Infrastructure & Observability

Architecture Overview

RawDrive uses a modern cloud-native infrastructure stack:

Key Files

Traefik Configuration

Rate Limiting Middleware

# infrastructure/docker/traefik/dynamic.yaml
http:
  middlewares:
    rate-limit-api:
      rateLimit:
        average: 50    # Requests per second
        burst: 100     # Burst allowance
        period: 1s
        sourceCriterion:
          ipStrategy:
            depth: 1

    rate-limit-uploads:
      rateLimit:
        average: 10    # Lower for uploads
        burst: 20
        period: 1s

Routing Rules

# API routing with middleware
http:
  routers:
    api-router:
      rule: "Host(`api.rawdrive.ai`) || PathPrefix(`/api`)"
      entryPoints:
        - websecure
      service: backend-service
      middlewares:
        - rate-limit-api
        - security-headers
        - cors-headers
      tls:
        certResolver: letsencrypt

Prometheus Metrics

Traefik exposes metrics for KEDA autoscaling:

# traefik.yaml - Enable Prometheus metrics
metrics:
  prometheus:
    entryPoint: metrics
    addEntryPointsLabels: true
    addRoutersLabels: true
    addServicesLabels: true

KEDA Autoscaling

Backend Scaler (Traefik Metrics)

# infrastructure/kubernetes/base/keda/scaledobjects.yaml
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: rawdrive-backend-scaler
spec:
  scaleTargetRef:
    name: rawdrive-backend
  minReplicaCount: 2
  maxReplicaCount: 100
  triggers:
    # Scale on request rate
    - type: prometheus
      metadata:
        serverAddress: http://prometheus:9090
        threshold: "100"
        query: |
          sum(rate(traefik_service_requests_total{service=~"rawdrive-backend.*"}[1m]))

    # Scale on latency
    - type: prometheus
      metadata:
        threshold: "1"
        query: |
          histogram_quantile(0.95, 
            sum(rate(traefik_service_request_duration_seconds_bucket[5m])) by (le)
          )

Worker Scaler (Redis Queues)

triggers:
  - type: redis
    metadata:
      address: redis:6379
      listName: face_processing_queue
      listLength: "5"

Prometheus Alerting

Key Alerts

# Traefik high error rate
- alert: TraefikHighErrorRate
  expr: |
    sum(rate(traefik_service_requests_total{code=~"5.."}[5m]))
    / sum(rate(traefik_service_requests_total[5m])) > 0.05
  for: 2m
  labels:
    severity: critical

# KEDA not scaling
- alert: KEDAScaledObjectNotReady
  expr: keda_scaledobject_status{status!="True"} == 1
  for: 5m
  labels:
    severity: warning

PostgreSQL + pgvectorscale

Database Image

# docker-compose.yml - Use TimescaleDB image for full vector support
services:
  postgres:
    image: timescale/timescaledb-ha:pg16
    # Includes: pgvector, pgvectorscale, StreamingDiskANN

Vector Search

# Enable extensions
CREATE EXTENSION IF NOT EXISTS vector;
CREATE EXTENSION IF NOT EXISTS vectorscale;

# Create DiskANN index for large datasets
CREATE INDEX ON photos
USING diskann (embedding vector_cosine_ops)
WITH (num_neighbors = 50);

Index Selection

Docker Commands

# Start with Traefik (uses File Provider routing from dynamic.yaml)
docker compose -f docker-compose.yml up -d

# Access Traefik dashboard (dev only)
open http://localhost:8080           # Direct access
open http://traefik.localhost        # Via router

# View Traefik metrics
curl http://localhost:8082/metrics

# Note: docker-compose.traefik.yml is DEPRECATED and should NOT be used
# All routing is defined in infrastructure/docker/traefik/dynamic.yaml

Kubernetes Commands

# Install KEDA
helm install keda kedacore/keda -n keda --create-namespace

# Install Traefik CRDs
kubectl apply -f https://raw.githubusercontent.com/traefik/traefik/v3.0/docs/content/reference/dynamic-configuration/kubernetes-crd-definition-v1.yml

# Deploy infrastructure
kubectl apply -k infrastructure/kubernetes/base/

# Check KEDA ScaledObjects
kubectl get scaledobjects -n rawdrive

# Check HPA status (managed by KEDA)
kubectl get hpa -n rawdrive

Monitoring Checklist

• Traefik dashboard accessible (dev: :8080)
• Prometheus scraping Traefik metrics (:8082)
• Grafana dashboards showing traffic
• KEDA ScaledObjects in "Ready" state
• Alerts configured in Alertmanager
• Log aggregation via Loki working

Performance Targets