[HDFS-16691] Use quorum instead of requiring full JN set for NN format

• Type: Improvement
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.4.0, 3.3.5
• Component/s: namenode

Environment

Demonstrated in a Kubernetes environment running Java 11.  Using an HA configuration:

1. Start new cluster, but short 1 JN (minimum quorum, and the missing JN won’t resolve). VERIFY:

• NN formats the 2 existing JN and stabilizes
• Messages show sync between JN-0 and JN-1, and NN -> JN

1. Scale JN stateful set to add missing JN.  NOTE: Requires HDFS-16690

• Target Version/s: 3.3.9
• Assignee: Steve Vaughan
• Reporter: Steve Vaughan
• Created: 25/Jul/22 16:23
• Updated: 29/Nov/22 15:51

Description

Currently a format request fails if any of the JournalNodes is unresolvable.  For dynamic cluster environments where a JournalNode may not be available during the initial formatting step but JournalNodes can self-heal, it makes sense to allow the format to succeed when a quorum of JournalNodes is available.

Issue Links

• is a child of: HADOOP-18396 Issues running in dynamic / managed environments
• is blocked by: HDFS-16690 Automatically format new unformatted JournalNodes using JournalNodeSyncer