Not sure it's that simple. I think this requires reflection based API
invocation based on hadoop version. Also the hadoop version itself needs to
be passed from build file into Java code.
Hadoop eco system components like Hive adopt a technique called shim
loaders to solve this type of problems.
So I compiled the 2.5 fgs against 2.6 when I was testing. If we abstract
this right it may just be an if statement or two.
On Aug 7, 2015 6:47 PM, Santosh Marella smare...@maprtech.com wrote:
Myriad code base compiled against hadoop 2.7 should work on hadoop 2.5
cluster as long as FGS (i.e.
Myriad code base compiled against hadoop 2.7 should work on hadoop 2.5
cluster as long as FGS (i.e. zero profile NM) is not used.
Verified the above. As long as FGS (zero profile NM) is not used,
Myriad compiled against hadoop 2.7 will work on hadoop 2.5.
Thanks,
Santosh
On Fri, Aug 7, 2015
It will make working on HA easier. However, one concern that's been
addressed previously is that FGS works for Hadoop 2.6.0+. Do we plan to
support 2.5.X (anything lower?) also as Santosh has a way to do that, if so
how do we facilitate that? Profiles?
Darin
On Fri, Aug 7, 2015 at 1:46 PM,
It will make working on HA easier
Oh Yes!
how do we facilitate that? Profiles?
Profiles might be one way. Currently, FGS is supported for zero profile
only.
And we have seen there was an API incompatibility from 2.5 to 2.6+ in FGS
code.
So, ideally (since I haven't tried it myself), when FGS is