You ought to rely on the hadoop-client dependency (it's considered
downstream facing) instead of directly on hadoop-hdfs (it's internal facing
and the contents might change).

Client - Server compatibility is covered under our "Wire Compatibility"
section of the compatibility docs:

http://hadoop.apache.org/docs/r2.7.6/hadoop-project-dist/hadoop-common/Compatibility.html#Wire_compatibility

the tl;dr version of that section is "any Hadoop 2 client should be able to
talk with any other Hadoop 2 cluster".

On Thu, Apr 26, 2018 at 8:14 AM, Christophe Jolif <[email protected]> wrote:

> Maybe a stupid question, but if I have a Java program using hadoop-hdfs
> jar to write to HDFS is there any sort of compatibility scheme with the
> actual Hadoop cluster it is talking to?
>
> In other words, should the version between the two be fully identical? Can
> any 2.8.x jar talk to any 2.8.y cluster? Or 2.8 can even talk to 2.7?
>
> Thanks!
> --
> Christophe
>



-- 
busbey

Reply via email to