AWS cho AI/Agent Developers — Day 5: Multi-Region Agent Routing với Route53

Agent chạy một region. User ở khắp nơi. Latency giết trải nghiệm — LLM call 2s từ us-east-1 ổn, nhưng từ Đông Nam Á thành 4s. Thêm DNS, TLS, network hops → mất user.

Hôm nay build multi-region routing:

Route53 latency-based — DNS resolve tới region gần user nhất
Active-passive failover — us-east-1 die → eu-west-1 tự tiếp quản
CloudFront — cache agent responses tại 450+ edge locations
Global Accelerator — TCP/UDP agents không xài được DNS routing

1
                    Route53
2
                   /        \
3
           us-east-1      eu-west-1
4
           (Primary)      (Secondary)
5
              │                │
6
           ALB ─── DynamoDB Global Tables ─── ALB
7
              │                                │
8
           ECS Fargate                    ECS Fargate

Step 1: Deploy Agent ở 2 Region#

Copy infrastructure từ Days 1-4, deploy lên us-east-1 và eu-west-1.

1
# us-east-1 (primary)
2
aws ecs create-service --cluster mcp-server-cluster --region us-east-1 ...
3

4
# eu-west-1 (secondary)
5
aws ecs create-service --cluster mcp-server-cluster --region eu-west-1 ...

State: DynamoDB Global Tables từ Day 2 tự replicate. Cache: ElastiCache là regional — khi failover, session đọc từ DynamoDB (chậm hơn tí nhưng vẫn chạy).

Step 2: Route53 Latency-Based#

1
# Tạo record latency-based cho us-east-1 (PRIMARY)
2
aws route53 change-resource-record-sets \
3
  --hosted-zone-id ZONE_ID \
4
  --change-batch '{
5
    "Changes": [
6
      {
7
        "Action": "CREATE",
8
        "ResourceRecordSet": {
9
          "Name": "agent.yourdomain.com.",
10
          "Type": "A",
11
          "SetIdentifier": "us-east-1",
12
          "Region": "us-east-1",
13
          "Failover": "PRIMARY",
14
          "AliasTarget": {
15
            "HostedZoneId": "<alb-zone-id>",
16
            "DNSName": "<alb-dns>",
17
            "EvaluateTargetHealth": true
18
          },
19
          "HealthCheckId": "<health-check-us>"
20
        }
21
      },
22
      {
23
        "Action": "CREATE",
24
        "ResourceRecordSet": {
25
          "Name": "agent.yourdomain.com.",
26
          "Type": "A",
27
          "SetIdentifier": "eu-west-1",
28
          "Region": "eu-west-1",
29
          "Failover": "SECONDARY",
30
          "AliasTarget": {
31
            "HostedZoneId": "<alb-zone-id>",
32
            "DNSName": "<alb-dns>",
33
            "EvaluateTargetHealth": true
34
          },
35
          "HealthCheckId": "<health-check-eu>"
36
        }
37
      }
38
    ]
39
  }'

Health Checks#

1
aws route53 create-health-check \
2
  --health-check-config '{
3
    "Type": "HTTPS",
4
    "FullyQualifiedDomainName": "<alb-dns-us-east-1>",
5
    "Port": 443,
6
    "ResourcePath": "/health",
7
    "RequestInterval": 30,
8
    "FailureThreshold": 3
9
  }'

3 lần fail → Route53 chuyển traffic sang eu-west-1 trong ~90s.

Step 3: CloudFront Edge Caching#

1
aws cloudfront create-distribution \
2
  --origin-domain-name "agent.yourdomain.com" \
3
  --default-cache-behavior '{
4
    "ViewerProtocolPolicy": "redirect-to-https",
5
    "MinTTL": 60,
6
    "DefaultTTL": 300,
7
    "MaxTTL": 3600
8
  }'

Cache ở edge: docs, FAQ, tool schemas, health checks. Không cache: user-specific prompts, write operations, WebSocket.

Step 4: Global Accelerator cho TCP/UDP#

DNS failover mất 30-60s (TTL). Global Accelerator dùng Anycast IP—<1s failover.

1
aws globalaccelerator create-accelerator --name agent-global-accelerator
2

3
# Tạo listener port 443 TCP
4
aws globalaccelerator create-listener \
5
  --accelerator-arn <arn> \
6
  --port-ranges '[{"FromPort":443,"ToPort":443}]' \
7
  --protocol TCP
8

9
# Add us-east-1 và eu-west-1 làm endpoints
10
aws globalaccelerator create-endpoint-group \
11
  --listener-arn <arn> \
12
  --endpoint-group-region us-east-1 \
13
  --endpoint-configurations '[{"EndpointId":"<alb-arn>","Weight":100}]'

	Route53	Global Accelerator
Failover	30-60s	<1s
Protocol	HTTP/HTTPS	TCP/UDP + HTTP
Cost	~$7/mo	~$18/mo

Step 5: Client-Side Region Detection#

1
const client = {
2
  async detectRegion() {
3
    const regions = ["us-east-1", "eu-west-1", "ap-southeast-1"];
4
    const latencies = await Promise.all(regions.map(async r => {
5
      const start = Date.now();
6
      await fetch(`https://agent-${r}.yourdomain.com/ping`);
7
      return { region: r, latency: Date.now() - start };
8
    }));
9
    return latencies.sort((a,b) => a.latency - b.latency)[0].region;
10
  },
11

12
  async ask(prompt: string) {
13
    const region = await this.detectRegion();
14
    return fetch(`https://agent-${region}.yourdomain.com/agent`, {
15
      method: "POST",
16
      body: JSON.stringify({ prompt, sessionId: this.sessionId, currentRegion: region }),
17
    });
18
  },
19
};

Chi phí DR#

Component	Monthly
Second region (ECS + ALB)	~$55
Global Tables replication	~$20-40
Route53 + health checks	~$7
CloudFront	~$5
Global Accelerator (optional)	~$18
Tổng	~$87-125/tháng

So với single region ~ $69/mo → ×1.5-2×. Đáng nếu downtime cost >$ 4/ngày.

Checklist#

Day	Chủ đề
1	Deploy MCP Server lên ECS Fargate ✅
2	Agent State với DynamoDB Global Tables ✅
3	LLM Caching với ElastiCache + Bedrock ✅
4	Serverless Agent với Lambda + Bedrock ✅
5	Multi-Region Agent Routing với Route53 ✅
6	CI/CD cho AI Agents với CodePipeline

Series: AWS cho AI/Agent Developers. Day 5: Multi-region routing — Route53 latency/failover, CloudFront edge caching, Global Accelerator.