Incremental Snapshots for Fast Leader Node Bring-Up

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In distributed computing systems, bringing up a replacement leader node after failure or disconnection is computationally expensive due to the need for extensive key-value scan operations, which consumes significant resources and time, especially as data size and node numbers increase.

Innovation Solution

Implementing incremental snapshots and delta scans to quickly bring up a new leader node by retrieving only the data that has changed since the last snapshot, reducing the need for extensive key-value scan operations and minimizing resource consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If value-by-value consensus scanning is performed to bring up a replacement leader node, then data accuracy is ensured, but computational cost and time consumption increase enormously

Engineering Contradiction:
Improvedata accuracyVSAvoidleader node bring-up speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

Incremental snapshots are taken periodically before leader failure, capturing the state of key-value stores at specific points in time. This preliminary action allows the replacement leader node to start with a recent snapshot instead of performing value-by-value consensus scanning from scratch, significantly reducing bring-up time while maintaining data accuracy through the use of pre-captured consistent state data.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The key-value store is divided into incremental snapshots that capture only the changes since the last snapshot. This segmentation allows the replacement leader to process only the incremental changes rather than scanning all values, reducing computational cost while ensuring data accuracy by applying consensus only to the segmented incremental portions.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If full data scanning is performed on the replacement leader node, then up-to-date consensus values are achieved, but memory usage and CPU cycles are consumed significantly

Engineering Contradiction:
Improveconsensus value accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The invention extracts only the incremental changes from the full key-value store by using incremental snapshots. Instead of scanning and processing all values to achieve consensus, the system extracts only the changed portions since the last snapshot, reducing memory usage and CPU cycles while maintaining consensus value accuracy through targeted processing of extracted incremental data.

Inventive Principle:
Principle #2Taking out (Extraction)

3Stability of the object's composition

If extensive key-value scan operations are performed during leader node bring-up, then data consistency is verified, but network bandwidth is consumed heavily

Engineering Contradiction:
Improvedata consistencyVSAvoidnetwork bandwidth consumption
Core Design Contradiction:
Stability of the object's compositionVSLoss of energy

Solution Approach 1:

Incremental snapshots are pre-captured and stored before leader failure occurs. These snapshots contain the consistent state at specific points in time. When a replacement leader node is brought up, it can retrieve these pre-prepared snapshots instead of performing extensive real-time scanning and consensus operations, reducing network bandwidth consumption while maintaining data consistency through the use of pre-verified snapshot data.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS11768809B2Managing incremental snapshots for fast leader node bring-up
Publication Date: 2023.09.26 NUTANIX INC
  • US11768809B2 patent drawing
  • US11768809B2 patent drawing
  • US11768809B2 patent drawing

AI summary

Methods, systems and computer program products for computer system data management. A leader node that is in a leader-follower relationship with a follower node is deployed to manage metadata across a plurality of interconnected computing nodes of the computer system. Upon a failure of any sort that results in a loss of operation or a disconnection of the leader node, the followers undertake steps to bring up a new leader. Such steps to bring up a new leader include electing a new leader to replace the lost or disconnected leader and initializing the data state of the new leader node by causing the leader to request incremental snapshots from followers. Additional metadata that might have changed since closing the incremental snapshots can also be retrieved from the followers. The metadata that might have changed since the time of closing the incremental snapshot is subjected to verification using a consensus algorithm.