Speaking as a user, if you are suggesting that you will retain the
functionality but move the methods to an optional dependency, it makes sense to
me. We have used the Hadoop integration for AvroParquetInputFormat and
CqlBulkOutputFormat in Flink (although we won't be using CqlBulkOutputFormat
any longer because it doesn't seem to be reliable).
From: Fabian Hueske <fhue...@gmail.com<mailto:fhue...@gmail.com>>
Date: Friday, October 14, 2016 at 4:29 AM
Subject: [DISCUSS] Deprecate Hadoop source method from (batch)
I would like to propose to deprecate the utility methods to read data with
Hadoop InputFormats from the (batch) ExecutionEnvironment.
The motivation for deprecating these methods is reduce Flink's dependency on
Hadoop but rather have Hadoop as an optional dependency for users that actually
need it (HDFS, MapRed-Compat, ...). Eventually, we want to have Flink
distribution that does not have a hard Hadoop dependency.
One step for this is to remove the Hadoop dependency from flink-java (Flink's
Java DataSet API) which is currently required due to the above utility methods
(see FLINK-4315). We recently received a PR that addresses FLINK-4315 and
removes the Hadoop methods from the ExecutionEnvironment. After some
discussion, it was decided to defer the PR to Flink 2.0 because it breaks the
API (these methods are delared @PublicEvolving).
I propose to accept this PR for Flink 1.2, but instead of removing the methods
This would help to migrate old code and prevent new usage of these methods.
For a later Flink release (1.3 or 2.0) we could remove these methods and the
Hadoop dependency on flink-java.
What do others think?