Monolithic Architecture

In monolithic architecture every component is on single instance.
The instance capacity = max capacity needed for an opration. E.g. If our app has A,B and C components and C requires max CPU (4 GB Ram) and B requires max storage(10 Gb storage), our instance should have minimum CPU that meets C's CPU requirement and B's storage requirement. i.e. min instance must be 4+ gb ram and 10+ gb storage.
That is not scalable because when we need to scale, we need to scale entire stack (4+ Gb Ram and 10+ Gb storage).
The option is tiered application, i.e. each component A, B and C live in their separate tiers. i.e. instance of A can have 1 Gb ram and 5 Gb storage, instance of B can have 1 Gb ram and 10 Gb Storage and instance of C can have 4 GB ram and 2 Gb storage.
Tiered instance must have one to one connection with each other and they need to be constants.
We can scale instances using Load balancers.
It's not possible to shut down every instance of one component even if the component is not used.

Event driven architecture

Compoents can have different requirements and communicate with each other using event buses.
Everything is an event, i.e. button click, API Calls, Object stored in Bucket etc…
Producers produce events and Consumers consume events.
Producers deliver messages to event buses and event buses deliver messages to consumers.
Consumers terminate themselves when there are no more messages to process.

Function as a service.
Event driven execution.
Piece of code in one language and one runtime.
Only billed for a duration of function runs.
Key component of Serverless architecture.
Container, that has function code, runtime environment
Uses execution role (IAM Role)
Invocation is event driven or manual.
When invoked, the code and additional info is downloaded in a compute hardware managed by aws.
Lambda can never run, run once or many times. Runs in one compute instance, runs in many paralleley.
Never store anything environment.
Public service by default.
They can access public resources by default.
They can be configured to be inside private VPC. When they're inside private VPC they can't access public resources.
Max 15 minutes execution time.
1Million Free request per month and 4,00,000 GB-Seconds of compute time per month are free.

Cloudwatch events is an old architecture, eventbridge is superset of cloudwatch and is recommended.
They both observe events (Eventbridge can accept events from third parties). And send messages to receivers.
Both have default event bus for an acount.
Cloudwatch event has only one event bus. Eventbridge can have additional buses.
Event rules that looks for a pattern defined by us and when the pattern is found, it delivered to the target. Pattern matching + schedule rules.
Rule recipients can have one or more targets.

Managed AWS Service.
Manages API Endpoint services.
Create, Publish, Monitor and Secure APIs.
Billed based on API Calls, Data Transfers and additional features such as caching.

Serverless is not a hardware acrhitecture, there is always a server running. But not all the components of app are hosted in single server/set of servers, they are all separated.
It's all about management. In serverless architecture, developers, admin and management care about 0 servers, sometimes a few when required.
App is collection of small & specialized functions.
Stateless and Ephemeral environments. Envs don't have their own state, and they are temporary esp they stop when they're done.
Billing is done as duration.
Event driven: Nothing is running unless it's required.
FaaS(Function as a service) is used where possible for compute.
Managed Service (Like S3 and RDS) when possible.

Highly available, secure, durable pub/sub messaging service.
Public AWS service. Network connectivity with public endpoint.
Co-ordinate sending and delivery of messages.
Messages <=256 KB
SNS Topics are base entity of SNS
SNS Topics have permissions and configuration of sending those messages.
Publisher sends a message to topic.
Topics have subscribers which receive messages.
Subscribers include: HTTP(S) endpoints, Emails, SQS, Mobile Push, SMS and Lambda.
Used across AWS for notification. Cloudwatch Alarms and CloudFormation
Offers Delivery Status- Some supported types like SQS, Lambda or HTTP(S) endpoints can send delivery status back to SNS.
Delivery retries.
Highly Available and Scalable services.
SNS is regionally resiliant.
Capable of Server side encryption.
Cross account access via topic policy.

Step Functions lets us create state machines.
State machines as service. State Machines have start, steps and end.
Each step is capable for receiving a data, do work on that data and output a data.
By using step function a lambda can increase it's running duration.
Standard workflow and Express workflow.
Maximum duration is 1 year for Standard, 5 minutes for Express
Started via API Gateway, IOT Rules, Eventbridge, Lambda etc…
States are defined as Amazon State Language (ASL). It's a JSON Template.
State machines assume IAM Role for permissions.
States:
- Succeed & Fail
- Wait: Wait for certain period of time and waits till certain point in time.
- Choice: Different path based on input.
- Parallel: Multiple execution at same time.
- Map: Accepts list, and executes something based on item of that list.
- Task: Allows the state machine to actually perform task.
  - Can be integrated with different AWS Services.

Managed message queues.
Standard and FIFO queues.
In FIFO Queue messages are receievd in order.
In Standard Queue messages can be received out of order.
Message size max 256 KB.
Received messages are hidden for a period of time (Visibility time out).
After visibility timeout if client hasn't deleted the message from queue, this message can re-appear at later point of time or retried immediately.
Dead Letter Queues, message fails multiple times can be moved to separate queue.
ASGs can be scaled using length of queue
In standard queue message can be delivered atleast once.
In FIFO queue message can be delivered exactly once.
FIFO 3000 messages per second with batching and 300 messages per second without.
Billed based on Request, message received by SQS via sender.
Short polling and Long polling.
- Short polling logs a request and pulls the queue immidiately. Even if there is no messages we're billed.
- In long polling we can wait for a period (Upto 20 seconds), if messages are available they will be delivered, if not they will wait.
Encryption at rest is supported using KMS.
Encryption in transit (Between SQS and clients) is supported by default.
Identity policy can be used for accessing Queue from same account.
Queue Policy can only be used to allow access to different account.

Kinesis is a scalable streaming service.
Desined to ingest a lot of data from lot of producers.
Producers send data to Kinesis Stream.
Streams can scale from low to infinite data rates.
Public service and highly available by design.
Streams store 24 Hour moving window of data.
- Any data which is older than 24 hours old will be replaced by new data.
Multiple consumer access data from that moving window with granular time control.
- i.e. one consumer can access real time data, one can access one data per hour.
Kinesis start with one shard. And based on performance requirements can add multiple shard.
Shard Capacity is 1 MB/Second Ingestion and 2 MB/Second Consumption.
More shard = more cost.
Another costing factor is window. 24 hr is default and given and can be increased upto 7 days. More storage window is required, more cost is there.
Data is stored in Kinesis Data Record which have maximum 1MB size.
Data Firehose can be used to pass Kinesis data into other AWS Service like S3.