Re: Merge IPv6

[Elliott Clark <ec...@apache.org>]

HADOOP-11890 Is the related jira.

On Fri, Aug 19, 2016 at 10:09 AM, Elliott Clark <ec...@apache.org> wrote:
> Is it time to merge IPv6 branch?
>
> We have been running the patches in several production and test
> clusters. We have tested it on ipv4 only clusters, on clusters where
> DFS is ipv4 and HBase is using ipv6, and we've tested it on ipv6 only
> clusters. We've seen no performance degredation, and have had no
> issues with network name resolution.
>
> Thoughts?
>
> Thanks

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-help@hadoop.apache.org

Re: Merge IPv6

[Allen Wittenauer <aw...@effectivemachines.com>]

> On Aug 19, 2016, at 4:38 PM, Elliott Clark <ec...@apache.org> wrote:
> 
> HADOOP-11890 Is the related jira.


I've added some comments to that JIRA regarding an actual code issue.

But I have a general question that probably isn't appropriate for the JIRA (yet?).

Way back in the 0.20 days, the NN didn't handle IP multi-pathing very well without a lot of extra DNS rigging such that all addresses used could be resolved back to the hostname in the includes file.  Packets from non-matching addresses were, of course, dropped.  Given the nature of IPv6, has the testing included only statically assigned addresses or has stateless autoconfig (either local or RA-based) been tested?  If so, what was the state of DNS vs. the include file?  If only static has been tested, is that documented so that users don't fall into a hole?

Thanks.
---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-dev-help@hadoop.apache.org

Re: Merge IPv6

[Allen Wittenauer <aw...@effectivemachines.com>]

> On Aug 19, 2016, at 4:38 PM, Elliott Clark <ec...@apache.org> wrote:
> 
> HADOOP-11890 Is the related jira.


I've added some comments to that JIRA regarding an actual code issue.

But I have a general question that probably isn't appropriate for the JIRA (yet?).

Way back in the 0.20 days, the NN didn't handle IP multi-pathing very well without a lot of extra DNS rigging such that all addresses used could be resolved back to the hostname in the includes file.  Packets from non-matching addresses were, of course, dropped.  Given the nature of IPv6, has the testing included only statically assigned addresses or has stateless autoconfig (either local or RA-based) been tested?  If so, what was the state of DNS vs. the include file?  If only static has been tested, is that documented so that users don't fall into a hole?

Thanks.
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-dev-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-dev-help@hadoop.apache.org