[ARTEMIS-3344] Embedding Zookeeper for Pluggable Quorum Vote

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Assignee: Francesco Nigro
• Reporter: Francesco Nigro
• Created: 14/Jun/21 09:05
• Updated: 14/Jun/21 09:05

Description

One of the nicer things of the Atomix RI, last minute dropped because of lack of OSS community, related to the single pair use case, was the ability to achieve a sufficient number of nodes in order to prevent split-brain just by adding a "witness" node (without being a full-fat broker) to the existing primary/backup ones.

It was possible because each broker node was already an active member of the quorum and able to participate to quorum votes.
This was saving users to configure any external service and just care about brokers and their reachability, on topology setup.

Providing a similar behavior for Apache Curator/Zookeeper quorum vote RI means:

• exposing some hook/map all the relevant configuration bits of Zookeeper
• consider the performance/stability hit and memory requirements changes while embedding a potentially "noisy" neighbor (network I/O, disk I/O, CPU and java heap/memory-wise)

There are few ways to achieve embedding ZK:

1. as a proper quorum service in the broker
2. by impl some plugin that treat it as an external service to be run along the broker

The latter approach is way less invasive because users can just implement such plugin deciding what/which bits configure and expose.

Issue Links

• is related to: ARTEMIS-2716 Implements pluggable Quorum Vote