This is pretty easy - no need for Sqoop. Add a Cascading Lingual step at the front of your job which executes a Redshift UNLOAD
command to S3:
UNLOAD ('select_statement')
TO 's3://object_path_prefix'
[ WITH ] CREDENTIALS [AS] 'aws_access_credentials'
[ option [ ... ] ]
Then you can either process the export directly on S3, or add an S3DistCp step to bring the data onto HDFS first.
This will be a lot more performant than adding Sqoop, and a lot simpler to maintain.