>
> The patches even going back as far as last September have all removed
> the old code path. I sort of assumed that, if you are taking time to
> review the patches, you would have noticed this... additionally,
> Colin's comments on the JIRA said as much... eg:
>

Todd, we have different ways of reviewing and tracking work that
happens in a branch. In fact if you look at snapshots branch, a
functionality was coded and optimized/rewritten multiple times.
My approach is to review final consolidated patch. That is the reason
why I have been keeping HADOOP-8562 updated, so reviewers
can see the latest patch.


>  "The old RPC is now deprecated and will always throw an
> AccessControlException, so that older clients will fall back to remote
> reads."
> "BlockReaderLocal: simpler implementation that uses raw FileChannel
> objects. We don't need to cache anything, or make RPCs to the
> DataNode."


Sorry this is not explicit. HDFS-347 has been around for a long time.
Stating "HDFS-2246 short circuit mechanism will be removed from
the client" is what I call explicitly stating it.

To summarize I want to retain old short circuit and will spend time
maintaining it for the following reasons:
- It is a functionality that is already available and it supports all
  all the platforms.
- I know great deal of testing has gone into HDFS-347. But if any
  issue is discovered, it is good to have old functionality as a fallback
  option.

Please consider these thoughts and do not take it as impeding the work.
I am actually trying to help here.

Regards,
Suresh


-- 
http://hortonworks.com/download/

Reply via email to