All containers launched on a single node
I’m using HDP2.1 (HDFS & YARN services) on a 4-node cluster (1xNN, 3xDN)
Trying to run TestDFSIO benchmarks I noticed all of the containers (100 in total) are being launched on a single node even though I’ve got two other perfectly working NodeManagers. This obviously gives very bad performance.
Browsing through the DFS I saw all control files (/benchmarks/TestDFSIO/io_control/*) were replicated successfully on 3 different datanodes.
The cluster is healthy according to all Ambari checks, 3/3 DN alive, 3/3 NM alive, all seems just fine.
What can cause this kind of behavior and how can I fix it?