[
https://issues.apache.org/jira/browse/HDFS-13642?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16495527#comment-16495527
]
Xiao Chen commented on HDFS-13642:
----------------------------------
Looking into DFSStripedOutputStream code and the design, I think this should be
rejected. With a block size smaller than cell size, RS can never happen. Even
if we fix the output stream to always replicate like if a small file is
created, it defeats the purpose of EC.
Default block size is 128MB, and I can't think of a use case where a block size
less than 1MB is desired. Given the min-block-size defaults to 1MB, and cell
size is 1MB, it also requires manual config change in NN to ride over the
min-block-size check. So this Jira will make NN behavior more explicit by
throwing at file creation, rather than failing when in output stream writing.
> Creating a file with block size smaller than EC policy's cell size should
> throw
> -------------------------------------------------------------------------------
>
> Key: HDFS-13642
> URL: https://issues.apache.org/jira/browse/HDFS-13642
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: erasure-coding
> Affects Versions: 3.0.0
> Reporter: Xiao Chen
> Assignee: Xiao Chen
> Priority: Major
> Attachments: HDFS-13642.01.patch
>
>
> The following command causes an exception:
> {noformat}
> hadoop fs -Ddfs.block.size=349696 -put -f lineitem_sixblocks.parquet
> /test-warehouse/tmp123ec
> {noformat}
> {noformat}
> 18/05/25 16:00:59 WARN hdfs.DataStreamer: DataStreamer Exception
> java.io.IOException: BlockSize 349696 < lastByteOffsetInBlock, #0:
> blk_-9223372036854574256_14634, packet seqno: 7 offsetInBlock: 349696
> lastPacketInBlock: false lastByteOffsetInBlock: 350208
> at org.apache.hadoop.hdfs.DataStreamer.run(DataStreamer.java:729)
> at
> org.apache.hadoop.hdfs.StripedDataStreamer.run(StripedDataStreamer.java:46)
> 18/05/25 16:00:59 WARN hdfs.DFSOutputStream: Failed: offset=4096, length=512,
> DFSStripedOutputStream:#0: failed, blk_-9223372036854574256_14634
> java.io.IOException: BlockSize 349696 < lastByteOffsetInBlock, #0:
> blk_-9223372036854574256_14634, packet seqno: 7 offsetInBlock: 349696
> lastPacketInBlock: false lastByteOffsetInBlock: 350208
> at org.apache.hadoop.hdfs.DataStreamer.run(DataStreamer.java:729)
> at
> org.apache.hadoop.hdfs.StripedDataStreamer.run(StripedDataStreamer.java:46)
> {noformat}
> Then the streamer is confused and hangs.
> The local file is under 6MB, the hdfs file has a RS-3-2-1024k EC policy.
>
> Credit to [~tarasbob] for reporting this issue.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
