Recoleta
GitHub
Source note

Mono2Sls: Automated Monolith-to-Serverless Migration via Multi-Stage Pipeline with Static Analysis

Authors
Xingyan Chen, Yuxin Su, Zishan Su, Yang Yu, Zibin Zheng
Source
arXiv
Published
Collection
Software Intelligence
Serverless MigrationCode IntelligenceMulti Agent SystemsSoftware ModernizationAws Sam
Open arXivMarkdown source

Summary

Mono2Sls automates migration of monolithic Flask and Express backends into deployable AWS SAM serverless applications. It combines static code analysis with four tool-using LLM agents that plan the architecture, generate Lambda code, write SAM templates, and check consistency.

Problem

  • Moving a monolith to serverless requires aligned changes across API routing, Lambda boundaries, application code, IAM, DynamoDB, Cognito, SQS/EventBridge, and SAM templates.
  • Manual migration is slow and error-prone because a small mismatch between code and infrastructure can block deployment or break API behavior.
  • General code assistants often lack a stable cross-artifact contract for serverless migration, so handler code, routes, permissions, and template resources can drift apart.

Approach

  • Static analysis extracts HTTP entry points, file tags, cross-file call edges, async hints, and DynamoDB schema candidates into analysis_report.json.
  • An Architect agent turns those facts into blueprint.json, mapping business endpoints to Lambda functions and choosing Cognito, synchronous Lambda calls, SQS, or EventBridge where needed.
  • A Code Developer agent rewrites Flask/Express handlers into Lambda handlers, adapts identity to Cognito claims, removes global state, and adds SDK calls for inter-Lambda communication.
  • A SAM Engineer agent generates template.yaml with DynamoDB, Cognito, API Gateway, Lambda, layer, queue, and event resources, then runs cfn-lint through a validation tool.
  • A Consistency Validator runs 11 cross-artifact checks across generated code, SAM, and the blueprint, then applies fixes and re-validates.

Results

  • The benchmark covers 6 applications, 10,478 lines of code, 76 observable business endpoints, 24 DynamoDB tables, authentication in 6/6 apps, and async patterns in 4/6 apps.
  • Mono2Sls reports 100% deployment success without manual fixes across the 6 benchmark applications.
  • End-to-end correctness reaches 66.1%, compared with 53.7% to 61.2% for the commercial baselines.
  • API-coverage F1 reaches 98.7%, compared with 88.4% for the commercial baselines.
  • The ablation study reports that static-analysis-guided architecture planning adds 23.4 percentage points to end-to-end correctness.
  • The paper also claims more consistent use of AWS-native authentication and asynchronous patterns, though the excerpt does not provide the detailed per-pattern counts.

Link