Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Not A Problem
-
2.0.3-alpha, 0.23.5
-
None
-
None
-
None
Description
As initially reported in MAPREDUCE-4801, the shuffle protocol allows a reducer to hangup the connection to a nodemanager after receiving the shuffle header. This can introduce a connection reset IOException or a ClosedChannelException on the nodemanager. Logging these is not very interesting, and adds a lot of data to the NM log.
MAPREDUCE-4801 missed some places where these exceptions can occur, will post sample exceptions below for reference.