Associated Datadog Illuminated episode: From Monolith to Microservices, Powered by LLMs
Datadog WebApp Monolith
- thousands of HTTP routes
- need to be migrated to Services Framework
Goal: Decommission Datadog monolith! Rewriting and/or getting rid of
Technologies: Python, Claude Code, GPT5, LangGraph, Temporal
flowchart TD
A["① Gather + Plan\n──────────────\ninput: current monolith docs,\nservices framework docs,\ninstructions (prompt),\ndestination repo\n──────────────\noutput: PLAN\n[Claude Code]"] --> B
B["② DO THE THING! Execute the Plan\n──────────────\n[Claude Code]\ninput: prompts to write code,\nprompts for test, organization, etc.\naccess to monolith + Services Framework\n──────────────\noutput: CODE!!!! + tests\n(probably not good yet)"] --> VP1
subgraph VP1["Verification Process"]
C["③ Run typechecker + fix issues"] --> D
D["④ Run tests + fix issues"] --> E
E["⑤ Start Server + fix issues"]
E -- fails --> C
D -- fails --> C
end
VP1 --> F["⑥ Code review by GPT5\n──────────────\ninput: code diff, prompts,\ngoal of migration,\naccess to monolith +\nservices framework\n──────────────\noutput: 1 of 2 things:\n• PASSES -or-\n• discrepancies"]
F -- passes --> K
F -- discrepancies --> G
G["⑦ Address discrepancies\n──────────────\n[Claude Code]\ninput: context, prompts, discrepancies\n──────────────\noutput: Code + tests"] --> VP2
subgraph VP2["Verification Process"]
H["⑧ run typechecker + fix issues"] --> I
I["⑨ run tests + fix issues"] --> J
J["⑩ start server + fix issues"]
I -- fails --> H
J -- fails --> H
end
VP2 --> K["⑪ open PR"]
K --> L["⑫ deploy to staging"]
L --> M["⑬ HUMAN REVIEW"]
M --> N["MERGED & DEPLOYED!!! HUZZAH!!! WOO HOO!!!"]
- not always passing integration testing
- How to measure success? + consistency
- modernize while migration
- exploratory + validating THIS WORKFLOW HAS GREAT POTENTIAL
- continuous improvement
- Workflow has run successfully
