High Availability (HA)

AWS provides 3 types of load balancers:
- Application load balancer (ALB): useful for balancing HTTP/HTTPS traffic, operate at Layer 7 and are application-aware;
- network load balancer: useful for balancing TCP traffic, operate at Layer 4 and can achieve extreme performance;
- Classic load balancer: legacy ELBs, can operate at Layer 7 (can include features such as X-Forwarded and sticky sessions, but not application-aware) and Layer 4.
ELBs in VPC support IPv4 addresses only, while ELBs in EC2-classic support both IPv4 and IPv6.
ELB returns 504 Gateway Timeout when either app server or db server is down.
X-Forwarded-For can be used to identify the actual source IP address of the requests.
ELBs use health check to determine whether traffic should be forwarded to a certain server.
ELBs always have a DNS name (i.e., domain name), but do not have an IP address.
Sticky session allows to bind a user's session to a specific EC2 instance.
- If you enable sticky session for ALBs, the traffic will only be sent at the target group level.
It is possible to enable cross-zone load balancing on ELBs.
- If you do not enable cross-zone LB, you can ask Route53 to split the traffic to different AZs and have one ELB per AZ as well.
It is also possible to create a listen with rules to forward traffic based on the path pattern of URLs, which is also known as path-based routing.
- This is useful for building your own API gateway for microservices.

HA Architecture Principles

Everything fails, thus, you should plan for failure.
There must be redundant resources for every level.
- Different regions, different AZs, different subsets, different instances (EC2, RDS).
To prepare for scalability, we can do either horizontal scalability (scale out) or veritical scalability (scale up).
We have to always seek balance between HA and cost effectiveness.

CloudFormation is a way of completely scripting a set of cloud resources.
- Other infrastructure-as-code (IaC) solutions like Terraform may be better.
CloudFormation also provides some pre-configured templates called Quickstarts.

Elastic Beanstalk helps to quickly deploy applications on AWS without being aware of the underlying resources used.
Elastic Beanstalk is a good starting point, but definitely not a good engineering practice at scale.