I found the cause of the error. Because of a corrupt block.
However, the length of the corrupt block and the GS are the same as a
normal Block.
Therefore, HDFS cannot recognize that block as a corrupt block.
I excluded the datanode that had a corrupt block.

Thanks.

2021년 6월 28일 (월) 오후 2:02, Minwoo Kang <[email protected]>님이 작성:

> Hello,
>
> I met a strange issue. However, I don't understand why it occurred.
> That is "On-disk size without header provided is 65347, but block header
> contains -620432417. Block offset: 117193180315, data starts with:".
>
> Call stack:
>   at
> org.apache.hadoop.hbase.io.hfile.HFileBlock.validateOnDiskSizeWithoutHeader(HFileBlock.java:521)
>   at
> org.apache.hadoop.hbase.io.hfile.HFileBlock.access$700(HFileBlock.java:88)
>   at
> org.apache.hadoop.hbase.io.hfile.HFileBlock$FSReaderImpl.readBlockDataInternal(HFileBlock.java:1671)
>   at
> org.apache.hadoop.hbase.io.hfile.HFileBlock$FSReaderImpl.readBlockData(HFileBlock.java:1538)
>   at
> org.apache.hadoop.hbase.io.hfile.HFileReaderV2.readBlock(HFileReaderV2.java:452)
>
> The HBase version is 1.2.6
> I know 1.2.6 reached EOL.
> (However, it is too hard migration to the new cluster. That the reason why
> I operate this cluster.)
>
> It looks like a block is wrong. When it occurred, Every read request that
> related to the error block didn't complete.
> I don't know how to resolve this. The only way to resolve this issue is
> major compaction.
> After major compaction (means a block that looks wrong becomes an invalid
> block), a Read request works fine.
>
> I found an issue https://issues.apache.org/jira/browse/HBASE-20761.
> I am not sure it is related.
> But HBASE-20761 mentioned "This will cause further attempts to read the
> block to fail since we will still retry the corrupt replica instead of
> reporting the corrupt replica and trying a different one.", this looks like
> a key to solve this issue.
>
> Our cluster configuration
>   - hbase.regionserver.checksum.verify=true
>   - dfs.client.read.shortcircuit.skip.checksum=false
>
> Does anyone have a similar situation?
>
> Thanks.
>
>

Reply via email to