[HDDS-1713] ReplicationManager fail to find proper node topology based on Datanode details from heartbeat

• Type: Sub-task
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 0.4.1
• Labels: pull-request-available
• Target Version/s: 0.4.1
• Assignee: Sammi Chen
• Reporter: Xiaoyu Yao
• Created: 20/Jun/19 18:35
• Updated: 05/Aug/19 14:52
• Resolved: 19/Jul/19 22:48

Description

DN does not have the topology info included in its heartbeat message for container report/pipeline report.

SCM is where the topology information is available. During the processing of heartbeat, we should not rely on the datanodedetails from report to choose datanodes for close container. Otherwise, all the datanode locations of existing container replicas will fallback to /default-rack.

 

The fix is to retrieve the corresponding datanode locations from scm nodemanager, which has authoritative network topology information. 

Issue Links

• is duplicated by: HDDS-1770 SCM crashes when ReplicationManager is trying to re-replicate under replicated containers
• is duplicated by: HDDS-1824 IllegalArgumentException in NetworkTopologyImpl causes SCM to shutdown
• links to: GitHub Pull Request #1008
• links to: GitHub Pull Request #1112