Opscraft
AWS core services · Lesson 3 of 8 · 3 min read

Storage

S3

  • Storage classes: Standard, Intelligent-Tiering, Standard-IA, One Zone-IA, Glacier Instant Retrieval, Glacier Flexible Retrieval, Glacier Deep Archive; lifecycle rules transition and expire objects (and old versions, incomplete multipart uploads).
  • Consistency: strong read-after-write for all operations.
  • Versioning (protects against overwrite/delete), MFA delete, Object Lock (WORM; governance/compliance modes) for ransomware-proof backups and compliance.
  • Security: Block Public Access on at account level; bucket policies + IAM; ACLs disabled (Object Ownership: bucket owner enforced); encryption at rest by default (SSE-S3), SSE-KMS for key control and audit (use S3 Bucket Keys to cut KMS costs); enforce TLS with aws:SecureTransport; restrict to VPC endpoint with aws:SourceVpce.
  • Replication: CRR/SRR (needs versioning) for DR/compliance; Replication Time Control for SLA.
  • Performance: thousands of requests/second per prefix; multipart upload for large objects; Transfer Acceleration; byte-range fetches.
  • Static sites: S3 + CloudFront with Origin Access Control (bucket stays private).
  • Events: to Lambda/SQS/SNS/EventBridge.
Enlarge diagramStandardhot dataIntelligent-Tieringunknown patternStandard-IAmonthly accessGlacier Instantquarterly, ms accessGlacier Flexibleminutes to hoursDeep Archive12+ hours, cheapestleft to right, top to bottom: cheaper to store,slower and costlier to read
Figure 1. S3 storage classes trade storage price against access cost and speed. Lifecycle rules move objects down the ladder automatically.
Swipe sideways to see all of it, or tap to open it full screen.
Enlarge diagramusersCloudFrontTLS (us-east-1cert), caching,WAFS3 bucketprivate · BlockPublic AccessonHTTPSOrigin Access Control
Figure 2. A static site with S3 and CloudFront: the bucket stays private and only CloudFront, through Origin Access Control, may read it.
Swipe sideways to see all of it, or tap to open it full screen.

EBS and EFS

EBS EFS FSx
Type Block volume Managed NFS Managed Windows/Lustre/NetApp/OpenZFS
Scope One AZ, typically one instance (multi-attach for io1/io2 only) Regional, many clients across AZs Varies
K8s ReadWriteOnce via EBS CSI ReadWriteMany via EFS CSI Specialised
Notes gp3 (baseline 3,000 IOPS/125 MB/s, provision more independently), io2 for high IOPS; snapshots to S3, cross-region copy; encryption Scales automatically; throughput modes; lifecycle to IA HPC, Windows shares

EBS gotcha in EKS: a pod with an EBS-backed PVC can only be scheduled in the volume's AZ — use WaitForFirstConsumer storage classes and keep StatefulSets spread properly.

Enlarge diagramAZ AAZ BnodeEBS volumepgdatanodepod cannot use the volume hereEFS is regional and shared, so it can be mounted from every AZ.
Figure 3. An EBS volume lives in one AZ, so a pod using it can only run on nodes in that AZ. WaitForFirstConsumer creates the volume where the pod lands.
Swipe sideways to see all of it, or tap to open it full screen.