For the TOP, check out
https://issues.apache.org/jira/browse/MAPREDUCE-4574 which we fixed in
Hadoop recently to allow full reuse with Avro.
On Wed, May 1, 2013 at 10:49 AM, Rahul Bhattacharjee
<[EMAIL PROTECTED]> wrote:
> Hi Harsh,
> Looks like a lot of other classes are also to be rewritten for avro. Like
> the total sort partitioner , which I think currently assumes writable as the
> io mechanism.
> I faced problem using with avro , so though of writing to the forum.
> Thanks a lot
> On Wed, May 1, 2013 at 10:35 AM, Harsh J <[EMAIL PROTECTED]> wrote:
>> Moving the question to Apache Avro's user@ lists. Please use the right
>> lists for the most relevant answers.
>> Avro is a different serialization technique that intends to replace
>> the Writable serialization defaults in Hadoop. MR accepts a list of
>> serializers it can use for its key/value structures and isn't limited
>> to Writable in any way. Look up the property "io.serializations" in
>> your Hadoop's core-default.xml for more information.
>> The Avro project also offers fast comparator classes that are used for
>> comparing the bytes/structures of Avro objects. This is mostly
>> auto-set for you when you use the MR framework as described at
>> (via AvroJob helper class).
>> On Tue, Apr 30, 2013 at 6:51 PM, Rahul Bhattacharjee
>> <[EMAIL PROTECTED]> wrote:
>> > Hi,
>> > When dealing with Avro data files in MR jobs ,we use AvroMapper , I
>> > noticed
>> > that the output of K and V of AvroMapper isnt writable and neither the
>> > key
>> > is comparable (these are AvroKey and AvroValue). As the general
>> > serialization mechanism is writable , how is the K,V pairs in case of
>> > avro ,
>> > travel across nodes?
>> > Thanks,
>> > Rahul
>> Harsh J