Resilient Architectures
Multi AZ in AWS
- Services where multi AZ must be enabled manually:
- EFS, ELB, ASG, Elastic Beanstalk
- RDS, OpenSearch: multi AZ used for failovers
- Aurora:
- Data is stored automatically across multi AZ
- It can have multi AZ for DB itself (multi master, read replicas, multi AZ deployments)
- OpenSearch (managed): multi master
- Jenkins (self-hosted): multi master
- Services where multi az is implicitly enabled:
- S3 (except OneZone-IA)
- DynamoDB: data is replicated by default
- All of AWS proprietary managed services
EBS - Multi AZ
- EBS is tied to a single AZ
- How can we make EBS multi AZ?
- We crate an ASG with 1 min/max/desired capacity
- We create a lifecycle hook for
Terminate: make a snapshot of the EBS volume
- We create a lifecycle hook fot
Start: copy the snapshot, create an EBS, attach it to the instance
- Note: for PIOPS volumes (IO1), to get max performance after snapshot, read the entire volume once (pre warming of IO blocks)
Blue-Green Architecture
- Blue-Green with ALB and Route53:
- Blue-Green with AWS Lambda:
Multi Region Services
- DynamoDB Global Tables: multi wau replication, enabled by Streams
- AWS Config Aggregators: multi region and multi account aggregated view in one account
- RDS Cross Region Read Replicas: used for Read and disaster recovery
- Aurora Global Database: one region is master, other is for read and disaster recovery
- EBS volumes snapshots, AMI, RDS snapshots can be copied to other regions
- VPC peering to allow private traffic between regions
- Route 53 uses a global network of DNS servers
- S3 Cross Region Replication
- CloudFront: CDN at the Edge Locations
- Lambda@Edge for Global Lambda function at Edge Location (A/B testing)
Multi Region with Route 53
- Routing to multiple regions: latency, geo-location, geo-proximity routing
- Health Check: we can have DNS failover to different regions
- Health Check types:
- Health checks monitoring an endpoint: application server, other AWS resource
- Health checks that monitor other health checks (calculated health checks)
- Health check that monitor CloudWatch alarms: example throttles of DynamoDB, custom metrics, etc.
- Create, update or delete stacks across multiple accounts and regions with a single operation
- Administrator account is allowed to create StackSets
- Trusted accounts are allowed to create, update and delete instances from StackSets
- When we update a StackSet in the administrator account, all the associated stack instances are updated through all accounts and regions
- Ability to set a maximum concurrent actions on targets (# or %)
- Ability to set failure tolerance (# or %)
Multi Region CodePipeline