AWS S3 Notes
S3 Bucket, Object, Encryption
- Object-based storage is spread across multiple devices and facilities. Storage that can be scaled indefinitely.
- File vault
- It must have a globally unique name.
- Defined at the Region level
- There is a naming convention
- no capital letters
- not underscore
- 3-63 characters
- No IP
- Must start with lower_case_letter or number
Objects have keys.
means full path
key consists of prefix + object name
There is no concept of a directory. (The UI is made of keys like directories)
The object value is the body of the content.
- Max Object Size : 5TB(5000GB)
- If you are uploading an object larger than 5GB, you should use a
Metadata (system metadata)
- Versioning of files inside S3 is possible.
- Available at the bucket level
- If you overwrite the same key, the version is incremented.
- Protection from accidental deletion
- Easy to roll back
- Files uploaded with versioning off are version null.
- When you remove versioning from a bucket, you don’t get rid of the old version, it removes the versioning for the new file.
- If you delete a file with versioning turned on, the Delete Marker will only be attached. (The previous version remains as is)
- There are 4 encryption methods
- SSE-S3: Encrypting objects using AWS-managed keys
- SSE-KMS: AWS Key Management Service 사용
- SSE-C: Unique Encryption Key
- Client Side Encryption: Encrypted by the client
Server side encryption
Encryption is managed by Amazon s3.
Uses AES-256 encryption
must be set in the header
Server side encryption
KMS handles encryption keys
Advantages of KMS: User Control + Tracking
must be set in the header.
Encrypt with key managed by Customer
S3 does not store encryption key
must be used unconditionally.
- The client proceeds with encryption
- Encrypted before sending to S3
- Responsibility for decryption is also on the client side.
S3 Security, Websites, CORS
- User based
- IAM Policies: which API calls should be allowed
- Resource based
- Bucket Policies
- Object Access Control List (ACL)
- Bucket Access Control List (ACL)
- Note: IAM policy can access s3 object
- If the user has permission or the resource policy is ALLOW
- If there is no explicit DENY
S3 Bucket Policies
- JSON based
- How to define policy via JSON
- Fully defined buckets and objects
- Allow / Deny
Security - Other
- Logging and Audit
- S3 Access Logs are stored in different buckets
- Author of API calls to AWS CloudTrail
- User Security
- MFA Delete : Use MFA to delete
- Pre-Signed URLs: URLs that are only valid for a limited time
- S3 can host a static website and can make it accessible from www
- website URL is
- If 403 is displayed, you can set the bucket policy to allow for the public read.
- Cross Origin Resource Sharing
- By default, web browsers allow only the same origin and block requests to hostnames of different origins.
- Same origin
- Another Origin (Cross Origin)
- If the correct CORS header is not found, the web browser blocks the request.
- CORS Header (ex: Access-Control-Allow-Origin) must be set.
- When a client sends a cross-origin request to our s3 bucket, we need to enable the correct CORS header.
S3 Consistency Model
- From 12/2020
- Maintain Consistency (visible immediately)
- read/write consistency
- list consistency
S3 Advanced 1
S3 MFA Delete
- Regulate Behavior Using Multi-Factor Authentication (MFA)
- You must enable versioning to use it.
- If you need MFA
- Permanently delete an object
- Suspending bucket versioning
- When MFA is not required
- versioning enable
- deleted version listing
- Only the bucket owner can set it.
- Not from the console, CLI only, root account only.
S3 Default Encryption vs Bucket policy
- If you want to force encryption, you can set a bucket policy to prevent it if there is no encryption header.
- Another way is to use the s3 default encryption option.
- Bucket policy is computed before default encryption
S3 access log
- To log all requests to access a bucket for auditing purposes.
- Stored in another bucket
- May be analyzed using Athena or other analysis tools.
- Do not use the same monitoring bucket and logging bucket. (Logging loop - huge capacity)
- When you want to copy a bucket to a bucket in another region.
- Versioning of source and destination must be enabled
- Cross Region Replication (CRR)
- Same Region Replication (SRR)
- Buckets can be in different accounts.
- Replication is asynchronous
- Appropriate IAM permissions must be granted to s3
- CRR: Regulatory Compliance, Faster Data Access Times
- SRR: log aggregation, replication of production_test data
- Only new objects after being activated are duplicated.
- DELETE case
- Delete marker can also be duplicated
- Deleting a version is not replicated
- No chain replication
- When bucket 1 is replicated to bucket 2 and bucket 2 is replicated to bucket 3
- Bucket 1 was not replicated in Bucket 3
S3 Presigned URLs
- How to temporarily approve upload to S3 or download from S3
S3 Advanced 2
S3 storage class
- Amazon S3 Standard - General Purpose
- Amazon S3 Standard-Infrequent Access(IA)
- Amazon S3 One Zone-Infrequent Access
- Amazon S3 Intelligent Tiering
- Amazon Glacier
- Amazon Glacier Deep Archive
- Amazon S3 Reduced Redundancy Storage (deprecated)
S3 Standard - General Purpose
- Durability, no object loss, availability: 99.9999999%
- Can withstand two dysfunctions at the same time
- Very general use
S3 Standard - Infrequent Access(IA)
- Data that is accessed infrequently but needs to be accessed quickly
- Less expensive than S3 Standard
- For disaster recovery, backup, and storage of unused data
S3 One Zone - Infrequent Access(IA)
- The IA function is the same, and it is stored in only one AZ.
- Availability is a bit low, but latency is low and high throughput can be expected.
- Data is lost when AZ is blown up
- Supports encryption.
- 20% cheaper than IA
- Used to save backup files, image thumbnails, etc.
S3 Intelligent Tiering
- Low latency, high performance same as S3 standard
- There is a tiering fee along with a small monitoring fee every month.
- Tiering is the storage of high performance / low performance according to usage.
- Data movement occurs automatically between universal S3 and S3 IA
low-cost object storage
For backup / archiving
For storage for a very long period of time (10 years)
An alternative to magnetic tape storage
It’s very cheap at $0.004 per GB, but it comes at a retrieval cost.
Each item is
called and can store up to 40TB.
Archives are stored in a safe called, not a bucket.
AWS Glacier & Glacier Deep Archive
- 3 recovery options
- Expedited (1~5분)
- Standard (3~5 hours)
- Bulk (5-12 hours)
- Minimum storage period: 90 days
- Deep Archive: Storage for a longer period of time
- Standard (12 hours)
- Bulk (48 hours)
- Minimum storage period: 180 days
S3 Lifecycle Rules
- Objects can be moved between storage classes
- Frequently accessed objects are STANDARD_IA
- Objects that do not need real-time are GLACIER or DEEP_ARCHIVE
- Although it is possible to move directly, it is also possible to move automatically using lifecycle constructs.
- Transition actions: define when an object will be moved to another storage class
- Transfer to Standard IA class after 60 days
- Move to Glacier class after 6 months
- Expiration actions: Deleting an object after a period of time has elapsed.
- Access logs can be set to be cleared after 365 days
- Can be used to delete old versions of files
- Can be used to clear incomplete multi-part uploads
- Rules can be created with specific prefixes
- Rules can be defined for specific tags
S3 Lifecycle Rule Scenario 1
- When the application creates a user profile, it is uploaded to s3.
- These thumbnails can be easily regenerated and need to be kept for 45 days.
- Original images must be able to be restored immediately for 45 days, and after 45 days, you can wait up to 6 hours.
=> Keep the original S3 in standard and send it to GLACIER after 45 days. => Thumbnails are placed in ONEZONE_IA and deleted after 45 days (because re-creation is possible)
S3 Lifecycle Rule Scenario 2
- Recover deleted s3 immediately for 15 days
- Objects deleted for up to 1 year can be restored within 48 hours
=> S3 versioning => If it is not the current version, move it to S3_IA (restore immediately) => After 15 days, move it to DEEP_ARCHIVE (can be restored within 48 hours)
S3 Analysis - Storage Class Analysis
- You can set up s3 analytics to decide when to send objects from standard to standard_ia.
- Only works with standard -> standard_ia
- It takes 24-48 hours to activate for the first time
- Good for improving lifecycle rules
S3 - Baseline Performance
- Amazon S3 automatically scales to handle a very large number of requests.
- The latency is very short.
- 3500 put/copy/post/delete, 5500 get per (second, prefix)
- object path => prefix
- bucket/folder1/sub1/file => /folder1/sub1/
- bucket/folder1/sub2/file => /folder1/sub2/
- bucket/1/file => /1/
- bucket/2/file => /2/
S3 - KMS Restrictions
- When using SSE-KMS, it is affected by KMS limit
- When uploading, the KMS API called GenerateDataKey is called.
- When downloading, the KMS API called Decrypt is called.
- KMS has a quota limit per second
- You can request a quota increase through the service quotas console.
S3 - Performance
- Multipart Upload
- 100 MB or more per file
- Used for files larger than 5GB
- Can be uploaded at the same time
- S3 Transfer Acceleration
- Move files to a nearby edge location
- Faster transfers from edge locations to buckets over the private network
- Compatible with split uploads
S3 Performance - S3 Byte-Range Fetches
- When requesting a GET request, a specific object is divided into small byte units in a byte range.
- request in parallel
- Provides better recovery from failures (requests only partial data)