Amazon Web Services services
2 / 24
Amazon Web Services

Object Storage — Amazon S3

Store unlimited objects durably and cheaply, with tiered classes and fine-grained access.

S3 stores objects, not files. You PUT and GET by key inside a bucket, over HTTP.

A gigantic valet coat-check that never runs out of hooks. You hand over an item and get a ticket (the key); rarely claimed coats are quietly moved to a cheaper basement room (Glacier), and nothing is ever lost.

Key Concepts

1
    s3://my-bucket/images/2026/06/photo.jpg
                   ^^^^^^^^^^^^^^^^^^^^^^^  one key, not folders
2
There are no real directories. images/2026/ is a prefix in the key, which is why listing is a prefix scan and why renaming a "folder" means copying every object.
images/2026/
3
Durability is 11 nines, achieved by replicating each object across devices in several AZs. That is durability, not availability — the object will not be lost, which is a different promise from always being reachable.
4
Objects go up to 5 TB, with multipart upload required past 5 GB.
5
Storage classes are where the cost control lives.
    Standard              hot data, highest storage price
    Intelligent-Tiering   moves objects automatically by access
    Standard-IA           infrequent, cheaper storage + retrieval fee
    One Zone-IA           same, single AZ, for reproducible data
    Glacier Flexible      archive, minutes to hours to restore
    Deep Archive          cheapest, up to 12 hours
6
Lifecycle rules apply those automatically — Standard to IA at 30 days, Glacier at 90, expire at a year. On a bucket that only grows, this is usually the single largest saving available.
7
Security is layered, and the default is closed.
    Block Public Access   on by default, at account and bucket level
    bucket policy + IAM   who may do what
    SSE-S3 / SSE-KMS      encryption at rest, KMS when you need
                          key control and an audit trail
    aws:SecureTransport   deny plain HTTP
8
Versioning protects against the realistic disaster, which is a mistaken delete or overwrite rather than hardware failure. With MFA delete it also protects against a compromised key.
9
Strong read-after-write consistency has applied to all operations since 2020 — a PUT is immediately readable, which older material still gets wrong.
10
What the interviewer is probing.1. "Is S3 a filesystem?" Probing: the mental model, which drives several follow-ups. Stalls: "Yes, with folders." Moves up: no — it is a key-value store over HTTP; images/2026/ is a prefix in the key, which is why listing is a prefix scan and renaming a folder copies every object.
11
2. "What does 11 nines of durability actually promise?" Probing: durability versus availability. Stalls: "It never goes down." Moves up: that the object will not be lost, by replicating across devices and AZs — availability is a separate, lower number.
12
3. "How would you cut the cost of a bucket that only grows?" Probing: lifecycle rules. Stalls: "Delete old data." Moves up: lifecycle transitions — Standard to IA at 30 days, Glacier at 90, expire at a retention limit — or Intelligent-Tiering when access is unpredictable.
13
4. "How do you stop a bucket being publicly readable?" Probing: the layered controls. Stalls: "Set the ACL." Moves up: Block Public Access at account and bucket level, which overrides any policy or ACL, plus a bucket policy denying non-TLS access.
Use cases:s3object-storagelifecycledurability