> On Aug 12, 2016, at 8:19 AM, Junping Du <j...@hortonworks.com> wrote:
> 
>  In this community, we are so aggressive to drop Java 7 support in 3.0.x 
> release. Here, why we are so conservative to keep releasing new bits to 
> support Java 6?

        I don't view a group of people putting bug fixes into a micro release 
as particularly conservative.  If a group within the community wasn't 
interested in doing it, 2.6.5 wouldn't be happening.

        But let's put the releases into context, because I think it tells a 
more interesting story.

                * hadoop 2.6.x = EOLed JREs (6,7) 
                * hadoop 2.7 -> hadoop 2.x = transitional (7,8)
                * hadoop 3.x = JRE 8
                * hadoop 4.x = JRE 9 

        There are groups of people still using JDK6 and they want bug fixes in 
a maintenance release.  Boom, there's 2.6.x.

        Hadoop 3.x has been pushed off for years for "reasons".  So we still 
have releases coming off of branch-2.  If 2.7 had been released as 3.x, this 
chart would look less weird. But it wasn't thus 2.x has this weird wart in the 
middle of that supports JDK7 and JDK8.  Given the public policy and roadmaps of 
at least one major vendor at the time of this writing, we should expect to see 
JDK7 support for at least the next two years after 3.x appears. Bang, there's 
2.x, where x is some large number.

        Then there is the future.  People using JRE 8 want to use newer 
dependencies.  A reasonable request. Some of these dependency updates won't 
work with JRE 7.   We can't do that in hadoop 2.x in any sort of compatible way 
without breaking the universe. (Tons of JIRAs on this point.) This means we can 
only do it in 3.x (re: Hadoop Compatibility Guidelines).  Kapow, there's 3.x

        The log4j community has stated that v1 won't work with JDK9. In turn, 
this means we'll need to upgrade to v2 at some point.  Upgrading to v2 will 
break the log4j properties file (and maybe other things?). Another incompatible 
change and it likely won't appear until Apache Hadoop v4 unless someone takes 
the initiative to fix it before v3 hits store shelves.  This makes JDK9 the 
likely target for Apache Hadoop v4.  

        Having major release cadences tied to JRE updates isn't necessarily a 
bad thing and definitely forces the community to a) actually stop beating 
around the bush on majors and b) actually makes it relatively easy to determine 
what the schedule looks like to some degree.





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

Reply via email to