Polo Legacy — Overview & Architecture

Polo v1 is a serverless infrastructure observability and cost analytics platform for Marqo Cloud, built as a Python/HTMX application running on AWS Lambda + API Gateway.

Tech Stack

Layer	Technology
Runtime	Python 3.12 on AWS Lambda (256 MB, 30s timeout)
Web	AWS Lambda Powertools APIGatewayRestResolver (prod), Flask (dev)
Rendering	htmxido (Python DSL for HTML generation) + HTMX
Styling	Tailwind CSS 3.4 with custom Marqo palette
JS	jQuery 3.7, DataTables, Highcharts, GoJS (diagrams), Hyperscript
Data Store	DynamoDB (ResourceTable + ReportTable)
Sync	Python collectors via Lambda, multi-account AWS API calls
Deploy	AWS SAM + CDK, Tailwind CLI for CSS
Auth	External auth proxy (/auth/logout redirect)

High-Level Architecture

                    ┌──────────────────────────┐
                    │   EventBridge Schedules   │
                    └──────┬───┬───┬───┬───────┘
                           │   │   │   │
              ┌────────────┘   │   │   └────────────┐
              ▼                ▼   ▼                ▼
        lambda_sync      lambda_report  lambda_cop  lambda_prune
        (collect AWS     (generate cost  (Slack     (purge deleted
         resources)       & action        alerts)    resources)
                          reports)
              │                │
              ▼                ▼
        ┌──────────────────────────┐
        │   DynamoDB               │
        │   - ResourceTable        │
        │   - ReportTable          │
        └──────────┬───────────────┘
                   │
                   ▼
        lambda_function (API Gateway)
              │
              ├── /data/*        → resource browser
              ├── /stats/*       → analytics (mostly stubs)
              ├── /actions/*     → cleanup recommendations
              ├── /dashboards/*  → cost reports & series
              ├── /budget/*      → budget tracking
              ├── /diagram/*     → infrastructure visualization
              └── /query/*       → freeform data queries

Lambda Functions

Function	Trigger	Purpose
lambda_function	API Gateway (HTTP)	Main web UI — routes, templates, data serving
lambda_static	API Gateway (/static)	Serves CSS, JS, images with MIME detection
lambda_sync	EventBridge schedule	Collects resources from 16 AWS accounts
lambda_report	EventBridge schedule	Generates cost, budget, and action reports
lambda_cop	EventBridge schedule	Detects infrastructure issues, posts to Slack
lambda_prune	EventBridge schedule	Deletes resources marked deleted_at > 1 day

AWS Accounts (16)

Account ID	Name	Environment
023568249301	controller	prod
651774330118	prod	prod
905418443936	marqtune_prod	prod
014498650341	models	prod
468036072962	staging	dev
339712831429	preprod	dev
707042731317	core	dev
424082663841	open_source	dev
010928202142	preprod_controller	dev
975050354766	marqtune	dev
504304931539	s2ui	dev
992382409372	polo	dev
311141537729	escalator	dev
311141526841	coach	dev
780949682512	commercial	dev
940994029740	ml	dev

Key Design Decisions

1. Server-side rendering with HTMX — no client-side framework. All HTML is generated in Python via htmxido, with HTMX handling partial page updates. This keeps the frontend extremely lightweight but tightly couples UI to the Python backend.

2. DynamoDB as primary store — single-table design with type (PK) and path (SK). Hierarchical paths encode resource relationships (e.g., A#{aws_id}/V#{vpc_id}/S#{subnet_id}/I#{instance_id}). Simple but limits ad-hoc querying — hence the /query page.

3. Multi-account sync via role assumption — a single Lambda assumes roles into 16 target AWS accounts, fetching resources in parallel with ThreadPoolExecutor.

4. Local dev parity — LocalDataService reads/writes a local_data.json file, mirroring the DynamoDB DataService interface. Devs run against cached data without AWS credentials.

5. Report pre-computation — cost breakdowns and action recommendations are pre-computed by lambda_report and stored in ReportTable, avoiding expensive on-the-fly calculations.

6. Cost tracking baseline — a "golden date" (June 21, 2025) serves as the cost baseline. All budget comparisons reference this date. Monthly target: $79,167 (~$2,639/day).