[SAA + SAP] 13. S3

SAA

• Max object size 5000 GB / 5 TB
• Uploading more thatn 5 GB, then multi part upload

• root account for MFA
• MFA delete need to use CLI

• Must enable versioning
• IAM permission
• After activating, only new objects are repliacated
• For Delete operations:
  • Can replicate delete markers from source to target (optional setting)
  • Deletions with a version ID are not repliacted (to avoid malicious deletes)
• There is no "chaning" of replication
  • if bucket 1 has replication into bucket 2, which has replication into bukcet 3
  • Then objects created in bucket 1 are not replicated to bucket 3

 

• Glacier under 90 days and 12 hours retriveal
• Deep archive under 180 days and 48 hours teriveal

• Thumbnail can be easily recreated, so can be saved into OneZone IA
• After 45 days should be retrive up to 6 hours.. then should use Glacier

• Recover deleted objects .. Enable versioning first
• Once enabled versioning, the "noncurrent version" can be saved into S3_IA
• 365d, 48h means DEEP ARCHIVE

• Help to analytic how many days should transfer to IA or Glacier

• If using KMS, the request is over 5500,10000,30000/s based on region, you need to increase Quotas

• Good for download large files
• Only get part of file

• Get less data (necessary data)
• SQL server-side filtering

 

S3 Select is NOT equal to Byte-range fetches. 

S3 select: Less data by SQL

Byte-range fetches: first amount of byte data of the file

• Enabling versioning for every successful write.

• For compliance

---

SAP

Anti patterns

• Lots of small files: because you need to pay more
• POSIX file system (use EFS instead), file locks
• Search features, queries, rapidly chaning data
• Website with dynamic content

Replication

Why need repliation?

• reduce latency
• disaster recovery
• security

S3- CloudWatch Events

• By default, CloudTrail records S3 bucket-level API calls
• CloudTrail logs for object-level can be enabled
• This helps us generate events for object-level API (GetObject, PutObject, PutObjectAcl,....)

• Organize the prefix can get better S3 performance

S3 Solution Architecture

1. Just EC2 with instance store: you might lost the data
2. CloudFront + EC2 + EBS: once EC2 fail, CloudFront might report failure if there is no cache
3. CloudFront + ALB + ASG + EFS: need to use EFS for cross AZs, expensive
4. CloudFront + S3: only static content is preferred, not dynamic content

As long as size is under 5.5TB, 1TB~3000IOPS

EBS cannot share content cross multi-AZ in runtime

Requester pay: then it requires players also have AWS account .. against public accessable to the game content, B - 4

Infrequence access which will cause company pay more menoy on S3. against lowest possible AWS Cost, C - 3

A: which is not scalable, A- 2

D - 1