hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Doug Cutting (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HADOOP-1134) Block level CRCs in HDFS
Date Fri, 01 Jun 2007 17:09:16 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-1134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12500763
] 

Doug Cutting commented on HADOOP-1134:
--------------------------------------

Calvin Yu noted on hadoop-user that join() seems to sometimes hang even if the thread has
been interrupted.  In other places we use the idiom of a 'running' flag that's checked in
a thread's loop in conjunction with an interrupt, rather than interrupt+join, and that seems
to be reliable.  So I think we should switch to that here to.

Also, in the current patch, I don't see why the thread is held in a field.  I worry that someone
might add code like 'if (sortProgressThread == null) ...', and that we might somehow not always
null this field.  If it is kept in a local variable around the call then this is much less
of a risk.

So I think we should convert the createProgressThread method to a nested class whose constructor
starts the thread and which has a stop() method that sets a flag.  It would also be good if
the 'try' block could be shared between 'collect()' and 'flush()'.  I think this calls for
a new method something like:

private void sortWithProgress() {
  ProgressThread progress = new ProgressThread();
  try {
     sortAndSpillToDisk();
  } finally {
     progress.stop();
  }
}



> Block level CRCs in HDFS
> ------------------------
>
>                 Key: HADOOP-1134
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1134
>             Project: Hadoop
>          Issue Type: New Feature
>          Components: dfs
>            Reporter: Raghu Angadi
>            Assignee: Raghu Angadi
>         Attachments: bc-no-upgrade-05302007.patch, DfsBlockCrcDesign-05305007.htm
>
>
> Currently CRCs are handled at FileSystem level and are transparent to core HDFS. See
recent improvement HADOOP-928 ( that can add checksums to a given filesystem ) regd more about
it. Though this served us well there a few disadvantages :
> 1) This doubles namespace in HDFS ( or other filesystem implementations ). In many cases,
it nearly doubles the number of blocks. Taking namenode out of CRCs would nearly double namespace
performance both in terms of CPU and memory.
> 2) Since CRCs are transparent to HDFS, it can not actively detect corrupted blocks. With
block level CRCs, Datanode can periodically verify the checksums and report corruptions to
namnode such that name replicas can be created.
> We propose to have CRCs maintained for all HDFS data in much the same way as in GFS.
I will update the jira with detailed requirements and design. This will include same guarantees
provided by current implementation and will include a upgrade of current data.
>  

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message