S3 Storage
AWS SAA SAA-C03 · Design Resilient Architectures
S3: object storage at any scale
Amazon Simple Storage Service (S3) is object storage: you store files (objects) in containers called buckets, and each object is identified by a key within its bucket. Buckets are globally unique in name but live in one Region, and objects can range from zero bytes to 5 TB. S3 is designed for 99.999999999% (eleven nines) durability, meaning that on average you might lose one object in 10,000 years for every 10,000 objects stored — a figure achieved by storing data redundantly across multiple devices in multiple Availability Zones. Availability is a separate number and varies by storage class. The exam tests durability versus availability, and storage classes, more than any other S3 detail.
Every S3 object consists of data, a key, metadata, and a version ID if versioning is enabled. Objects are accessed over HTTP(S) through the S3 API, the console, or the CLI, and since December 2020 S3 has provided strong read-after-write consistency for all operations — once a write succeeds, any subsequent read sees the new data. The older eventual-consistency behavior for overwrites and deletes no longer applies, and the exam's current questions reflect that.
Storage classes: pay for the access pattern you have
S3 offers a ladder of storage classes trading storage price against retrieval price and speed:
| Storage class | Designed for | Availability | Notes |
|---|---|---|---|
| S3 Standard | Frequently accessed data | 99.99% | Default class; millisecond access, no retrieval fee |
| S3 Intelligent-Tiering | Data with unknown or changing access patterns | 99.9% | Moves objects between frequent and infrequent tiers automatically |
| S3 Standard-IA | Infrequently accessed, but needed quickly when accessed | 99.9% | Lower storage cost, retrieval fee, 30-day minimum |
| S3 One Zone-IA | Infrequently accessed, re-creatable data | 99.5% | Stored in a single AZ — cheaper, but lost if that AZ is destroyed |
| S3 Glacier Instant Retrieval | Archive data accessed quarterly, needs millisecond retrieval | 99.9% | 90-day minimum storage duration |
| S3 Glacier Flexible Retrieval | Archive accessed once or twice a year | 99.99% | Retrieval in minutes to hours; 90-day minimum |
| S3 Glacier Deep Archive | Long-term archive, rarely accessed | 99.99% | Lowest cost; retrieval in hours; 180-day minimum |
Glacier retrieval times are worth learning precisely, because the exam uses them to test whether a storage class fits a requirement. Glacier Flexible Retrieval offers Expedited (typically 1–5 minutes), Standard (typically 3–5 hours), and Bulk (typically 5–12 hours, and free). Glacier Deep Archive offers Standard (typically within 12 hours) and Bulk (typically within 48 hours); Expedited retrieval is not available for Deep Archive. If a question says archived data must be retrievable within an hour, Deep Archive fails and Flexible Retrieval with Expedited retrieval is the answer.
Lifecycle policies and cost control
A lifecycle policy automates moving objects between storage classes and eventually expiring them, based on rules you define — for example, move objects to Standard-IA after 30 days, to Glacier Flexible Retrieval after 90 days, and delete them after 7 years. Lifecycle transitions are the exam's standard answer for any scenario where data ages and access drops off. Note the minimum storage durations: moving an object out of Standard-IA before 30 days, or out of Glacier Flexible Retrieval before 90 days, still incurs the charge for the full minimum period. S3 Intelligent-Tiering is the alternative when you cannot predict access patterns: it monitors each object and moves it between frequent-access and infrequent-access tiers automatically, with optional archive tiers, for a small monthly monitoring fee per object.
Protecting data: versioning, replication, and encryption
Versioning keeps every version of an object, so an accidental overwrite or delete can be undone by restoring an earlier version. Once versioning is enabled on a bucket, it can be suspended but never fully disabled. Versioning is the prerequisite for cross-Region replication (CRR) and same-Region replication (SRR), which automatically copy new objects to another bucket — CRR for disaster recovery and compliance across Regions, SRR for aggregating logs or meeting data-residency rules within one Region. Replication copies new objects by default; existing objects need S3 Batch Replication.
For accidental deletion protection, MFA Delete requires multi-factor authentication to delete object versions or suspend versioning, and S3 Object Lock makes objects immutable for a fixed retention period (or indefinitely in legal hold), which is the answer for compliance scenarios like financial records under WORM (write once, read many) requirements. Encryption is straightforward: S3 encrypts all new objects at rest by default using SSE-S3 (Amazon-managed keys), and you can choose SSE-KMS for keys you control and audit through AWS Key Management Service, or SSE-C to provide your own keys. In transit, S3 uses HTTPS. A bucket policy can enforce encryption by denying uploads that do not specify it.
Access control and serving content
S3 buckets are private by default. Access is controlled by IAM policies, bucket policies, and access control lists (ACLs — largely legacy; the exam prefers policies). S3 Block Public Access is a set of account- and bucket-level settings that prevent buckets from becoming public, and it should stay enabled unless there is a specific reason to serve public content. When public access IS needed — for example, hosting a static website — the secure pattern is to serve the bucket through CloudFront with Origin Access Control (OAC), keeping the bucket itself private. Pre-signed URLs give time-limited access to a private object without changing the bucket policy, which is the standard way to let a user download a specific file. For auditing, S3 server access logging records requests, and CloudTrail records API activity including data events.
Moving and processing data in S3
Several S3 features handle data movement and processing at scale. Multipart upload splits a large object into parts that are uploaded in parallel and reassembled by S3 — it is required for objects over 5 GB, improves throughput for large uploads, and lets you retry a failed part instead of the whole upload. S3 Transfer Acceleration routes uploads through CloudFront edge locations and the AWS global network, which can dramatically speed up transfers from distant clients. S3 Batch Operations performs large-scale operations on lists of objects — copying, tagging, changing storage classes, invoking Lambda functions, or restoring archived objects — managed as a single job with progress tracking and a completion report.
For processing data without moving it, S3 Select lets you retrieve a subset of an object using simple SQL expressions, so an application can read only the rows and columns it needs instead of downloading an entire CSV or JSON file. S3 Object Lambda goes further: it lets you add your own code to transform data as it is retrieved, for example redacting personally identifiable information before the object is returned. S3 Inventory produces scheduled reports of everything in a bucket — useful for auditing and for planning lifecycle policies — and S3 Storage Lens gives organization-wide visibility into storage usage and activity, highlighting opportunities to save cost, such as buckets with incomplete multipart uploads consuming storage. Incomplete multipart uploads are a classic hidden cost: parts that were never completed or aborted keep consuming storage until a lifecycle rule cleans them up.
Exam traps to watch for
- Choosing S3 One Zone-IA for data that cannot be re-created. It stores data in a single AZ — if that AZ is destroyed, the data is gone.
- Picking Glacier Deep Archive when retrieval must happen within hours: Deep Archive Standard retrieval typically takes up to 12 hours, and Expedited is not available for it.
- Forgetting that replication requires versioning enabled on both the source and destination buckets, and that it does not copy objects that already existed (that needs Batch Replication).
- Making a bucket public to serve content instead of using CloudFront with Origin Access Control, or forgetting S3 Block Public Access exists.
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.
Q1.A company stores financial records in S3 that must be kept for 7 years and are almost never accessed after the first month. The records must be retrievable within 5 hours when requested. Which storage solution is MOST cost-effective while meeting the retrieval requirement?
Explanation: Glacier Flexible Retrieval is archive-priced and its Standard retrieval typically completes within 3–5 hours, which meets the 5-hour requirement; Expedited retrieval (1–5 minutes) is available if an urgent request ever needs it. Option A keeps rarely accessed data in the most expensive class for 7 years. Option B fails the requirement: Deep Archive Standard retrieval typically takes up to 12 hours and has no Expedited option. Option D is not archive pricing and stores the data in a single AZ, which is a poor choice for records that must survive 7 years.
Exam objective: SAA-C03 Task Statement 4.1: Design cost-optimized storage solutions
Q2.A company has an S3 bucket containing critical documents. A user accidentally overwrote an important object with an empty file. The company wants to be able to recover the previous version of any object. Which feature should the solutions architect enable?
Explanation: Versioning keeps every version of an object, so an overwritten or deleted object can be restored from its earlier version. Option A records who accessed the bucket but cannot restore data. Option B controls public access and has nothing to do with recovery. Option D speeds up uploads from distant locations using CloudFront edge locations and does not protect against overwrites.
Exam objective: SAA-C03 Task Statement 2.2: Design highly available and/or fault-tolerant architectures
Q3.A media company stores video files in S3. The access pattern is unpredictable: some videos become popular months after upload, while others are never watched again. The company wants storage costs to adapt automatically without manual lifecycle rules. Which storage class should the solutions architect recommend?
Explanation: Intelligent-Tiering monitors each object's access pattern and moves it automatically between frequent-access and infrequent-access tiers (and optional archive tiers), so costs adapt to unpredictable patterns without manual rules. Option D charges the highest storage price for every object regardless of access. Option C is archive storage with retrieval delays of minutes to hours — unsuitable for videos that might suddenly become popular. Option A stores data in one AZ and is for infrequently accessed, re-creatable data, not for content whose popularity is unknown.
Exam objective: SAA-C03 Task Statement 4.1: Design cost-optimized storage solutions
Q4.A company needs to serve a static website stored in a private S3 bucket to users worldwide with low latency. The bucket must not be publicly accessible. Which solution meets these requirements?
Explanation: CloudFront with Origin Access Control (OAC) is the secure pattern: the bucket stays private, only CloudFront can read it, and content is cached at edge locations worldwide for low latency. Option B makes the bucket public, violating the requirement. Option A speeds up uploads to S3 but does not serve a website or keep the bucket private. Option C is unworkable for a website: pre-signed URLs expire, and generating one per object per visitor is not a serving architecture.
Exam objective: SAA-C03 Task Statement 1.2: Design secure workloads and applications
Q5.A financial services company must retain transaction records in S3 so that no one — not even an administrator — can delete or modify them during a 5-year retention period. Which S3 feature meets this requirement?
Explanation: S3 Object Lock in compliance mode makes objects immutable for the retention period, and the retention settings cannot be shortened or removed by anyone, including the root user — which is exactly the write-once-read-many (WORM) guarantee regulators require. Option B preserves old versions but does not prevent deletion of versions. Option A copies objects to another Region but does not make them immutable. Option C can be changed by an administrator with permission to edit the bucket policy, so it is not an enforceable retention control.
Exam objective: SAA-C03 Task Statement 1.3: Determine appropriate data security controls
Q6.A company wants to automatically copy all new objects uploaded to an S3 bucket in us-east-1 to a bucket in eu-west-1 for disaster recovery. The company has enabled versioning on the source bucket only. What else is required for cross-Region replication to work?
Explanation: Cross-Region replication requires versioning enabled on BOTH the source and destination buckets, plus an IAM role granting S3 permission to read from the source and write to the destination. Option D is wrong because the destination bucket's versioning is also required. Option A is an unrelated deletion-protection feature. Option B changes the storage class and has nothing to do with replication.
Exam objective: SAA-C03 Task Statement 2.2: Design highly available and/or fault-tolerant architectures
Q7.A user needs to download a single private file from an S3 bucket. The company does not want to change the bucket policy or make the object public. Which solution provides temporary access to just that object?
Explanation: A pre-signed URL grants time-limited access to a specific object without changing any bucket policy or making anything public — the standard mechanism for sharing one private file. Option A protects against overwrites but grants no access. Option C is a content-delivery architecture, far more than needed for one file. Option B grants full S3 access to the user permanently, which is far broader than the requirement.
Exam objective: SAA-C03 Task Statement 1.1: Design secure access to AWS resources
Q8.A company stores application logs in S3. The logs are written daily, are rarely read after 30 days, and must be deleted after 1 year to meet a data retention policy. Which solution automates this MOST cost-effectively?
Explanation: A lifecycle policy automates both steps — the transition to the cheaper Standard-IA class after 30 days and the deletion after 365 days — with no manual work. Option A relies on manual effort every month, which is error-prone and not automated. Option C automates the tiering but still requires manual deletion. Option D fails because logs written today may still be read in their first month, and Deep Archive has a 180-day minimum storage duration plus hours-long retrieval.
Exam objective: SAA-C03 Task Statement 4.1: Design cost-optimized storage solutions