Skip to content

Technical Architecture

This document catalogs the application topologies, system protocols, and API integrations orchestrating the microservices ecosystem. It focuses specifically on the interaction layers between logical components.

1. Interaction Mapping

The architecture implements a hybrid communication framework splitting operations into synchronous blocking calls during ingress validation, and asynchronous decoupled events for core fulfillment scaling.

Phase 1: Synchronous Ingress & Validation

When a frontend user initiates a cart checkout, the checkoutservice invokes real-time RPC calls across internal domains to establish state consistency before concluding the HTTP transaction.

  1. CRM Lookup: A synchronous retrieval of user account metadata executing explicitly via GET /customers across an internal network border to the CRM Service.
  2. Fraud Analytics: The orchestrator triggers an external transaction risk calculation physically spanning across Google Cloud to Microsoft Azure. An HTTP POST /metrics passes { transactionType, durationMs, success } payloads which natively returns an instantaneous boolean risk score.
  3. Inventory Reservation: To prevent race-condition overselling, the application interrogates the Inventory Service via a synchronous GET /inventory/{productId} request to explicitly verify physical stock capacity prior to persisting the cart event.

Phase 2: Asynchronous Event Distribution

Following successful synchronous validation, the checkoutservice finalizes the payment gate, permanently commits the user session, and returns an HTTP 200 explicitly to the client browser.

Simultaneously, the frontend tier publishes an OrderConfirmedEvent schema object onto a dedicated operational topic managed structurally by Google Cloud Pub/Sub.

Phase 3: Distributed Order Fulfillment

Pub/Sub performs horizontal scale-out payload distribution, routing immutable event messages globally to downstream backend workloads triggered via push subscriptions.

  1. Order Management Execution: Cloud Run environments running the OrderManagement Node.js process ingest the push event array dynamically and begin executing conditional logic routing.
  2. Accounting Ledger Sync: OrderManagement executes a subsequent synchronous POST /transactions specifically into the Accounting Service. Natively, the Accounting Service sequentially conducts a retroactive GET request specifically against the original CRM Service logic to map transaction identities firmly to loyalty databases.
  3. Warehouse Dispatching: The OrderManagement tier posts instructions utilizing POST /shipments directly to the Warehouse APIs. The Warehouse service logically triggers physical packing routines and initiates a final stock reconciliation backward mapping to the core Inventory Service utilizing a rigid PUT /inventory/{productId} structural edit.

Phase 4: Conversational Workflow (CRM Concierge)

Users interacting with the storefront can seamlessly engage an AI agent built on Vertex AI and the Google Agent Development Kit (ADK). By acting autonomously, the crm-agent issues explicit Model Context Protocol (MCP) tool requests mapped to the CRM REST API (e.g., executing backend lookups or parsing history metadata dynamically), empowering a deeply conversational, context-aware user support layer natively.

Phase 5: Deep Packet Inspection (DPI) Modeling

During all phases of traffic routing, a physical copy of the native network packets is explicitly intercepted via GCP Packet Mirroring. The packets are encapsulated directly via the Geneve protocol and streamed toward a cluster of Deep Packet Inspection (Traffic Collector) engines. Operating on Python, the collector immediately decapsulates the UDP 6081 framing to parse the actual inner IP and TCP headers, checking for string-based vulnerabilities natively (e.g. SQLi, XSS) displaying instantaneous analytical records on its dedicated monitoring dashboard.

2. API & Component Registry

The table below catalogs the deployment profiles, core structural protocols, and data mechanisms for each discrete application function.

Service Name Compute Target Language Framework Core Endpoints State Pattern
Checkout GCP GKE Go / Python Internal Integrator Stateless (Redis Cache)
Event Broker GCP Pub/Sub Managed PaaS Publish, Subscribe Managed Queues
OrderManagement GCP Cloud Run Node.js (Express) Triggered via POST / Stateless Event Processor
Apigee API Google Cloud API SaaS Matrix Reverse Proxying / Auth Token Caches
Partner API GCP Cloud Run Node.js (TypeScript) POST /tracking, GET /catalog Serverless Scale
CRM Concierge GCP Vertex AI TypeScript (ADK) POST /v1/chat LLM Session
CRM GCP Compute VM Node.js (Express) GET /customers, POST Persistent (Local SQLite/CloudSQL)
Inventory GCP Compute VM Node.js (Express) GET /inventory, PUT Operational Memory / Disk
Warehouse GCP Cloud Run Node.js (Express) POST /shipments Stateless Computing
Accounting GCP Cloud Run Node.js (Express) GET, POST, PUT Persistent (Local Memory / Disk)
Fraud Engine Azure Virtual Machine Node.js (Express) POST /metrics, GET Transitory Memory Cache
Data Warehouse GCP BigQuery Managed PaaS Direct Streaming Columnar Disk
Traffic Collector GCP Compute VM Python (Flask / Raw Socket) UDP 6081 (Geneve) Volatile SQLite (DPI)