Advanced Topics

Multi-Cloud Setup

Deploy Cassandra across AWS, GCP, Azure - active-active, disaster recovery, and hybrid cloud architectures!

☁️ Why Multi-Cloud Cassandra?

The Multi-Cloud Challenge 🌍

Your SaaS company serves customers globally:

🌎 Users in US-East, EU-West, Asia-Pacific
⚡ Need < 50ms latency for all regions
🔥 Must survive entire cloud provider outage
💰 Want to optimize costs across providers
📜 Compliance requires data sovereignty (GDPR)
🎯 Avoid vendor lock-in

Solution: Deploy Cassandra across multiple cloud providers for global availability, low latency, and resilience!

Multi-Cloud Benefits

Multi-cloud = Running infrastructure across multiple cloud providers (AWS, GCP, Azure) as a unified system.

Key Advantages:

🌍 Global reach: Deploy close to users worldwide
🛡️ High availability: Survive provider-level outages
⚡ Low latency: Serve from nearest datacenter
💰 Cost optimization: Use cheapest resources per region
🔒 Data sovereignty: Keep data in required countries
🚫 Avoid lock-in: Not dependent on single vendor
🎯 Best-of-breed: Use each cloud's strengths

Use Cases

🌍

Global Applications

Netflix (worldwide streaming)
Uber (multi-region dispatch)
Gaming platforms
Social networks
E-commerce

🔥

Disaster Recovery

Financial services
Healthcare systems
Government
Critical infrastructure
Enterprise apps

📜

Compliance

GDPR (EU data residency)
HIPAA (US healthcare)
PCI-DSS (payment data)
Data localization laws
Industry regulations

💡

Hybrid Cloud

On-prem + cloud
Edge computing
Migration strategy
Legacy integration
Cost optimization

🏗️ Multi-Cloud Architectures

Architecture 1: Active-Active (Multi-Region)

ACTIVE-ACTIVE ARCHITECTURE ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ │ AWS US-EAST │◄───►│ GCP EU-WEST │◄───►│ AZURE ASIA │ │ Cassandra DC1 │ │ Cassandra DC2 │ │ Cassandra DC3 │ │ RF=3, 6 nodes │ │ RF=3, 6 nodes │ │ RF=3, 6 nodes │ └─────────────────┘ └─────────────────┘ └─────────────────┘ ▲ ▲ ▲ │ │ │ US Users EU Users Asia Users ✅ All DCs accept reads AND writes ✅ Data replicated globally ✅ Users connect to nearest DC (low latency) ✅ Survives entire DC/provider failure

Active-Active Configuration

# cassandra.yaml (each datacenter)

# AWS US-EAST nodes:
cluster_name: 'GlobalCassandra'
endpoint_snitch: GossipingPropertyFileSnitch
seed_provider:
  - class_name: org.apache.cassandra.locator.SimpleSeedProvider
    parameters:
      - seeds: "aws-seed1,gcp-seed1,azure-seed1"

# cassandra-rackdc.properties (AWS nodes)
dc=aws_us_east
rack=rack1

# GCP EU-WEST nodes:
dc=gcp_eu_west
rack=rack1

# Azure Asia nodes:
dc=azure_asia_pacific
rack=rack1

# Create keyspace with replication across all DCs
CREATE KEYSPACE global_app
WITH REPLICATION = {
  'class': 'NetworkTopologyStrategy',
  'aws_us_east': 3,
  'gcp_eu_west': 3,
  'azure_asia_pacific': 3
};

-- Total: 9 copies of data (3 per DC)

Architecture 2: Active-Passive (Disaster Recovery)

ACTIVE-PASSIVE ARCHITECTURE ┌─────────────────┐ ┌─────────────────┐ │ AWS PRIMARY │ │ AZURE STANDBY │ │ Cassandra DC1 │───── async ──────►│ Cassandra DC2 │ │ RF=3, 9 nodes │ replication │ RF=3, 6 nodes │ └─────────────────┘ └─────────────────┘ ▲ ▲ │ │ ALL Traffic Backup only (no client traffic) Primary DC: Serves all production traffic Standby DC: Receives async replication, ready for failover Failover: Manual switch if primary fails

# Active-Passive with async replication

CREATE KEYSPACE production
WITH REPLICATION = {
  'class': 'NetworkTopologyStrategy',
  'aws_primary': 3,      ← Active
  'azure_standby': 3     ← Passive backup
};

-- Client connects ONLY to aws_primary normally
-- In disaster: Switch clients to azure_standby

Architecture 3: Hybrid Cloud (On-Prem + Cloud)

HYBRID ARCHITECTURE ┌──────────────────┐ ┌─────────────────┐ │ ON-PREMISES │ │ AWS CLOUD │ │ Cassandra DC1 │◄────────►│ Cassandra DC2 │ │ Legacy systems │ VPN │ New workloads │ │ Compliance data │ │ Elastic scale │ └──────────────────┘ └─────────────────┘ Use case: Gradual cloud migration Benefit: Keep sensitive data on-prem, scale in cloud

⚙️ Multi-Cloud Setup Guide

Step 1: Network Configuration

Critical: Cross-Cloud Networking

Challenge: Cassandra nodes across different clouds must communicate reliably.

Solutions:

🔗 VPN: Site-to-site VPN between clouds
🌐 Public IPs: Nodes communicate over internet (cheaper but less secure)
⚡ Direct Connect: AWS Direct Connect + Azure ExpressRoute + GCP Interconnect
🔒 VPC Peering: Cloud-native peering (limited cross-provider support)

AWS Setup (DC1)

# 1. Create VPC
aws ec2 create-vpc --cidr-block 10.0.0.0/16 --region us-east-1

# 2. Launch EC2 instances (Cassandra nodes)
aws ec2 run-instances \
  --image-id ami-12345 \
  --instance-type r5.2xlarge \
  --count 6 \
  --subnet-id subnet-123 \
  --security-group-ids sg-456

# 3. Security group: Allow Cassandra ports from all DCs
# - 7000 (inter-node)
# - 7001 (TLS inter-node)
# - 9042 (CQL)
# - 7199 (JMX)

# 4. Assign Elastic IPs (for cross-cloud communication)
aws ec2 allocate-address --domain vpc

GCP Setup (DC2)

# 1. Create VPC
gcloud compute networks create cassandra-net \
  --subnet-mode=custom

gcloud compute networks subnets create cassandra-subnet \
  --network=cassandra-net \
  --range=10.1.0.0/16 \
  --region=europe-west1

# 2. Create VM instances
gcloud compute instances create cassandra-node-{1..6} \
  --zone=europe-west1-b \
  --machine-type=n2-highmem-8 \
  --subnet=cassandra-subnet

# 3. Firewall rules
gcloud compute firewall-rules create cassandra-internal \
  --network=cassandra-net \
  --allow=tcp:7000,tcp:7001,tcp:9042

Azure Setup (DC3)

# 1. Create Resource Group and VNet
az group create --name cassandra-rg --location southeastasia

az network vnet create \
  --resource-group cassandra-rg \
  --name cassandra-vnet \
  --address-prefix 10.2.0.0/16 \
  --subnet-name cassandra-subnet \
  --subnet-prefix 10.2.0.0/24

# 2. Create VMs
az vm create \
  --resource-group cassandra-rg \
  --name cassandra-node-1 \
  --image UbuntuLTS \
  --size Standard_E8s_v3 \
  --vnet-name cassandra-vnet \
  --subnet cassandra-subnet

# 3. Network Security Group
az network nsg rule create \
  --resource-group cassandra-rg \
  --nsg-name cassandra-nsg \
  --name AllowCassandra \
  --priority 100 \
  --destination-port-ranges 7000 7001 9042

Step 2: Install Cassandra

# Run on ALL nodes (AWS, GCP, Azure)

# 1. Install Java 11
sudo apt update
sudo apt install openjdk-11-jdk -y

# 2. Add Cassandra repository
echo "deb https://debian.cassandra.apache.org 41x main" | \
  sudo tee /etc/apt/sources.list.d/cassandra.list

# 3. Install Cassandra
sudo apt update
sudo apt install cassandra -y

# 4. Stop service (configure first)
sudo systemctl stop cassandra

Step 3: Configure Each Datacenter

# Edit /etc/cassandra/cassandra.yaml on EACH node

# Common settings (all nodes):
cluster_name: 'GlobalCassandra'
num_tokens: 16
endpoint_snitch: GossipingPropertyFileSnitch

# Seeds: Include one from each DC
seed_provider:
  - class_name: org.apache.cassandra.locator.SimpleSeedProvider
    parameters:
      - seeds: "52.1.2.3,35.4.5.6,13.7.8.9"
               # AWS     GCP      Azure

# Listen addresses (use PUBLIC IP for cross-cloud)
listen_address: # Node's private IP
broadcast_address: # Node's public IP (for cross-cloud)

# Edit /etc/cassandra/cassandra-rackdc.properties

# AWS nodes:
dc=aws_us_east
rack=rack1

# GCP nodes:
dc=gcp_eu_west
rack=rack1

# Azure nodes:
dc=azure_asia
rack=rack1

Step 4: Start Cluster

# Start nodes ONE AT A TIME, waiting for each to join

# 1. Start seed nodes first (one per DC)
sudo systemctl start cassandra

# 2. Check if node joined
nodetool status
# Should show: UN (Up/Normal)

# 3. Wait 2 minutes, then start next node

# 4. Repeat until all nodes running

# 5. Verify 3 datacenters
nodetool status
# Output:
# Datacenter: aws_us_east (6 nodes)
# Datacenter: gcp_eu_west (6 nodes)
# Datacenter: azure_asia (6 nodes)

Step 5: Create Multi-DC Keyspace

CREATE KEYSPACE global_app
WITH REPLICATION = {
  'class': 'NetworkTopologyStrategy',
  'aws_us_east': 3,
  'gcp_eu_west': 3,
  'azure_asia': 3
};

CREATE TABLE global_app.users (
    user_id uuid PRIMARY KEY,
    email text,
    username text,
    created_at timestamp
);

-- Data now replicated to ALL three clouds!

🌐 Cloud Provider Specifics

AWS Recommendations

☁️ AWS Best Practices

Instance: r5.2xlarge or r6i.2xlarge (memory-optimized)
Storage: EBS gp3 (3000 IOPS, 125 MB/s)
Network: Enhanced networking (SR-IOV)
AZs: Spread nodes across 3 availability zones
Security: VPC, security groups, IAM roles
Monitoring: CloudWatch metrics + custom JMX

GCP Recommendations

☁️ GCP Best Practices

Instance: n2-highmem-8 (8 vCPUs, 64GB RAM)
Storage: SSD persistent disks (pd-ssd)
Network: Premium tier networking
Zones: Spread across 3 zones
Security: VPC, firewall rules, service accounts
Monitoring: Cloud Monitoring + Stackdriver

Azure Recommendations

☁️ Azure Best Practices

Instance: Standard_E8s_v3 (8 vCPUs, 64GB)
Storage: Premium SSD (P30 or P40)
Network: Accelerated networking
AZs: Availability zones (3 zones)
Security: VNet, NSG, managed identity
Monitoring: Azure Monitor + Application Insights

🔥 Disaster Recovery Strategies

Strategy 1: Multi-DC Active-Active

Built-in DR with Active-Active

How it works: All datacenters accept traffic. If one DC fails, others automatically handle load.

# Client configuration with fallback

from cassandra.cluster import Cluster
from cassandra.policies import DCAwareRoundRobinPolicy

# Connect to local DC first, fallback to remote
cluster = Cluster(
    contact_points=['aws-node1', 'gcp-node1', 'azure-node1'],
    load_balancing_policy=DCAwareRoundRobinPolicy(
        local_dc='aws_us_east',
        used_hosts_per_remote_dc=3  ← Fallback to remote DCs
    )
)

# If AWS fails:
# - Driver automatically tries GCP
# - Then Azure if GCP also fails
# - Transparent to application!

RTO: Seconds (automatic)

RPO: Zero (synchronous replication)

Strategy 2: Regular Backups

# Automated backup script

# 1. Take snapshot on all nodes
nodetool snapshot -t backup-$(date +%Y%m%d)

# 2. Upload to S3 (from AWS), GCS (from GCP), Blob (from Azure)
aws s3 sync /var/lib/cassandra/data/ s3://my-backups/

# 3. Schedule daily with cron
0 2 * * * /scripts/backup-cassandra.sh

# 4. Retention: Keep 30 days, monthly archives for 1 year

Strategy 3: Cross-Region Replication

CROSS-REGION DR ARCHITECTURE Production Region: ┌────────────────────────────────────────┐ │ AWS US-EAST (3 DCs) │ │ DC1: us-east-1a, DC2: us-east-1b │ │ DC3: us-east-1c │ └────────────────────────────────────────┘ ↓ async replication DR Region (Standby): ┌────────────────────────────────────────┐ │ AWS EU-WEST (1 DC, cold standby) │ │ Receives async replication │ └────────────────────────────────────────┘ Failover: Promote EU-WEST to active if US-EAST entire region fails

Disaster Scenarios & Response

Scenario	Impact	Response	RTO
Single node fails	None ✅	Automatic - other replicas serve	< 1 sec
AZ/Zone fails	Minimal ✅	Other AZs handle traffic	< 30 sec
Entire DC fails	Degraded	Failover to other DCs	1-5 min
Cloud provider outage	Serious	Route to other cloud providers	5-15 min

✅ Multi-Cloud Best Practices

✅ DO These

Use NetworkTopologyStrategy
Configure LOCAL_QUORUM reads
Enable authentication & encryption
Monitor cross-DC latency
Use 3+ nodes per DC
Spread across availability zones
Test failover regularly
Automate backups
Document runbooks
Use consistent instance types

❌ DON'T Do These

Use SimpleStrategy (single DC only)
Run QUORUM across WAN
Mix different Cassandra versions
Ignore cross-cloud network costs
Deploy < 3 nodes per DC
Skip encryption over internet
Forget about latency
Use single seed per DC
Ignore time sync (NTP)
Deploy without monitoring

Performance Tips

🎯 LOCAL_QUORUM: Read/write within local DC only (fast)
⚡ Async replication: Cross-DC replication is async by default
💰 Data transfer costs: Cross-cloud egress expensive (~$0.09/GB)
🔒 Encryption: Use TLS for cross-cloud (adds latency)
⏰ NTP sync: Critical for distributed operations
📊 Monitoring: Track cross-DC replication lag

Cost Considerations

Multi-cloud costs more than single cloud:

💰 Compute: 3x instances (3 clouds × nodes per cloud)
💾 Storage: 3x data (replicated everywhere)
🌐 Network: Cross-cloud egress fees
👨‍💼 Operations: More complexity = more ops cost

Typical savings vs benefits:

✅ Cost of downtime > multi-cloud cost
✅ Vendor negotiation leverage
✅ Use spot/preemptible instances where possible

🎯 Multi-Cloud Summary

You now know how to deploy Cassandra globally!

📚 Key Takeaways:

☁️ Multi-cloud = AWS + GCP + Azure unified
🌍 Active-active for global low latency
🔥 Survives entire cloud provider outages
🔗 VPN or public IPs for cross-cloud
📊 NetworkTopologyStrategy with RF per DC
⚡ LOCAL_QUORUM for performance
🛡️ Built-in disaster recovery
💰 Consider egress costs

🎉 CONGRATULATIONS! 🎉

You've completed ALL advanced Cassandra topics!

🚀 You're now a Cassandra expert! 🚀

Responsive Ad