[SAMZA-1013] Add YARN Node label support

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.12.0
• Assignee: Unassigned
• Reporter: Maxim Logvinenko
• Created: 05/Sep/16 17:01
• Updated: 21/Oct/16 01:04
• Resolved: 21/Oct/16 01:04

Description

YARN Node labels were introduced in Hadoop version 2.6, which allows to group nodes with similar characteristics and allows applications to specify where to run. This patch adds support for YARN node labels in Samza.

In this implementation, node labels are defined directly in yarnConfig in YarnClusterResourceManager. It might be better to have node labels as a part of SamzaResourceRequest and SamzaResource classes, but org.apache.hadoop.yarn.api.records.Container class doesn't contain node label and hence we have nothing to pass to the SamzaResource constructor in onContainersAllocated method of YarnClusterResourceManager class.

RB Code review: https://reviews.apache.org/r/51633/