[
https://issues.apache.org/jira/browse/HDFS-8375?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14551208#comment-14551208
]
Zhe Zhang commented on HDFS-8375:
---------------------------------
It's actually a tricky and deep-hidden bug not related in this patch. We are
not handling {{LocatedBlock#offset}} properly for internal parity blocks in a
group.
{{StripedBlockUtil#constructInternalBlock}} assigns the start offset of each
internal block as {{bg.getStartOffset() + idxInBlockGroup * cellSize}}, which
makes sense for internal data blocks but not parity ones. It happens to work
with 256K cell size because {{(256 x 1024) % 9 == 1}}, which is not true for
64K.
I'm +1 on the patch with {{BLOCK_STRIPED_CELL_SIZE}} set back to 256K for now.
I will file a separate JIRA to address the above start offset issue.
> Add cellSize as an XAttr to ECZone
> ----------------------------------
>
> Key: HDFS-8375
> URL: https://issues.apache.org/jira/browse/HDFS-8375
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Vinayakumar B
> Assignee: Vinayakumar B
> Attachments: HDFS-8375-HDFS-7285-01.patch,
> HDFS-8375-HDFS-7285-02.patch, HDFS-8375-HDFS-7285-03.patch
>
>
> Add {{cellSize}} as an Xattr for ECZone. as discussed
> [here|https://issues.apache.org/jira/browse/HDFS-8347?focusedCommentId=14539108&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14539108]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
