[
https://issues.apache.org/jira/browse/HDFS-8493?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14596306#comment-14596306
]
Haohui Mai commented on HDFS-8493:
----------------------------------
bq. IMO, I think this is okay, especially for write ops, provided fsn writelock
is held. And I can see many places where this resolution is done under fsn lock
held, but not fsd lock.
Can you list these places and file jiras? They are critical bugs and should be
fixed.
bq. This triggered thoughts, Why two separate locks, fsdir lock and
fsnamesystem locks.? Almost all ops are go through fsn with lock (read/write)
held, and then go on-to get fsdir locks.
Though most of the time the fsd lock is acquired within the fsn lock.
BlockManager and LeaseManager only requires the fsn lock but not the fsd lock.
We're in the process of cleaning up the locks of both fsn and fsd locks. At the
end of the day the NN should be able to process block reports w/o blocking
requests to the namespace.
> Consolidate truncate() related implementation in a single class
> ---------------------------------------------------------------
>
> Key: HDFS-8493
> URL: https://issues.apache.org/jira/browse/HDFS-8493
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Haohui Mai
> Assignee: Rakesh R
> Attachments: HDFS-8493-001.patch, HDFS-8493-002.patch,
> HDFS-8493-003.patch, HDFS-8493-004.patch, HDFS-8493-005.patch,
> HDFS-8493-006.patch, HDFS-8493-007.patch, HDFS-8493-007.patch
>
>
> This jira proposes to consolidate truncate() related methods into a single
> class.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
