Cloud diagram

Cloud Architecture Diagram Generator

Generate professional cloud architecture diagrams using the Python diagrams library (mingrammer/diagrams) with Graphviz rendering. Outputs PNG images with official icons for Azure, AWS, GCP, Kubernetes, on-premises, and 10+ other providers.

Prerequisites

The script needs two dependencies. Install them at runtime if missing:

pip install diagrams --break-system-packages -q
apt-get install -y graphviz -q 2>/dev/null || true

Workflow

Step 1: Understand the Architecture

Before writing any code, clarify what the user wants:

• Which cloud provider(s)? Azure, AWS, GCP, Kubernetes, on-prem, or a multi-cloud mix
• What type of diagram? Architecture overview, network topology, landing zone, data flow, CI/CD pipeline
• Which services? Map user descriptions to specific node classes for the chosen provider
• How complex? Simple (3-5 nodes), medium (6-15 nodes), complex (15+ nodes with multiple clusters)
• Layout direction? Left-to-right (LR) for wide diagrams, top-to-bottom (TB) for tall ones

If the user is vague (e.g., "draw me a web app architecture"), make reasonable assumptions and generate a solid default -- don't over-ask. You can always iterate.

Step 2: Map Services to Node Classes

Consult references/<provider>-nodes.md for the full node list per provider. Below are the most common mappings.

Azure (from diagrams.azure.<module> import <Class>)

Compute: AppServices, FunctionApps, KubernetesServices (alias AKS), VM, VMScaleSet (alias VMSS), ContainerInstances, ContainerApps, BatchAccounts

Networking: VirtualNetworks, ApplicationGateway, LoadBalancers, Firewall, FrontDoors, CDNProfiles, DNSZones, ExpressrouteCircuits, VirtualNetworkGateways, TrafficManagerProfiles, Subnets, PrivateEndpoint, PublicIpAddresses

Database: SQLDatabases, CosmosDb, CacheForRedis, DatabaseForPostgresqlServers, DatabaseForMysqlServers

Storage: BlobStorage, StorageAccounts, DataLakeStorage, QueueStorage, FileStorage

Security: KeyVaults, Sentinel, SecurityCenter

Identity: ActiveDirectory, ManagedIdentities, ConditionalAccess

Integration: ServiceBus, EventGridTopics, LogicApps, APIManagement

DevOps/Monitoring: ApplicationInsights, Pipelines, Repos, AzureDevops

AI/ML: CognitiveServices, MachineLearning, AzureOpenai

AWS (from diagrams.aws.<module> import <Class>)

Compute: EC2, ECS, EKS, Lambda, Fargate, Batch, ElasticBeanstalk

Networking: ELB, ALB, NLB, CloudFront, Route53, VPC, APIGateway, DirectConnect

Database: RDS, Aurora, DynamoDB, ElastiCache, Redshift, Neptune, DocumentDB

Storage: S3, EFS, EBS, FSx, Glacier

Security: WAF, Shield, IAMRole, IAM, KMS, SecretsManager, Cognito

Integration: SQS, SNS, Kinesis, StepFunctions, EventBridge

Management: Cloudwatch, Cloudtrail, Config, SystemsManager

GCP (from diagrams.gcp.<module> import <Class>)

Compute: Run, Functions, GKE, ComputeEngine, AppEngine

Networking: CDN, DNS, LoadBalancing, Armor, VPC

Database: SQL, Spanner, Bigtable, Firestore, Memorystore

Storage: GCS, Filestore, PersistentDisk

Analytics: BigQuery, Dataflow, Dataproc, PubSub, Composer

Kubernetes (from diagrams.k8s.<module> import <Class>)

Compute: Pod, Deployment, ReplicaSet, StatefulSet, DaemonSet, Job, CronJob

Networking: Ingress, Service, NetworkPolicy

Storage: PV, PVC, StorageClass

Other: HPA, Namespace, ConfigMap, Secret, ServiceAccount

On-Premises / Open Source (from diagrams.onprem.<module> import <Class>)

Compute: Docker, Nomad

Database: PostgreSQL, MySQL, MongoDB, Redis, Cassandra, ClickHouse, Elasticsearch

Network: Nginx, HAProxy, Traefik, Envoy, Istio, Kong

Queue: Kafka, RabbitMQ, Celery

Monitoring: Grafana, Prometheus, Datadog, Splunk

CI/CD: Jenkins, GitlabCI, GithubActions, ArgoCD

Step 3: Write the Diagram Code

Use this pattern:

from diagrams import Diagram, Cluster, Edge
# Import from any provider:
from diagrams.azure.<module> import <NodeClass>
from diagrams.aws.<module> import <NodeClass>
from diagrams.gcp.<module> import <NodeClass>
from diagrams.k8s.<module> import <NodeClass>
from diagrams.onprem.<module> import <NodeClass>

with Diagram("<Title>", show=False, filename="<output_name>", outformat="png", direction="LR"):
    # Optional: group related resources in Clusters
    with Cluster("Resource Group / VPC / Region"):
        node1 = NodeClass("Label")
        node2 = NodeClass("Label")

    # Connect with >> (data flow direction)
    node1 >> node2

    # Fan out to multiple targets
    node1 >> [node2, node3]

    # Custom edge labels and styles
    node1 >> Edge(label="HTTPS", color="blue") >> node2

Multi-provider example (mix providers freely in one diagram):

from diagrams.aws.network import CloudFront
from diagrams.azure.compute import AppServices
from diagrams.gcp.database import Spanner
from diagrams.onprem.monitoring import Grafana

# All of these can coexist in a single Diagram context

Key patterns:

• Clusters = Resource Groups, VPCs, VNets, Subnets, Regions, Namespaces, or any logical grouping
• Nested Clusters = VNet containing Subnets, Region containing Resource Groups
• Edge direction = >> flows left-to-right or top-to-bottom depending on direction
• Edge labels = Use Edge(label="protocol/description") for clarity
• Lists = source >> [target1, target2] fans out connections
• Bidirectional = Use node1 >> node2 and node2 >> node1 separately, or node1 - node2 for undirected

Direction guidelines:

• LR (left-to-right): Best for data flow, request paths, pipelines
• TB (top-to-bottom): Best for hierarchical designs, landing zones, network topology
• RL or BT: Rarely used but available

Diagram() parameters:

• show=False -- Always set this (don't open a viewer)
• filename -- Output path without extension
• outformat -- Use "png" (default) or "svg"
• direction -- "LR", "TB", "RL", "BT"
• graph_attr -- Dict of Graphviz attributes for fine-tuning (e.g., {"fontsize": "20", "bgcolor": "white", "pad": "0.5"})

Step 4: Generate and Present

1. Write the Python script to a temp file
2. Run it: python3 /tmp/cloud_diagram.py
3. The output PNG will be at the filename path
4. Present the image to the user

Step 5: Iterate

If the user wants changes:

• Add/remove services
• Change layout or grouping
• Adjust labels or edge descriptions
• Change output format
• Switch or add cloud providers

Modify the script and regenerate. Each run is fast (~2-5 seconds).

Common Architecture Patterns

Azure: Hub-Spoke Network

Hub VNet: Firewall, VPN Gateway, Bastion
  +-- Spoke 1: Web tier (App Gateway -> App Services)
  +-- Spoke 2: Data tier (SQL, Cosmos DB)
  +-- Spoke 3: AKS workloads

Azure: 3-Tier Web Application

Users -> Front Door/CDN -> App Gateway -> App Services/AKS -> SQL/Cosmos DB
                                                            -> Redis Cache
                                                            -> Blob Storage

Azure: Serverless Event-Driven

Event Source -> Event Grid/Service Bus -> Function Apps -> Cosmos DB
                                                        -> Blob Storage
                                      -> Logic Apps -> External APIs

AWS: 3-Tier Web Application

Route 53 -> CloudFront -> WAF -> ALB -> ECS/EC2 -> RDS (Multi-AZ)
                                                 -> ElastiCache
                                                 -> S3

AWS: Serverless

API Gateway -> Lambda -> DynamoDB
                      -> S3
            -> Step Functions -> Lambda -> SQS -> Lambda
EventBridge -> Lambda -> SNS -> Subscribers

AWS: Data Lake

S3 (raw) -> Glue ETL -> S3 (curated) -> Athena -> QuickSight
                                       -> Redshift Spectrum
Kinesis Data Streams -> Kinesis Firehose -> S3

GCP: Web Application

Cloud DNS -> Cloud CDN -> Cloud Load Balancing -> Cloud Run / GKE
                                               -> Cloud SQL / Spanner
                                               -> Memorystore
Cloud Armor (WAF) -|

GCP: Data Analytics

Pub/Sub -> Dataflow -> BigQuery -> Looker
                    -> Cloud Storage
Cloud Composer (orchestration)

Kubernetes: Microservices

Ingress -> Service A -> Pod (Deployment) -> PVC -> PV
        -> Service B -> Pod (StatefulSet) -> ConfigMap, Secret
HPA auto-scales Deployments
NetworkPolicy controls east-west traffic

Multi-Cloud: Active-Active

Global LB
  +-- AWS (us-east-1): CloudFront -> ECS -> RDS
  +-- Azure (East US): Front Door -> App Service -> Cosmos DB
  +-- GCP (us-central1): CDN -> Cloud Run -> Spanner
Cross-region replication between databases

Hybrid: On-Prem + Cloud

On-Prem: Nginx -> Docker containers -> PostgreSQL
  |-- VPN / ExpressRoute / Interconnect --|
Cloud: API Gateway -> Kubernetes -> Managed DB
Monitoring: Prometheus + Grafana (on-prem) collecting from both

Using Custom Icons

For services not in the diagrams library, use Custom nodes:

from diagrams.custom import Custom
from urllib.request import urlretrieve

icon_url = "https://example.com/my-service-icon.png"
icon_file = "custom_icon.png"
urlretrieve(icon_url, icon_file)

custom_node = Custom("My Service", icon_file)

Edge Styling Reference

Edge(label="HTTPS", color="blue", style="bold")
Edge(label="async", color="orange", style="dashed")
Edge(label="private link", color="darkgreen", style="dotted")
Edge(color="firebrick", style="bold")   # Error/alert path

Troubleshooting

Problem	Fix
ModuleNotFoundError: diagrams	pip install diagrams --break-system-packages
ExecutableNotFound: dot	apt-get install -y graphviz
Nodes overlapping	Add graph_attr={"nodesep": "1.0", "ranksep": "1.5"}
Diagram too wide/tall	Switch direction between LR and TB
Import error for a node	Check references/<provider>-nodes.md for the exact class name and module
Unknown provider module	Run python scripts/generate_node_refs.py --provider <name> --output-dir references/