Separate the result reader out of the client class which sets up connection.

It is sufficient to have only one interface connection to read any number of
results. Creating a new client interface connection every time for RPC creates
new ByteBuffers for network layer communication which leads to memory leak on
the CC there by leading OutOfMemory exception on the CC. Separating the connection
and reader out and making a connection only once and asking it to create the
reader for every job resolves this problem and also the most sensible solution.

git-svn-id: https://hyracks.googlecode.com/svn/branches/fullstack_hyracks_result_distribution@3054 123451ca-8445-de46-9d55-352943316053
4 files changed
tree: 780ef915d49cbbe66c7f76f4cba621c6a97e4ef0
  1. algebricks/
  2. hivesterix/
  3. hyracks/
  4. pom.xml
  5. pregelix/