holding the lock on
    on FSDataset.volumes, and all of the other block service requests
    stall behind this lock.

    "DataNode: [/data/dfs-video-18/dfs/data]" daemon prio=10
    tid=0x4d7ad400 nid=0x7c40 runnable [0x4c698000..0x4c6990d0]
    java.lang.Thread.State: RUNNABLE
    at java.lang.String.lastIndexOf(String.java:1628)
    at java.io.File.getName(File.java:399)
    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getGenerationStampFromFile(FSDataset.java:148)

    at
    org.apache.hadoop.dfs.FSDataset$FSDir.getBlockInfo(FSDataset.java:181)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolume.getBlockInfo(FSDataset.java:412)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getBlockInfo(FSDataset.java:511)

    - locked <0x551e8d48> (a org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.getBlockReport(FSDataset.java:1053)
    at org.apache.hadoop.dfs.DataNode.offerService(DataNode.java:708)
    at org.apache.hadoop.dfs.DataNode.run(DataNode.java:2890)
    at java.lang.Thread.run(Thread.java:619)

    This is basically taking a stat on every hdfs block on the datanode,
    which in our case is ~ 2million, and can take 10+ minutes (we may be
    experiencing problems with our raid controller but have no visibility
    into it) at the OS level the file system seems fine and operations
    eventually finish.

    It appears that a couple of different data structures are being locked
    with the single object FSDataset$Volume.

    Then this happens:
    "org.apache.hadoop.dfs.DataNode$DataXceiver@1bcee17" daemon prio=10
    tid=0x4da8d000 nid=0x7ae4 waiting for monitor entry
    [0x459fe000..0x459ff0d0]
    java.lang.Thread.State: BLOCKED (on object monitor)
    at
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet.getNextVolume(FSDataset.java:473)

    - waiting to lock <0x551e8d48> (a
    org.apache.hadoop.dfs.FSDataset$FSVolumeSet)
    at org.apache.hadoop.dfs.FSDataset.writeToBlock(FSDataset.java:934)
    - locked <0x54e550e0> (a org.apache.hadoop.dfs.FSDataset)
    at
    org.apache.hadoop.dfs.DataNode$BlockReceiver.<init>(DataNode.java:2322)
    at
    org.apache.hadoop.dfs.DataNode$DataXceiver.writeBlock(DataNode.java:1187)
    at org.apache.hadoop.dfs.DataNode$DataXceiver.run(DataNode.java:1045)
    at java.lang.Thread.run(Thread.java:619)

    which locks the FSDataset while waiting on the volume object

    and now all of the Datanode operations stall waiting on the FSDataset
    object.
    ----------

    Our particular installation doesn't use multiple directories for hdfs,
    so a first simple hack for a local fix would be to modify
    getNextVolume to just return the single volume and not be synchronized

    A richer alternative would be to make the locking more fine grained on
    FSDataset$FSVolumeSet.

    Of course we are also trying to fix the file system performance and
    dfs block loading that results in the block report taking a long time.

    Any suggestions or warnings?

    Thanks.

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
groupcommon-user @
categorieshadoop
postedJan 9, '09 at 12:05a
activeJan 12, '09 at 9:03p
posts11
users4
websitehadoop.apache.org...
irc#hadoop

People

Translate

site design / logo © 2022 Grokbase