Two-Node Cluster Failover via On-Demand Witness Arbitration

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Implementing disaster recovery capabilities in two-node computing clusters poses challenges, particularly in leader selection and failover processes, leading to suspended operations and the need for human intervention when communication between nodes fails.

Innovation Solution

A disaster recovery framework that includes an on-demand arbiter for leader selection, using periodic heartbeat monitoring and a witness node to form a quorum for electing a leader, allowing for automated failover and failback operations without additional computing resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If a leader selection protocol is implemented in a two-node cluster, then automated disaster recovery operations can be performed, but the system complexity increases and additional nodes are required to form a quorum

Engineering Contradiction:
Improveautomated failover operationsVSAvoidcluster configuration complexity
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

A witness node is introduced as an intermediary component that participates in quorum formation without requiring a full cluster node configuration. The witness node provides arbitration capabilities for leader selection while maintaining simpler system requirements compared to traditional multi-node quorum systems.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The witness node serves multiple functions: it acts as a quorum participant for leader election, provides arbitration during split-brain scenarios, and enables automated disaster recovery operations. This multi-functional approach reduces the need for dedicated components for each function.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If additional nodes are added to enable leader selection and quorum formation, then automated disaster recovery is possible, but the computing resources (memory, processing power, network bandwidth) increase

Engineering Contradiction:
Improvedisaster recovery capabilityVSAvoidcomputing resources
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The witness node is designed as a lightweight, minimal-resource component that provides essential quorum functionality without requiring full node resources. It consumes significantly less memory, processing power, and network bandwidth compared to complete cluster nodes, enabling disaster recovery capability with minimal resource overhead.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Reliability

If communication between nodes fails, then each node may consider the other failed and request to become leader, but this leads to suspended operations until human intervention

Engineering Contradiction:
Improvehigh availabilityVSAvoidoperational continuity
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The witness node provides continuous feedback and arbitration during communication failures. When nodes cannot communicate with each other, the witness node receives leadership requests from both nodes and provides deterministic feedback to select a single leader, preventing indefinite suspension of operations and eliminating the need for human intervention in split-brain scenarios.

Inventive Principle:
Principle #23Feedback

Data Source

PatentUS11194680B2Two node clusters recovery on a failure
Publication Date: 2021.12.07 NUTANIX INC
  • US11194680B2 patent drawing
  • US11194680B2 patent drawing
  • US11194680B2 patent drawing

AI summary

Systems and methods for high availability computing systems. Systems and methods include disaster recovery of two-node computing clusters. A method embodiment commences upon identifying a computing cluster having two nodes, the two nodes corresponding to a first node and a second node that each send and receive heartbeat indications periodically while performing storage I/O operations. One or both of the two nodes detect a heartbeat failure between the two nodes, and in response to detecting the heartbeat failure, one or both of the nodes temporarily cease storage I/O operations. A witness node is accessed in an on-demand basis as a result of detecting the heartbeat failure. The witness performs a leadership election operation to provide a leadership lock to only one requestor. The leader then resumes storage I/O operations and performs one or more disaster remediation operations. After remediation, the computing cluster is restored to a configuration having two nodes.