Request Replication for Microservice Fault Isolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In microservice architectures, it is challenging to isolate and identify faults when upgrading container images, as the continuous integration environment with changing container images makes it difficult to ensure compatibility and stability, leading to performance degradation and errors, such as seen in the online shopping website Web-Mart after upgrading features.
Innovation Solution
The system replicates user requests to both a stable production system with validated container images and modified production systems with unvalidated images, comparing their responses to identify faults and determine the best combination of container images for optimal performance without disrupting the original production system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If container images are continuously updated in a microservice architecture, then system functionality and features are improved, but system stability and reliability deteriorate due to compatibility issues and faults
Solution Approach 1:
The system segments the production environment into multiple isolated containers, each running a different version of container images. This allows simultaneous operation of stable and updated versions, enabling functionality improvement while maintaining stability through version isolation.
Solution Approach 2:
The system performs preliminary validation by dispatching replicated requests to containers with updated container images before full deployment. The fault detector compares responses from updated and stable versions in advance, identifying potential issues before they affect the entire system.
2Difficulty of detecting and measuring
If multiple versions of container images are tested simultaneously, then fault identification capability is improved, but system complexity increases due to multiple production systems
Solution Approach 1:
The system creates copies of the production system architecture, with each copy running a specific version of container images. These replicated systems allow parallel testing of multiple versions while maintaining a standardized structure, making fault detection easier without proportionally increasing overall system complexity.
Solution Approach 2:
The fault detector acts as an intermediary component that receives and compares responses from multiple production systems. This centralized mediation simplifies the complexity by providing a single point of comparison and fault identification, rather than requiring complex distributed monitoring across all systems.
3Measurement precision
If replicated requests are dispatched to multiple production systems, then validation accuracy is improved, but processing time increases due to parallel request handling
Solution Approach 1:
The system maintains continuous operation by having the stable production system continue serving requests while updated versions are validated in parallel. The fault detection process occurs continuously in the background, ensuring validation accuracy without causing service interruptions or significant delays.
Solution Approach 2:
The system applies partial validation by selectively dispatching replicated requests to updated production systems based on risk assessment and confidence levels. Not all requests are routed for validation, allowing the system to maintain high validation accuracy for critical operations while reducing processing overhead for routine operations.
Data Source
AI summary
A method for identifying and isolating faults in versioned microservices includes a request replicator receiving an original request, and determining whether to replicate the original request. The request replicator replicates the original request creating one or more replicated requests, including a first replicated request. In an example, the request replicator dispatches the original request to a stable production system, and dispatches the first replicated request to a first modified production system. The stable production system produces a first reply to the original request. The first modified production system produces a second reply to the first replicated request. A fault detector performs a comparison of the second reply and the first reply and determines, based on the comparison, that the first modified production system has a verification status. Then, the stable production system is replaced with first modified production system.