HadoopConceptsNote

Resource Request

  • Request resources of each container.
    • CPU
    • Memory
  • Locality constraints for the constainers.
    • According to Network topology
    • Allow application to specify locality.
  • Generally, a container processes an HDFS block.
  • Request resources at any time.
    • All requests at beginning.
      • Spark in this way.
    • Or requests resources dynamically to meet the changing needs.
  • In MapReduce,
    • Map tasks request containers up front,
    • Reduce tasks containers are not started until later.
    • If any tasks fail, additional container would be requested.

TODO

  1. How to know the hosts of all containers.