S3 stores data as objects in buckets. Versioning retains earlier copies, replication copies objects between buckets, and Object Lock protects retained versions. Access policies, encryption, lifecycle rules, and events control how that data is used and managed.
Buckets, objects, and durability
Amazon S3 stores data as objectsAn object is stored data together with metadata that describes it. in bucketsA bucket is a container for objects stored in S3.. It is built for any amount of data with very high durability, so it is the default place on AWS for files, backups, data lakes and static website content.
How it works
- Create a bucket in a Region.
- Upload objects. Each one is a file plus metadata, identified by a keyAn object key is the name used to identify an object within its bucket..
- S3 stores each object across at least three Availability Zones in most storage classesA storage class determines an object storage option with particular access, placement, and billing characteristics..
- Grant access explicitly with policies; everything starts private.
| Storage type | AWS example | Holds |
|---|---|---|
| Object | Amazon S3 | Files plus metadata, accessed over HTTP by key |
| Block | Amazon EBS, instance store | Volumes attached to an instance |
| File | Amazon EFS, Amazon FSx | Shared file systems mounted over a network |
Exam cues
- If the question says store unlimited files, backups or static web content, think Amazon S3.
- If the question says highest durability, think S3’s 11 nines.
Remember
- Durability: 99.999999999%. S3 Standard availability: 99.99%.
- Objects live in buckets; buckets live in a Region.
Trap: S3 is not a disk you mount on an instance. For a boot or database volume use block storage (Amazon EBS); for a shared file system use Amazon EFS or Amazon FSx.
From the AWS docs (15)
- Amazon S3 is an object storage service for storing and protecting any amount of data for uses such as data lakes, websites, mobile applications, backup and restore, archive and big data analytics.
- In Amazon S3 an object is a file and any metadata that describes it, and a bucket is a container for objects; each object has a key that identifies it uniquely within the bucket.
- To store data in Amazon S3 you create a bucket, choosing its name and AWS Region, then upload data to it as objects.
- S3 buckets and objects are private and accessible only when access is explicitly granted, through bucket policies, IAM policies, ACLs or access points.
- Amazon S3 is designed for 99.999999999% (11 nines) durability, and S3 Standard for 99.99% availability.
- S3 Standard, S3 Intelligent-Tiering, S3 Standard-IA and the S3 Glacier storage classes store objects redundantly on multiple devices across at least three Availability Zones in a Region.
- S3 Versioning keeps multiple versions of an object in the same bucket, so accidentally deleted or overwritten objects can be restored.
- S3 buckets have a flat structure, not a file system hierarchy; console folders are shared key name prefixes.
- S3 Transfer Acceleration speeds up long-distance transfers to a bucket by routing them through CloudFront edge locations.
- CloudTrail S3 data events record object-level API activity such as reads and writes.
- CloudTrail does not log data events by default; they must be enabled and incur additional charges.
- S3 Inventory produces daily or weekly object and metadata listings for an entire bucket or for keys with a selected prefix.
- Multipart upload sends an object as independent parts, in any order and in parallel, which raises throughput and lets you retry only the parts that failed on an unreliable network.
- Block Public Access settings constrain permissions granted through policies or ACLs; removing a block does not itself grant public access.
- One source bucket can replicate to a single destination bucket or to multiple destination buckets, in the same Region or in different Regions.
Storage classes
Amazon S3 storage classesA storage class determines an object storage option with particular access, placement, and billing characteristics. trade access speed and frequency for price. All are designed for the same 11 nines of durability; they differ in availability, retrieval time, retrieval fees and minimum storage durationA minimum storage duration is the billed retention period that can still incur charges when data leaves a storage class early..
How it works
- Pick a class per objectAn object is stored data together with metadata that describes it., by how often and how fast it is read.
- The less often data is read, the cheaper the storage, with retrieval fees and minimum durations added.
- Lifecycle rules can move objects to cheaper classes as they age.
| Class | Access pattern | Retrieval | Min duration | AZs |
|---|---|---|---|---|
| S3 Standard | Frequent | Milliseconds | None | 3 or more |
| S3 Intelligent-Tiering | Unknown or changing | Milliseconds; no retrieval fees | None | 3 or more |
| S3 Express One Zone | Latency-sensitive | Single-digit milliseconds | None | 1 |
| S3 Standard-IA | About once a month | Milliseconds | 30 days | 3 or more |
| S3 One Zone-IA | Once a month, re-creatable | Milliseconds | 30 days | 1 |
| S3 Glacier Instant Retrieval | About once a quarter | Milliseconds | 90 days | 3 or more |
| S3 Glacier Flexible Retrieval | About once a year | Minutes to hours | 90 days | 3 or more |
| S3 Glacier Deep Archive | Less than once a year | Hours | 180 days | 3 or more |
Exam cues
- If the question says unknown or changing access, think Intelligent-Tiering.
- If the question says rarely accessed but needs immediate access, think Standard-IA or Glacier Instant Retrieval.
- If the question says long-term archive, cheapest, hours are fine, think Glacier Deep Archive.
- If the question says re-creatable data at lowest cost, think One Zone-IA.
Remember
- All classes: 11 nines durability.
- Minimums: Standard-IA and One Zone-IA 30 days, Glacier IR and Flexible 90 days, Deep Archive 180 days.
- Availability: Standard 99.99%, Standard-IA and Intelligent-Tiering 99.9%, One Zone-IA 99.5%.
Trap: Cheaper classes are not always cheaper overall. Frequently read data in Standard-IA or Glacier pays retrieval fees that can exceed the storage savings.
From the AWS docs (13)
- S3 Standard is the default storage class, for frequently accessed data (more than once a month) with millisecond access, no minimum storage duration and 99.99% designed availability.
- S3 Intelligent-Tiering is for data with unknown, changing or unpredictable access patterns; it charges per-object monitoring and automation fees but no retrieval fees.
- S3 Standard-IA is for long-lived, infrequently accessed data (about once a month) with millisecond access, a 30-day minimum storage duration and per-GB retrieval fees.
- S3 One Zone-IA stores data in one Availability Zone for re-creatable, infrequently accessed data; it has a 30-day minimum and is not resilient to the loss of that AZ.
- S3 Express One Zone gives single-digit millisecond access for latency-sensitive applications in one Availability Zone you choose.
- S3 Glacier Instant Retrieval is for archive data accessed about once a quarter that needs millisecond access, with a 90-day minimum storage duration.
- S3 Glacier Flexible Retrieval is for archives accessed about once a year, with retrieval in minutes to hours, a 90-day minimum, and objects restored before access.
- S3 Glacier Deep Archive is for archives accessed less than once a year, with retrieval in hours and a 180-day minimum storage duration; it is the lowest-cost class.
- All of these storage classes are designed for 99.999999999% durability.
- Amazon S3 archives data at the lowest costs in S3 Glacier Instant Retrieval, S3 Glacier Flexible Retrieval and S3 Glacier Deep Archive.
- S3 storage classes are designed for different availability: S3 Standard 99.99%, Standard-IA, Intelligent-Tiering and Glacier Instant Retrieval 99.9%, and One Zone-IA 99.5%.
- S3 Express One Zone stores objects in directory buckets in one Availability Zone you choose, so compute in that zone can be co-located with its storage for the lowest latency.
- S3 Storage Class Analysis watches access patterns to help decide when to move data to STANDARD_IA; results appear after 24 to 48 hours and update daily, with optional daily export to a bucket.
Versioning and delete markers
S3 Versioning keeps earlier copies when an objectAn object is stored data together with metadata that describes it. is overwritten or deleted. An object keyAn object key is the name used to identify an object within its bucket. is its name within a bucketA bucket is a container for objects stored in S3.; a version IDA version ID identifies a particular saved version of an object. selects one saved copy of that name.
How it works
- Enable versioning on the bucket. Existing objects keep a
nullversion ID; later writes create versions with unique IDs. - Overwrite a key to create a new current versionThe current version is the most recent version returned when a request does not specify a version ID.. The earlier copy becomes noncurrent, meaning it is no longer the copy returned by an ordinary read.
- Delete without a version ID to add a delete markerA delete marker is a version that makes an object appear deleted while earlier versions can remain stored., a current version that hides the object. Delete that marker by its version ID to expose the preceding version.
- Use a version ID to target permanent deletion of one version. MFA deleteMFA delete requires an authentication code for permanent version deletion and bucket versioning-state changes. can add a requirement for an authentication code.
Exam cues
- If the question says an overwritten S3 file must be recovered, think versioning and retrieval of the earlier version, because each write retains a separate copy.
- If the question says an S3 object vanished after an ordinary delete, think delete-marker removal, because the earlier versions remain.
- If the question says permanent version deletion must require MFAMulti-factor authentication requires an additional proof of identity beyond security credentials., think root-enabled MFA delete, because this protection covers version deletion and versioning-state changes.
Remember
- Versioning starts disabled; after enabling it, you can only enable or suspend it.
- Suspension preserves existing versions but new writes can overwrite the
nullversion. - Every version incurs storage charges for a full object.
- MFA delete is configured through the CLI or API and cannot be used with lifecycle configurations.
Trap: Suspending versioning does not remove the stored history. Deleting a specific version permanently is different from placing a delete marker.
A DELETE without a version ID adds a delete marker and hides the saved object.
Deleting the current marker by its version ID exposes the preceding version.
From the AWS docs (17)
- S3 Versioning preserves separate versions of objects to recover from accidental overwrites or deletions.
- Buckets start unversioned.
- Once enabled, bucket versioning can be suspended but cannot return to the unversioned state.
- A write in a versioning-enabled bucket creates a version with a unique version ID.
- Objects present before versioning is enabled keep their null version ID until changed by a later write.
- A DELETE without a version ID in a versioning-enabled bucket adds a delete marker as the current version.
- Permanently deleting the current delete marker by its version ID exposes the preceding version.
- A DELETE with a version ID permanently removes that version when authorized.
- Suspending versioning leaves existing versions unchanged.
- New writes in a versioning-suspended bucket have the null version ID and replace an existing null version.
- Each stored version is a full object and incurs normal storage charges.
- Only the bucket owner root account can enable MFA delete.
- MFA delete is configured through the CLI or API, not the console.
- MFA delete requires an authentication code for permanent version deletion and changes to the bucket versioning state.
- MFA delete cannot be used with lifecycle configurations.
- An object key is the name used to identify an object in a bucket.
- The current version is returned by default; a previous version becomes noncurrent when a newer version becomes current.
Replication
S3 Replication creates copies in other bucketsA bucket is a container for objects stored in S3. without making the application upload each copy. It is asynchronous: the original write can finish before its replicaA replica is an object copy that S3 creates in a replication destination bucket. is available.
How it works
- Enable versioning on both buckets. Choose Cross-Region ReplicationCross-Region Replication copies matching objects to buckets in a different AWS Region. (CRR) for another Region, including copies nearer distant readers, or Same-Region ReplicationSame-Region Replication copies matching objects to buckets in the same AWS Region. (SRR) for the same Region, such as consolidating logs.
- Create a source rule selecting objectsAn object is stored data together with metadata that describes it. and a destination. Select all source objects or filter by prefixA prefix is the beginning of an object key used to select related object names., tags, or both. Choose a replica storage classA storage class determines an object storage option with particular access, placement, and billing characteristics. if it should differ from the source. Configure rules in both directions when both buckets accept writes. Give S3 an IAM roleAn IAM role is an identity with permissions that a trusted service or caller can assume., a permission identity it can assume to read source versions and write replicas.
- For a different destination account, grant replication access in its bucket policyA bucket policy is a resource policy that grants or denies access to a bucket and its objects.. Use destination Bucket owner enforced Object Ownership when that account must own replicas.
- For server-side encryptionServer-side encryption lets S3 encrypt object data during storage and decrypt it for authorized access. using KMS keysAn object key is the name used to identify an object within its bucket. (SSE-KMSSSE-KMS is server-side object encryption using keys managed in AWS KMS.), or its dual-layer form (DSSE-KMSDSSE-KMS applies two layers of server-side encryption using KMS keys.), explicitly enable encrypted-object replication and specify a destination-Region KMS key. Grant source decrypt and destination encrypt access.
- S3 copies new matching versions. Configure destination lifecycle and other bucket settings separately.
- To backfill existing objects or retry failures, run S3 Batch Replication with a manifestA manifest lists the object keys and, when required, version IDs that a batch job processes. and an existing replication configuration.
- The caller setting replication needs
s3:PutReplicationConfigurationandiam:PassRole.
Exam cues
- If the question says new S3 writes need copies in another Region, think CRR, because it copies matching versions asynchronously across Regions.
- If the question says S3 replicas must survive version-specific deletes in the source, think replication, because these deletes are not propagated.
- If the question says a replication chain misses copies from its first bucket, think direct source-to-destination rules, because live replication skips replicas from another rule.
- If the question requires a replication time target and CloudWatch visibility, think S3 Replication Time Control, because it enables replication metrics and replicates 99.9 percent of uploaded objects within 15 minutes.
- If the question says an SSE-KMS replica must be readable in another account, think a destination-owned customer managed KMS key and key policy access for the replication role.
- If the question says copy every new object to buckets in several Regions, think S3 Replication to multiple destinations, because one source bucket can replicate to many.
Remember
- Replication is directional; buckets that both accept writes need two-way rules.
- A destination storage-class setting overrides the source class for replicas.
- Both buckets need versioning. Existing objects need Batch Replication.
- V2 delete-marker replication is off by default and cannot be enabled for tag-filtered rules. V1 rules replicate markers from user deletes.
- Lifecycle actions, bucket settings, and archived source objects are excluded.
- An Object Lock source requires Object Lock on the destination and permissions to read retention and legal holdsA legal hold protects an object version without an expiry date until an authorized caller removes it..
- SSE-S3SSE-S3 is server-side object encryption with keys managed by S3. and SSE-CSSE-C is server-side encryption using a key that the caller must supply and retain. do not need the SSE-KMS opt-in.
- Replicas of SSE-S3, SSE-KMS, or DSSE-KMS objects keep the source encryption type rather than using the destination bucket default.
- Cross-account destinations cannot use Requester Pays.
Trap: A destination bucket does not automatically inherit the source lifecycle configuration. A version-specific source delete also does not remove the replica.
A new object version matches the replication rule, and S3 assumes the configured IAM role.
S3 copies the version asynchronously, subject to bucket and KMS permissions.
Box details (3)
- Source bucket
A live replication rule selects new versions. Use Batch Replication for existing, failed, or previously replicated objects.
- IAM role
S3 assumes this role to read source versions and replicate them. SSE-KMS requires source key decrypt and destination key encrypt permissions.
- Destination
A cross-account destination bucket policy grants replication access. Bucket owner enforced Object Ownership makes the destination account own replicas.
From the AWS docs (29)
- S3 Replication copies objects asynchronously between buckets in the same or different accounts.
- Cross-Region Replication copies objects between AWS Regions, including copies in Regions closer to geographically distributed readers to reduce access latency.
- Same-Region Replication copies objects between buckets in one AWS Region and can consolidate logs from several source buckets into one destination.
- Live replication covers new matching objects and does not copy objects that predate the rule.
- Both source and destination buckets must have versioning enabled.
- S3 needs an IAM role granting permission to read source versions and replicate them to the destination.
- A cross-account destination requires a bucket policy granting the source account replication permissions.
- Bucket owner enforced Object Ownership on the destination makes the destination account own replicas without an owner override.
- Version-specific deletions are not replicated to the destination.
- V2 rules with a Filter element do not replicate delete markers by default; marker replication can be enabled for rules without tag filters.
- V1 rules without a Filter element replicate delete markers caused by user actions.
- Live replication does not copy replicas created by another rule.
- Objects stored in Glacier Flexible Retrieval, Glacier Deep Archive, and Intelligent-Tiering archive access tiers are not copied by live replication.
- Lifecycle actions and bucket-level configurations are not replicated.
- Replication of SSE-KMS and DSSE-KMS objects requires explicit opt-in and a destination KMS key in the destination Region.
- SSE-KMS replication requires kms:Decrypt permission on the source key.
- SSE-KMS replication requires kms:Encrypt permission on the destination key.
- SSE-S3 and SSE-C replication do not require the SSE-KMS opt-in.
- Replication from an Object Lock bucket requires Object Lock on the destination and source permissions s3:GetObjectRetention and s3:GetObjectLegalHold.
- SSE-KMS uses AWS KMS keys for server-side encryption; DSSE-KMS applies two layers of encryption.
- S3 Replication Time Control replicates 99.9 percent of uploaded objects within 15 minutes and enables replication metrics.
- S3 Batch Replication uses an existing replication configuration and a manifest to retry failed objects or copy existing and previously replicated objects.
- For cross-account SSE-KMS replication, use a customer managed key owned by the destination account and grant the replication role access in its key policy.
- A cross-account replication destination cannot be a Requester Pays bucket.
- Setting a replication configuration requires s3:PutReplicationConfiguration and iam:PassRole for the caller.
- Replicas of SSE-S3, SSE-KMS or DSSE-KMS source objects use the source encryption type rather than the destination bucket default.
- Synchronizing buckets that both receive writes requires two-way replication rules; a rule from one source bucket to a destination does not configure the reverse direction.
- A replication rule can choose a storage class for replicas independently of the source; without that setting, replicas use the source storage class.
- An S3 replication rule can cover all objects or filter them by a key prefix, tags, or both; combined prefix and tag filters must all match.
Lifecycle and retention
S3 Lifecycle applies rules to move aging objectsAn object is stored data together with metadata that describes it. between storage classesA storage class determines an object storage option with particular access, placement, and billing characteristics. or remove them. A storage class is a choice of access speed, storage placement, and billing terms; choose it before scheduling transitions.
How it works
- Choose a class for the access pattern; it can be selected during upload. Standard suits frequent reads. IA means infrequent access: Standard-IA retains zone resilience; One Zone-IA suits data that can be recreated after a zone loss.
- Use lifecycle rules when age determines the move. Select the whole bucketA bucket is a container for objects stored in S3. or filter by prefixA prefix is the beginning of an object key used to select related object names. or tags. Use Intelligent-Tiering when changing access patterns should determine the access tier; reads from its Infrequent Access or Archive Instant Access tiers return objects to Frequent Access.
- Configure current-version and noncurrent-version actions separately. In a versioning-enabled bucket, current-version expiration adds a delete markerA delete marker is a version that makes an object appear deleted while earlier versions can remain stored.; noncurrent expiration removes old versions permanently.
- Add
AbortIncompleteMultipartUploadfor abandoned uploads. It removes unfinished parts after the configured age. - Allow each class’s minimum storage durationA minimum storage duration is the billed retention period that can still incur charges when data leaves a storage class early. before another transition in the same rule. Separate rules can move objects sooner but still incur remaining-duration charges. Glacier Instant Retrieval can transition to Flexible Retrieval or Deep Archive. Intelligent-Tiering destinations depend on the access tier. Lifecycle cannot move Deep Archive objects back to Standard; restoreA restore makes archived object data readable, temporarily for Glacier Flexible Retrieval and Deep Archive. and copy them.
- Rules apply to existing objects as well as new ones.
| Choice | Suitable access pattern or tradeoff |
|---|---|
| Standard | Frequent access; default class |
| Standard-IA | Infrequent reads with data across zones |
| One Zone-IA | Re-creatable data in one zone |
| Intelligent-Tiering | Changing or unknown access patterns |
| Glacier classes | Rare access; restore requirements depend on class |
| Intelligent-Tiering access tier | Lifecycle destinations |
|---|---|
| Frequent Access or Infrequent Access | One Zone-IA or any Glacier class |
| Archive Instant Access | Any Glacier class |
| Archive Access | Glacier Flexible Retrieval or Deep Archive |
| Deep Archive Access | Glacier Deep Archive |
Exam cues
- If the question says old versions remain after S3 logs expire, think
NoncurrentVersionExpiration, because current-version expiration keeps the older data. - If the question says S3 access patterns are unknown, think Intelligent-Tiering, because it monitors access and changes tiers automatically.
- If the question says S3 logs should move to IA immediately, think a day-zero lifecycle ruleA lifecycle rule selects objects and schedules actions such as storage transitions or expiration., because July 2026 removed the old transition wait.
- If an S3 bucket policyA bucket policy is a resource policy that grants or denies access to a bucket and its objects. denies deletion, think lifecycle configuration separately, because bucket policies do not prevent lifecycle actions.
- If the question says act only on S3 objects that another process marked, think a Lifecycle tag filter, because it matches an exact tag keyAn object key is the name used to identify an object within its bucket. and value.
Remember
- In a single rule, wait 30 days in an IA class or 90 days in Glacier Instant Retrieval before the next transition. Separate rules do not avoid minimum-duration charges.
- Intelligent-Tiering cost includes monitoring fees and savings from moving objects to cheaper tiers.
- From September 2024, new or modified configurations block transitions below 128 KB by default; size filters can override this. Older configurations keep their earlier behavior until changed.
- Minimum billing: IA classes 30 days; Glacier Instant and Flexible 90 days; Deep Archive 180 days. Early exit can still incur charges.
- Intelligent-Tiering moves unaccessed objects after 30 days to Infrequent Access and 90 days to Archive Instant Access.
- Optional Archive Access and Deep Archive Access need activation, with minimum no-access periods of 90 and 180 days. Objects below 128 KB stay in Frequent Access. The optional archive tiers require restoration before reading.
- Lifecycle does not expire locked noncurrent versionsA noncurrent version is a saved object version that is no longer current. or act on versions with pending or failed replication.
- Physical removal may occur after expiration.
Trap: The 30-day IA billing minimum is not a 30-day wait in S3 Standard. The wait was removed in July 2026.
From the AWS docs (36)
- S3 Lifecycle rules schedule transitions or expiration for all objects in a bucket, or select a subset with filters such as a key prefix or object tags.
- Current and noncurrent versions have separate transition and expiration actions.
- Expiration in a versioning-enabled bucket adds a delete marker; NoncurrentVersionExpiration permanently removes older versions.
- Since September 2024, new or modified lifecycle configurations block transitions of objects smaller than 128 KB by default.
- Configurations created before September 2024 retain the older small-object behavior until modified.
- ObjectSizeGreaterThan or ObjectSizeLessThan filters can override the default minimum transition size.
- In July 2026, S3 removed the 30-day wait before transitions into Standard-IA and One Zone-IA; rules can use 0 days after creation.
- Standard-IA and One Zone-IA each have a 30-day minimum storage billing duration.
- Glacier Instant Retrieval has a 90-day minimum storage billing duration.
- Glacier Flexible Retrieval has a 90-day minimum storage billing duration.
- Glacier Deep Archive has a 180-day minimum storage billing duration.
- Deleting, overwriting, or transitioning objects out of a class before its minimum storage duration can incur remaining-duration charges.
- AbortIncompleteMultipartUpload removes unfinished upload parts after the configured number of days.
- Lifecycle cannot transition Deep Archive objects to another class; restore a temporary copy and copy it with the required storage class.
- S3 Standard is the default storage class for frequently accessed objects.
- Standard-IA stores data redundantly across multiple Availability Zones.
- One Zone-IA stores data in one Availability Zone and is not resilient to physical loss of that zone.
- Intelligent-Tiering charges a monitoring and automation fee for eligible objects and automatically moves unread data into cheaper access tiers, so cost depends on both that fee and the tier savings.
- Intelligent-Tiering moves objects to Infrequent Access after 30 consecutive days without access.
- Intelligent-Tiering moves objects to Archive Instant Access after 90 consecutive days without access.
- Intelligent-Tiering optional Archive Access and Deep Archive Access tiers require activation and have minimum no-access periods of 90 and 180 days respectively.
- Intelligent-Tiering objects smaller than 128 KB are not monitored or automatically tiered and remain in Frequent Access.
- Lifecycle does not expire noncurrent versions protected by Object Lock.
- Lifecycle does not act on current or noncurrent versions with PENDING or FAILED replication status.
- Objects in Intelligent-Tiering Archive Access or Deep Archive Access must be restored before reading; restoration moves them to Frequent Access.
- A storage class defines object storage characteristics including access time, placement, and minimum billing terms.
- Lifecycle rules apply to both existing and newly added objects.
- A bucket policy cannot prevent deletions or transitions performed by an S3 Lifecycle rule.
- S3 may remove an object after its lifecycle expiration date rather than at the exact expiration time.
- A Lifecycle rule tag filter matches only objects whose tag key and value both match exactly. Several tags go inside And, and the object must carry all of them; extra tags do not stop the match.
- S3 Batch Operations applies a tag set to every object in a supplied list in one job, which can cover billions of objects.
- S3 Lifecycle can move Standard objects to Standard-IA, Intelligent-Tiering, One Zone-IA or Glacier classes. Standard-IA can move to Intelligent-Tiering, One Zone-IA or Glacier classes. Glacier Instant Retrieval can move to Flexible Retrieval or Deep Archive; One Zone-IA can move only to Flexible Retrieval or Deep Archive. No class transitions back to Standard.
- Each lifecycle transition is billed as a transition request per object, which is why S3 by default skips transitioning objects smaller than 128 KB.
- A single S3 Lifecycle rule must keep an object in a class for its minimum storage duration before another transition: 30 days for the IA classes and 90 days for Glacier Instant Retrieval. Separate rules can move it earlier, with remaining-duration charges.
- Reading an object in the Intelligent-Tiering Infrequent Access or Archive Instant Access tier automatically returns it to Frequent Access.
- An S3 upload can specify its storage class, including Standard-IA or One Zone-IA; omitting the choice uses Standard.
Glacier retrieval
S3 Glacier storage classesA storage class determines an object storage option with particular access, placement, and billing characteristics. trade access speed for archive storage costs. Instant Retrieval stays directly readable; Flexible Retrieval and Deep Archive require a restoreA restore makes archived object data readable, temporarily for Glacier Flexible Retrieval and Deep Archive. job before an application can read the data.
How it works
- Check the storage class. Use a normal
GETfor Instant Retrieval; useRestoreObjectfor Flexible Retrieval or Deep Archive. - Select a retrieval tier with its speed and charges in mind. Standard is the default if the request does not specify one. Deep Archive does not offer Expedited. Bulk retrieval from Flexible Retrieval is free.
- For Flexible Retrieval Expedited requests that need reserved capacity, purchase provisioned capacityProvisioned capacity reserves capacity for Expedited retrievals from Glacier Flexible Retrieval.. Without it, high demand can cause rejection.
- Read the temporary restored copy for the requested number of days. Copy it with a different storage class if it must remain readable permanently.
| Storage class | Expedited | Standard | Bulk |
|---|---|---|---|
| Glacier Instant Retrieval | Not a restore path; use GET | Not a restore path | Not a restore path |
| Glacier Flexible Retrieval | Typically 1 to 5 minutes below 250 MB | Typically 3 to 5 hours | Typically 5 to 12 hours |
| Glacier Deep Archive | Unavailable | Typically within 12 hours | Typically within 48 hours |
Exam cues
- If the question says an S3 archive must be immediately readable, think Glacier Instant Retrieval, because it needs no restore.
- If the question says S3 Flexible Retrieval needs reserved urgent retrieval capacity, think provisioned capacity, because on-demand Expedited requests can be rejected.
- If the question says an S3 restore must become a permanent Standard objectAn object is stored data together with metadata that describes it., think restore then copy, because restoration alone is temporary.
Remember
- Retrieval charges vary by tier; Flexible Retrieval Bulk retrieval is free.
- Flexible Retrieval Expedited is typically 1 to 5 minutes for objects smaller than 250 MB. Large datasets can take longer.
- Each provisioned unit supports at least three Expedited requests per 5 minutes and up to 300 MB/s.
- Minimum billed storage is 90 days for Instant and Flexible, 180 days for Deep Archive.
- Intelligent-Tiering archive restores move data to Frequent Access and do not accept a
Daysvalue.
Trap: Provisioned capacity does not add Expedited retrieval to Deep Archive.
RestoreObject makes a temporary readable copy of a Flexible Retrieval or Deep Archive object.
Copy the restored data with S3 Standard as the storage class to keep it permanently readable.
From the AWS docs (23)
- Glacier Instant Retrieval objects are immediately readable with GET.
- Glacier Flexible Retrieval and Deep Archive objects require RestoreObject before reading.
- Flexible Retrieval Expedited typically makes objects smaller than 250 MB available in 1 to 5 minutes.
- Flexible Retrieval Standard retrievals typically finish in 3 to 5 hours.
- Flexible Retrieval Bulk retrievals typically finish in 5 to 12 hours.
- Deep Archive Standard retrievals typically finish within 12 hours.
- Deep Archive Bulk retrievals typically finish within 48 hours.
- Deep Archive has no Expedited retrieval option.
- Standard is the default retrieval tier when none is specified.
- Each provisioned capacity unit supports at least three Expedited retrievals per 5 minutes.
- Each provisioned capacity unit supports up to 300 MB/s of retrieval throughput.
- Without provisioned capacity, Expedited requests might be rejected during high demand.
- Restoring Flexible Retrieval or Deep Archive creates a temporary readable copy for the requested number of days and leaves the archive in its class.
- To retain a restored archive in a different storage class, copy it while specifying the target class.
- Glacier Instant Retrieval has a 90-day minimum storage billing duration.
- Glacier Flexible Retrieval has a 90-day minimum storage billing duration.
- Glacier Deep Archive has a 180-day minimum storage billing duration.
- Restoring an Intelligent-Tiering Archive Access or Deep Archive Access object moves it to Frequent Access without a Days value.
- Large archive datasets can take longer to restore than typical retrieval times.
- Bulk is the lowest-cost S3 Glacier retrieval option; Bulk retrievals from Glacier Flexible Retrieval are free.
- Each object in Glacier Flexible Retrieval or Glacier Deep Archive adds 40 KB of metadata: 32 KB billed at the archive class rate and 8 KB at the S3 Standard rate.
- Glacier Instant Retrieval bills each object as at least 128 KB.
- Charges for restoring S3 Glacier Flexible Retrieval and Deep Archive objects vary with the selected retrieval tier.
Object Lock
S3 ObjectAn object is stored data together with metadata that describes it. Lock prevents protected object versions from being overwritten or deleted. Its write-once-read-many model means data can be read repeatedly while the saved version remains protected.
How it works
- Enable Object Lock and versioning. You can enable Object Lock on an existing bucketA bucket is a container for objects stored in S3., but cannot later disable it or suspend versioning.
- Choose a retention mode. Governance allows an explicit authorized bypass; compliance protects the version even from the root user.
- Apply a bucket default in days or years, or explicit retention to individual versions. Individual settings override the default for that version.
- Use a legal holdA legal hold protects an object version without an expiry date until an authorized caller removes it. when the release date is unknown. Removing a hold does not remove an independent retention periodA retention period protects an object version until its retain-until date..
- Keep the version IDA version ID identifies a particular saved version of an object. for recovery. A new write or ordinary delete can change what an ordinary read sees while leaving the locked version protected.
| Control | Release condition |
|---|---|
| Governance retention | Expiry or authorized explicit bypass |
| Compliance retention | Retention expiry; no user bypass |
| Legal hold | Explicit removal by an authorized caller |
Exam cues
- If the question says even the S3 account root user must not delete a backup version during retention, think compliance modeCompliance mode prevents any account user, including root, from deleting a protected version or shortening its retention., because it has no governance-style bypass.
- If the question says S3 audit data must remain locked until an investigator releases it, think legal hold, because it has no expiry.
- If the question says a successful S3 delete appears to defeat Object Lock, think delete markerA delete marker is a version that makes an object appear deleted while earlier versions can remain stored., because hiding a keyAn object key is the name used to identify an object within its bucket. does not delete its protected version.
Remember
- Governance bypass needs both
s3:BypassGovernanceRetentionand the explicit bypass header. - Retention can be extended; compliance retention cannot be shortened.
- Legal hold removal requires
s3:PutObjectLegalHold.
Trap: Object Lock protects versions, so it does not prevent another version or delete marker from becoming current.
From the AWS docs (13)
- S3 Object Lock uses write-once-read-many protection against deletion or overwrite of object versions.
- Object Lock requires S3 Versioning and protects individual versions rather than the object name.
- A protected version does not prevent new versions or delete markers from being added to the same key.
- Governance retention can be bypassed with s3:BypassGovernanceRetention and the x-amz-bypass-governance-retention:true request header.
- Compliance retention prevents any user, including root, from deleting the protected version or shortening its retention.
- An authorized user can extend an object retention period.
- A legal hold has no expiry and remains until a caller with s3:PutObjectLegalHold removes it.
- Legal holds and retention periods are independent, and a version can have both.
- A bucket default retention duration can be specified in days or years and applies to new object versions.
- Explicit per-version retention settings override the bucket default for that version.
- Object Lock can be enabled on an existing bucket.
- After enabling Object Lock, it cannot be disabled and versioning cannot be suspended on that bucket.
- An ordinary DELETE of a protected object can succeed by adding a delete marker while the protected version remains locked.
Encryption and Bucket Keys
S3 server-side encryptionServer-side encryption lets S3 encrypt object data during storage and decrypt it for authorized access. protects objectAn object is stored data together with metadata that describes it. data at rest, meaning while stored on disk. S3 encrypts data during storage and decrypts it for authorized reads; the encryption option determines who controls the keysAn object key is the name used to identify an object within its bucket..
How it works
- Choose S3-managed keys for the base encryption behavior or AWS Key Management Service (KMS) keys for key access control and auditing. Previously unencrypted objects need a separate copy operation to encrypt them.
- Use a customer managed KMS key when sharing SSE-KMSSSE-KMS is server-side object encryption using keys managed in AWS KMS. objects across accounts. The AWS managed S3 key cannot provide that cross-account sharing.
- Set bucketA bucket is a container for objects stored in S3. default encryption for uploads without explicit encryption settings. Enforce strict requirements with a bucket-policy deny, because request settings can override the default.
- Enable Bucket Keys for SSE-KMS request savings. Existing eligible SSE-KMS objects can use
UpdateObjectEncryptionto enable Bucket Keys without a data copy. Fewer KMS calls also mean fewer KMS CloudTrail events. - If SSE-CSSE-C is server-side encryption using a key that the caller must supply and retain. is required, provide the key over HTTPS. S3 does not keep that key.
| Option | Key control |
|---|---|
| SSE-S3SSE-S3 is server-side object encryption with keys managed by S3. | S3 manages keys |
| SSE-KMS | KMS manages keys with key policies and auditing |
| DSSE-KMSDSSE-KMS applies two layers of server-side encryption using KMS keys. | Two layers of server-side encryption with KMS keys |
| SSE-C | Caller supplies and retains the key |
Exam cues
- If the question says S3 KMS request costs are high, think Bucket Keys, because a short-lived bucket-level key reduces calls to KMS.
- If the question says S3 uploads must explicitly request KMS encryption, think a bucket-policy deny for a missing or different encryption header, because a default is overridable.
- If the question says S3 encrypted objects must be shared across accounts, think a customer managed KMS key, because the AWS managed S3 key cannot support that sharing.
Remember
- Changing bucket default encryption leaves existing objects unchanged.
- SSE-KMS can use the AWS managed
aws/s3key; cross-account sharing requires a customer managed keyA customer managed key is a KMS key whose policies and lifecycle are controlled by the customer.. - New uploads have base SSE-S3 encryption using AES-256 since January 2023; it cannot be disabled and adds no encryption charge.
- Bucket Keys can reduce KMS request costs by up to 99%; savings are not guaranteed for each workload.
- Bucket Keys do not support DSSE-KMS.
Trap: Default SSE-KMS does not require every caller to send a KMS header. A policy that requires that header also rejects an upload that would otherwise inherit the default.
From the AWS docs (31)
- Server-side encryption protects object data stored on disk; S3 encrypts uploads and decrypts authorized reads.
- Since January 2023, all new S3 uploads receive server-side encryption.
- SSE-S3 is the base default when no other encryption is selected.
- Default SSE-S3 encryption adds no encryption charge.
- SSE-S3 uses AES-256 with S3-managed encryption keys.
- Base encryption for new uploads cannot be disabled.
- Automatic encryption of new uploads does not encrypt previously stored unencrypted objects.
- Customer managed KMS keys allow control of key access, rotation, disablement, and auditing.
- Cross-account sharing of SSE-KMS objects requires a customer managed key rather than the AWS managed S3 key.
- A bucket policy can deny PutObject when s3:x-amz-server-side-encryption is not aws:kms.
- A bucket policy can require a specific KMS key with s3:x-amz-server-side-encryption-aws-kms-key-id.
- Upload-request encryption settings can override the bucket default encryption configuration.
- Bucket Keys use short-lived bucket-level keys to reduce SSE-KMS calls to KMS.
- Bucket Keys can reduce KMS request costs by up to 99%.
- Turning on Bucket Keys for a bucket does not change existing objects. For eligible SSE-KMS objects, UpdateObjectEncryption enables a Bucket Key without copying object data.
- Bucket Keys use the bucket ARN as encryption context, so policies expecting an object ARN must be updated.
- Bucket Keys do not support DSSE-KMS.
- SSE-C requires the caller to supply the encryption key for upload and retrieval; S3 does not retain the key.
- Losing an SSE-C key means losing access to its object.
- SSE-C requests require HTTPS.
- An Amazon Resource Name (ARN) uniquely identifies an AWS resource.
- An encryption context supplies additional identifying information to KMS and can be used in key access policies.
- DSSE-KMS applies two layers of server-side encryption to uploaded objects.
- Each upload or download of an SSE-KMS object calls KMS and counts against the Region's KMS request quota, so heavy traffic can be throttled; SSE-S3 makes no KMS calls.
- SSE-S3 encrypts each object with its own unique key and encrypts that key with a root key that S3 rotates regularly.
- Client-side encryption encrypts data before it is sent to S3, for example with the Amazon S3 Encryption Client, so S3 only ever stores ciphertext and never sees the keys.
- Uploading an SSE-KMS object needs kms:GenerateDataKey on the key and downloading it needs kms:Decrypt, so S3 access alone does not let someone read the data.
- With SSE-KMS, the KMS key must be in the same AWS Region as the S3 bucket.
- S3 Bucket Keys reduce the number of KMS CloudTrail records for SSE-KMS objects because S3 reuses temporary key material instead of contacting KMS for every request.
- SSE-KMS can use the AWS managed key aws/s3 when no customer managed key is supplied.
- Setting bucket default encryption leaves the encryption of objects already stored in the bucket unchanged.
Public access and object ownership
S3 separates public-access safeguards from objectAn object is stored data together with metadata that describes it. ownership. Block Public Access limits public grants; Object Ownership decides who owns uploaded objects and whether access control listsAn access control list is a set of grants attached to a bucket or object. (ACLs), lists of grants attached to resources, are used.
How it works
- Apply Block Public Access at the account level or through an organization policy for a shared baseline. Turning off a block does not itself grant access. S3 combines it with bucketA bucket is a container for objects stored in S3. and access point settings using the most restrictive result.
- Keep Bucket owner enforced for policy-based access. It makes the bucket owner own all objects and disables ACL authorization.
- For a legacy ACL workflow, distinguish Bucket owner preferred from Object writer. Preferred needs
bucket-owner-full-controlon a new cross-account upload to transfer ownership. - Deny insecure client requests with
aws:SecureTransport: false. When appropriate, exclude AWS service principalsA principal is the person, application identity, or service making an AWS request. usingaws:PrincipalIsAWSService: falseso redacted service network context does not block delivery. - Treat website endpoints separately from REST endpoints: they serve publicly readable website content over HTTP, including the configured index document. Use CloudFront when HTTPS delivery is required.
| Object Ownership | Who owns an uploaded object? | ACLs |
|---|---|---|
| Bucket owner enforced | Bucket owner | Disabled; default |
| Bucket owner preferred | Bucket owner for new cross-account uploads with bucket-owner-full-control |
Enabled |
| Object writer | Uploading account | Enabled |
Exam cues
- If the question says an S3 bucket must stay private even after a public grant, think Block Public Access, because its applicable settings override public permissions.
- If the question says S3 cross-account uploads must belong to the bucket owner without ACL management, think Bucket owner enforced, because it disables ACLs and gives that owner all objects.
- If the question says an S3 website needs HTTPS, think CloudFront, because the website endpoint itself supports HTTP only.
Remember
BlockPublicAclsrejects new public ACLs;IgnorePublicAclsignores their grants.BlockPublicPolicyrejects new public policies;RestrictPublicBucketsrestricts access when a policy is public.- Bucket owner enforced accepts no ACL or
bucket-owner-full-control; other upload ACLs fail withAccessControlListNotSupported. - The legacy
authenticated-readACL is not restricted to an organization.
Trap: An authenticated AWS account is not necessarily a trusted account. The AuthenticatedUsers ACL group includes accounts outside your organization.
From the AWS docs (19)
- S3 Block Public Access overrides public grants in policies and ACLs when a blocking setting applies.
- Block Public Access can be managed at organization, account, bucket, and access point levels; S3 uses the most restrictive applicable combination.
- BlockPublicAcls rejects requests that set public ACLs without changing existing ACLs.
- IgnorePublicAcls disregards public ACL grants without preventing those ACLs from being stored.
- BlockPublicPolicy rejects new public bucket policies without changing existing policies.
- RestrictPublicBuckets limits a bucket with a public policy to AWS service principals and authorized users in the owner account, blocking other cross-account access.
- Bucket owner enforced is the default Object Ownership setting: ACLs are disabled and the bucket owner owns all objects.
- With ACLs disabled, policies determine access rather than ACL grants.
- Bucket owner preferred transfers ownership of a new cross-account upload to the bucket owner when it uses bucket-owner-full-control.
- Object writer leaves ownership with the uploading account and keeps ACLs enabled.
- Bucket owner enforced accepts uploads without an ACL or with bucket-owner-full-control; other ACL uploads fail with AccessControlListNotSupported.
- The authenticated-read canned ACL grants object READ access to the AuthenticatedUsers group, which includes any authenticated AWS account.
- A bucket policy deny with aws:SecureTransport false rejects insecure client requests.
- Using aws:PrincipalIsAWSService false in a network-based deny excludes AWS service principals whose network context can be redacted.
- S3 website endpoints support only publicly readable content and do not support HTTPS.
- A website endpoint returns its configured index document at the bucket root rather than the REST endpoint object listing.
- CloudFront can provide HTTPS for a static website hosted on S3.
- An ACL is a list of grants attached to a bucket or object.
- IAM Access Analyzer for S3 reports general purpose buckets that allow access to the public or to other AWS accounts, and a public bucket can be blocked from that page with one action.
Access points and CORS
S3 access points give a shared bucketA bucket is a container for objects stored in S3. separate named entry points and permission controls. CORSCORS uses HTTP headers to tell browsers which cross-origin requests a server permits., or cross-origin resource sharing, lets browser code use permitted resources from another originAn origin is the combination of a URL protocol, host, and port.; an origin combines protocol, host, and port.
How it works
- Create an access point for each required access policy or private-network boundary. For a VPCA VPC is a logically isolated virtual network in AWS., a private AWS network, restrict the point to that VPC. The access point policy and bucket policyA bucket policy is a resource policy that grants or denies access to a bucket and its objects. work together.
- For browser access, configure CORS. When a preflight check is needed, the browser sends
OPTIONSbefore the actual request. - Authorize the actual request independently. A VPC endpointA VPC endpoint enables private connectivity from a VPC to supported AWS services. provides private connectivity to AWS services.
- For temporary sharing, issue a presigned URLA presigned URL delegates time-limited access to a specified object operation using its signer permissions. for a specific objectAn object is stored data together with metadata that describes it. and operation. A presigned
PUTlets a recipient upload without AWS credentials, but does not outlive its signing credentials. - For eligible existing users, S3 Object Lambda invokes Lambda through an Object Lambda Access Point to transform returned data. A supporting access point provides access to the original object.
Exam cues
- If the question says different teams need different S3 policies for a shared dataset, think access points, because each endpoint has its own access controls.
- If the question says authorized browser code cannot read S3 from another origin, think CORS matching, because browser permission and object authorization are separate checks.
- If the question says temporary S3 uploads must work without handing out credentials, think a presigned PUT URL, because it delegates one permitted operation.
- If the question says an existing Object Lambda user needs filtered S3 responses, think a response transformation, because the original object can remain unchanged.
Remember
- CORS does not grant object permissions. A different protocol, host, or port is a different origin.
- Access points are for object operations, not bucket administration.
- Signature Version 4Signature Version 4 adds authentication information to AWS API requests. (SigV4) signs AWS API requests. CLI or SDK SigV4 presigned URLs can last up to 7 days only while their signing credentials remain valid. Treat URLs as bearer tokensA bearer token grants its permitted access to anyone who possesses it while it remains valid..
- A presigned upload to an existing keyAn object key is the name used to identify an object within its bucket. replaces its current data; versioning determines whether earlier versions remain.
- Since November 2025, Object Lambda is restricted to existing customers and selected partner solutions.
Trap: CORS permits a cross-origin browser request but does not grant permission to access the object.
A request through the Object Lambda Access Point invokes Lambda, with a supporting access point providing access to original data.
Lambda transforms the returned data for the application without changing the stored original.
From the AWS docs (25)
- An S3 access point is a named endpoint with its own policy and network controls for object operations.
- An access point policy works together with the underlying bucket policy.
- An access point can be restricted to requests from a virtual private cloud (VPC).
- Each access point can have its own Block Public Access settings.
- Access points support object operations rather than bucket operations such as creating replication configurations.
- CORS lets browser applications access permitted resources across origins while existing authorization policies still apply.
- An origin is the combination of protocol, host, and port; changing one makes a request cross-origin.
- A browser can send an OPTIONS preflight request to check whether a cross-origin method and headers are allowed before the actual request.
- Presigned URLs grant time-limited access to an object operation using the permissions of the signing identity.
- A presigned PUT URL permits an upload without giving the recipient AWS credentials.
- A presigned upload replaces the current object at its specified key.
- Bucket versioning retains prior versions when an upload replaces the current object.
- A presigned URL expires when its signing credentials expire or are revoked, even if its requested lifetime is longer.
- SigV4 presigned URLs created through the CLI or SDKs can last up to 7 days when signing credentials remain valid.
- Presigned URLs are bearer tokens: anyone possessing a valid URL can use its permitted operation.
- S3 Object Lambda uses Lambda to transform GET, LIST, or HEAD responses.
- An Object Lambda Access Point uses a supporting access point to access original data.
- Object Lambda can filter or redact returned data without requiring changes to the stored original.
- Since November 7, 2025, S3 Object Lambda is limited to existing customers and selected partner solutions.
- A VPC is a logically isolated virtual network in AWS.
- A VPC endpoint provides private connectivity to supported AWS services.
- AWS Signature Version 4 (SigV4) adds authentication information to AWS API requests.
- S3 Access Grants maps read-only, write-only or read-write access to S3 prefixes, buckets or objects for IAM principals and directly for users and groups from a corporate directory.
- AWS recommends a bucket policy that delegates access control to access points, so each access point policy grants access without repeating it in the bucket policy.
- To use an access point from a VPC, the S3 VPC endpoint policy must allow the access point and its underlying bucket.
Event notifications
S3 event notifications start work when selected bucketA bucket is a container for objects stored in S3. events occur. They let a file arrival trigger processing without an application repeatedly listing the bucket.
How it works
- Select an event such as
s3:ObjectCreated:*and optionally narrow it with an object-key prefixA prefix is the beginning of an object key used to select related object names. or suffixA suffix is the ending of an object key used to select matching object names.. A prefix is the beginning of the objectAn object is stored data together with metadata that describes it. name; a suffix is its ending. - Choose Lambda, standard SNS, or standard SQS in the same Region for a native notification. Grant S3 permission on the destination.
- For EventBridge routing, enable EventBridge on the bucket. All supported events go to the bus; rules there select the events and targets. This path supports first-in, first-out (FIFOFIFO is a queue model that preserves message order within each message group.) queues, which preserve message order within a message group.
- Make processing idempotent, meaning a repeated event does not duplicate its effects. S3 notifications can arrive more than once or out of order.
- Keep output writes outside the triggering bucket or input prefix so processing does not recursively trigger itself.
- For SNS or SQS, grant the S3 service principalA principal is the person, application identity, or service making an AWS request. publish or send access in the destination policy, scoped to the source bucket ARNAn ARN is an AWS identifier for a resource. and account. A customer managed KMS keyAn object key is the name used to identify an object within its bucket. also needs a key policy granting S3 data key generation and decryption.
Exam cues
- If the question says S3 arrivals must reach an SQS FIFO queue, think EventBridge, because native notifications do not support that destination.
- If the question says S3 processing must survive consumer downtime, think SQS, because the queue stores notifications until consumers can handle them.
- If the question says an S3-triggered Lambda repeatedly processes its own output, think input-prefix isolation or separate buckets, because output uploads can trigger the same notification.
- If the question says an
s3:ObjectRemoved:Deletenotification arrived, think an object deletion, not an empty bucket, because the event describes the removed object. - If a notification must report S3 Lifecycle deletions, select lifecycle expiration events, because ObjectRemoved excludes automatic lifecycle deletions.
Remember
-
s3:ObjectRemoved:DeleteMarkerCreatedreports a new delete markerA delete marker is a version that makes an object appear deleted while earlier versions can remain stored.;s3:ObjectRemoved:Deletereports an object deletion. -
Native Lambda, SNS, and SQS targets must be in the bucket Region. SNS and SQS targets must be standard types.
-
Delivery is at least once, with possible duplicates and reordering.
-
Prefix and suffix filters do not support wildcards; overlapping filters for the same event types can be invalid.
-
s3:ObjectCreated:*includes completed multipart uploadsMultipart upload sends an object in independent parts and assembles them when the upload is completed..
Trap: A FIFO destination does not retroactively make S3 emit events in object-operation order.
Enabling EventBridge sends supported S3 events to the event bus.
An EventBridge rule selects matching events and sends them to the FIFO queue.
From the AWS docs (20)
- S3 event notifications send messages when configured bucket events occur.
- Native S3 event notifications can target Lambda, standard SNS topics, or standard SQS queues in the bucket Region.
- S3 needs permission to publish to SNS or SQS or invoke the target Lambda function.
- S3 notifications are delivered at least once and can be duplicated or arrive out of order.
- Native notifications can filter object key names by prefix and suffix; wildcard characters are unsupported in those filters.
- Overlapping prefix or suffix filters for the same event types can make a notification configuration invalid.
- Native S3 notifications cannot target SQS FIFO; EventBridge can route S3 events to a FIFO queue.
- Enabling EventBridge on an S3 bucket sends all supported events to EventBridge, where rules select events and targets.
- A Lambda function writing to the same bucket and triggering prefix can invoke itself repeatedly; use separate buckets or distinct input prefixes.
- s3:ObjectCreated:* covers object creation events including completed multipart uploads.
- s3:ObjectRemoved:Delete reports an object deletion; s3:ObjectRemoved:DeleteMarkerCreated reports creation of a delete marker in a versioned bucket. Neither event means that the bucket is empty.
- SQS queues store notification messages until consumers can process them.
- AWS recommends idempotent event consumers that track processed events to handle duplicate notifications.
- A prefix selects the beginning of an object key; a suffix selects its ending.
- FIFO queues preserve message order within each message group.
- SNS and SQS destination policies grant the S3 service principal SNS:Publish or SQS:SendMessage; AWS examples restrict the source bucket ARN and account.
- For an SNS topic or SQS queue encrypted with a customer managed KMS key, its key policy must allow S3 kms:GenerateDataKey and kms:Decrypt.
- S3 ObjectRemoved notifications do not report automatic lifecycle deletions; lifecycle expiration has its own event types.
- The s3:ObjectAcl:Put event notifies when an ACL is put on an object or an existing ACL changes.
- S3 ObjectRestore event types report when a restore from S3 Glacier Flexible Retrieval or Deep Archive starts, completes, and when the temporary restored copy expires.
Artifact and log bucket management
S3 bucketsA bucket is a container for objects stored in S3. can hold source packages, build artifactsAn artifact is a file produced or consumed during a software pipeline., and delivered log files. An artifact is a file produced or consumed by a build or deployment; bucket settings must match the producing service as well as retention requirements.
How it works
- Enable versioning for a CodePipeline S3 source and provide ZIP input when downstream actions require it. The artifact bucket is separate from this source bucket; artifact encryption can use a symmetric customer managed KMS keyAn object key is the name used to identify an object within its bucket..
- Grant each log producer delivery access. An ALB log bucket must be in the load balancer Region and supports SSE-S3SSE-S3 is server-side object encryption with keys managed by S3. only. Scope CloudTrail delivery grants to the required trails with
aws:SourceArn. - Use S3 arrival events to invoke Lambda for custom parsing. Give the function source read access and destination write access; it can index records in OpenSearch or send batches to a CloudWatch Logs stream with
PutLogEvents. - Apply lifecycle transitions and expiration, including noncurrent-version actions. Protect retained log versions with ObjectAn object is stored data together with metadata that describes it. Lock compliance retention when deletion must be prevented.
- For large uploads, initiate multipart uploadMultipart upload sends an object in independent parts and assembles them when the upload is completed., send parts in parallel, and complete it. Retry failed parts individually; complete or abort abandoned uploads to stop part-storage charges.
- For distant uploaders, consider Transfer Acceleration. Enable it and use the accelerate endpoint. A checksumA checksum is a calculated value used to detect whether data changed during transfer or storage. is a calculated value used to detect data changes. Validate transfers with checksums rather than assuming an ETagAn ETag is an object content identifier that is an MD5 digest only for certain upload and encryption combinations., the object entity tag, is an MD5MD5 is a digest algorithm that produces a value used to compare data content. digest of the file.
Exam cues
- If the question says an S3 CodePipeline source must identify object revisions, think source-bucket versioning, because the source action requires it.
- If the question says S3 log files need searchable parsed records on arrival, think S3 events plus Lambda, because custom code can read and index the records.
- If the question says an S3 upload suffers repeated network interruptions, think multipart upload, because only interrupted parts need retrying.
- If the question says S3 uploads travel long distances to one bucket, think Transfer Acceleration, because edge ingress provides an optimized path.
Remember
- ALB delivery requires a same-Region destination, a delivery policy, and SSE-S3. CloudTrail has its own delivery policy.
- Lifecycle needs separate noncurrent-version actions to remove older versionsA noncurrent version is a saved object version that is no longer current..
bucket-name.s3-accelerate.amazonaws.comis the Transfer Acceleration endpoint.- S3 rejects a supplied checksum mismatch with
BadDigest; request stored checksums when downloading. - MD5 is a checksum algorithm. An ETag is the file MD5 only for qualifying single-part operations using plaintext or SSE-S3, not for multipart, SSE-KMSSSE-KMS is server-side object encryption using keys managed in AWS KMS., or SSE-CSSE-C is server-side encryption using a key that the caller must supply and retain. objects.
Trap: A single blanket encryption policy can break a log producer. ALB access log delivery supports SSE-S3 rather than SSE-KMS.
Initiation returns an upload ID used to send and independently retry parts.
CompleteMultipartUpload assembles the uploaded parts into the object.
From the AWS docs (26)
- A CodePipeline S3 source action requires versioning on its source bucket.
- An S3 source can contain a ZIP archive; a single unzipped file fails downstream actions that require ZIP input.
- CodePipeline artifact encryption supports symmetric KMS keys and can use a customer managed key.
- S3 Lifecycle can transition or expire stored objects, with separate actions for noncurrent versions.
- Application Load Balancer access logs require a destination S3 bucket policy granting log-delivery access.
- An ALB log bucket must be in the load balancer Region and can belong to another account.
- SSE-S3 is the only supported server-side encryption option for ALB access log delivery.
- CloudTrail log delivery requires an S3 bucket policy granting the service access; aws:SourceArn can restrict the grant to specific trails.
- S3 arrival events can invoke Lambda code that reads and parses a log object and indexes records into OpenSearch Service.
- The log-processing Lambda role needs source S3 read permissions and destination OpenSearch write permissions.
- PutLogEvents uploads log-event batches to a specified CloudWatch Logs log stream.
- Object Lock compliance retention protects saved log versions from deletion during their retention period.
- Multipart upload initiates an upload, uploads separate parts, then completes the object from those parts.
- Multipart upload allows parallel part uploads and retrying interrupted parts without restarting the full object.
- Uploaded multipart parts incur storage charges until the upload is completed or aborted.
- Transfer Acceleration routes distant client uploads through CloudFront edge locations and an optimized network path to S3.
- Transfer Acceleration must be enabled on the bucket and uses bucket-name.s3-accelerate.amazonaws.com.
- A checksum is a calculated value for detecting data changes; S3 compares uploaded checksums and rejects mismatches with BadDigest.
- Stored S3 checksums can be requested during downloads to verify integrity.
- For qualifying single-part PUT, POST, or Copy objects encrypted with SSE-S3 or unencrypted, the ETag is an MD5 digest of object data.
- Multipart upload ETags are not MD5 digests of the entire object, regardless of encryption.
- SSE-KMS and SSE-C objects do not have object-data MD5 ETags.
- Artifacts are files consumed or produced by pipeline actions.
- The CodePipeline artifact bucket is distinct from an S3 source bucket.
- Multipart initiation returns an upload ID used to upload, list, complete, or abort parts.
- S3 arrival events can notify a Lambda function to process a new object.
Inventory and access logging
S3 Inventory answers what objectsAn object is stored data together with metadata that describes it. exist and which metadata they have. Access logging and CloudTrail data eventsA data event records resource activity such as reading or writing an S3 object. answer what requests were made; they serve different audit needs.
How it works
- Configure Inventory for a daily or weekly report for the entire bucketA bucket is a container for objects stored in S3. or a selected keyAn object key is the name used to identify an object within its bucket. prefixA prefix is the beginning of an object key used to select related object names.. Choose current versionsThe current version is the most recent version returned when a request does not specify a version ID. or all versions and metadata such as encryption and replication status.
- Choose an output format and destination bucket for the report. Query reports with Athena when needed.
- Enable server access logging for request records with requester, time, operation, and response-status fields. A separate destination without access logging avoids a recursive stream of logs.
- Enable CloudTrail S3 data events on a trail for object-level API auditing. Select the required buckets; these events are not included in Event history.
| Requirement | Feature |
|---|---|
| Scheduled object and metadata list | S3 Inventory |
| Best-effort request traffic records | Server access logging |
| Object API activity for auditing | CloudTrail data events |
Exam cues
- If the question says an S3 audit needs a scheduled object catalog, think Inventory, because it lists objects and selected metadata.
- If the question says an S3 audit needs object read and write calls, think CloudTrail data events, because Inventory is not a request log.
- If the question says an S3 server access log is delayed or absent, think best-effort delivery, because completeness and timeliness are not guaranteed.
Remember
- Inventory can select a key prefix.
- Server access logging is off by default and can also deliver structured records to CloudWatch Logs.
- A direct S3 access-log destination cannot use Object Lock or Requester Pays.
- CloudTrail data events are off by default, incur charges, and do not appear in Event history.
Trap: Inventory lists objects and metadata; it does not record object API requests.
From the AWS docs (20)
- S3 Inventory produces daily or weekly object and metadata listings for an entire bucket or for keys with a selected prefix.
- Inventory output formats are CSV, Apache ORC, and Apache Parquet.
- Inventory can list only current versions or all object versions.
- Inventory can include encryption status and replication status metadata.
- Athena can query S3 Inventory files using SQL.
- S3 server access logging records details about requests made to a bucket.
- Server access logs are delivered on a best-effort basis, usually within hours; completeness and timeliness are not guaranteed and records may be duplicated.
- Server access logging is not enabled by default.
- Use a separate access-log destination bucket without its own access logging to avoid recursive logs.
- A server access-log destination bucket cannot have Object Lock or Requester Pays enabled.
- Server access logs can also be delivered to CloudWatch Logs in structured format.
- CloudTrail S3 data events record object-level API activity such as reads and writes.
- CloudTrail does not log data events by default; they must be enabled and incur additional charges.
- CloudTrail Event history does not include data events.
- A CloudTrail trail can be configured to record data events for selected S3 buckets or all buckets in the account.
- ORC and Parquet are column-oriented formats used to query data efficiently.
- CSV is a comma-separated-values text format.
- S3 Inventory reports can include an Object ACL field that lists the grantees and permissions of each object.
- Server access log delivery needs the destination bucket policy to grant s3:PutObject to the logging service principal logging.s3.amazonaws.com.
- S3 server access log records include fields for requester identity, request time, operation, and HTTP response status.
Batch Replication
S3 Batch Replication fills gaps that live replication does not cover, such as data written before a rule existed. It runs an explicit job over a manifestA manifest lists the object keys and, when required, version IDs that a batch job processes., a list of objectsAn object is stored data together with metadata that describes it. to process.
How it works
- Keep a replication configuration on the source bucketA bucket is a container for objects stored in S3. and give the Batch Operations job roleAn IAM role is an identity with permissions that a trusted service or caller can assume. the required replication permissions.
- Generate a manifest from that configuration or supply CSVCSV is a text report format whose values are separated by commas. or S3 Inventory. Include version IDsA version ID identifies a particular saved version of an object. in a caller-supplied manifest for versioned objects.
- Filter by replication status to select never-attempted objects, failures, previously replicated source objects, or replicasA replica is an object copy that S3 creates in a replication destination bucket..
- Run the job and request a CSV completion reportA completion report lists batch job results for the processed objects.. Fix the cause of any failure before submitting a retry.
| Status filter | Objects selected |
|---|---|
NONE |
No prior replication attempt |
FAILED |
Failed attempt |
COMPLETED |
Previously replicated source objects |
REPLICA |
Copies made from another source |
Exam cues
- If the question says an S3 bucket already holds data when CRRCross-Region Replication copies matching objects to buckets in a different AWS Region. is enabled, think Batch Replication, because live replication does not backfill.
- If the question says an S3 replication permission problem was fixed, think a job filtered to
FAILED, because it targets failed attempts. - If the question says an S3 replica was permanently deleted by version ID, think copying the source in place, because Batch Replication cannot re-replicate that deleted replica.
Remember
- An S3-generated manifest stays in the source Region.
- RTC does not apply to batch jobs.
- Glacier Flexible Retrieval and Deep Archive source objects are unsupported. RestoreA restore makes archived object data readable, temporarily for Glacier Flexible Retrieval and Deep Archive. Intelligent-Tiering archive tiers to Frequent Access first.
Trap: A Batch Replication retry does not repair the permissions that caused the failure. Correct the cause before retrying.
From the AWS docs (17)
- S3 Batch Replication copies existing, previously replicated, or failed objects on demand through a Batch Operations job.
- The source bucket must already have a replication configuration.
- The Batch Operations job role needs permissions for the underlying replication operation.
- A Batch Replication manifest lists object keys for the job.
- A supplied Batch Replication manifest can be CSV or an S3 Inventory report.
- S3 can generate a Batch Replication manifest from the replication configuration.
- A caller-supplied manifest for versioned objects must include version IDs.
- An S3-generated manifest must be stored in the source bucket Region.
- The NONE status filter selects objects for which S3 has never attempted replication.
- The FAILED status filter selects objects whose previous replication attempt failed.
- The COMPLETED status filter selects previously replicated source objects, including objects to backfill into another destination.
- The REPLICA status filter selects replicas made from another source bucket.
- A requested CSV completion report lists objects and their success or failure results.
- Replication Time Control does not apply to Batch Replication.
- Batch Replication does not support objects stored in Glacier Flexible Retrieval or Deep Archive.
- Intelligent-Tiering Archive Access and Deep Archive Access objects must be restored to Frequent Access before Batch Replication.
- Batch Replication cannot re-replicate a replica deleted from the destination by version ID; copying the source in place creates a new version that initiates live replication.
Replication Time Control and monitoring
S3 Replication Time Control (RTC) makes live replication timing measurable against a 15-minute threshold. A service-level agreementA service-level agreement defines the service commitment and conditions for service credits. (SLA) sets the conditions for service credits; it does not make asynchronous replicationAsynchronous replication copies data separately from the original write, so the copy can become available later. instantaneous.
How it works
- Enable RTC on the live replication rule when predictable timing is required. For monitoring alone, enable replication metrics independently.
- Watch bytes and operations waiting to replicate, maximum lag, and failed operations in CloudWatch.
- Subscribe to missed-threshold and replicated-after-threshold notifications to distinguish late objectsAn object is stored data together with metadata that describes it. from completed ones.
- With replication metrics enabled, route failure notifications to Lambda, SNS, or SQS. The event identifies the object version and failure reason.
- Use automation to inspect failures and submit Batch Replication retries after addressing their cause.
| CloudWatch metric | Meaning |
|---|---|
BytesPendingReplication |
Bytes waiting to replicate |
OperationsPendingReplication |
Operations waiting to replicate |
ReplicationLatency |
Maximum lag in seconds |
OperationsFailedReplication |
Failed operations |
Exam cues
- If the question says S3 replication must report when an object is late, think RTC plus
OperationMissedThreshold, because it exposes the 15-minute threshold. - If the question says S3 replication failures need automated handling, think replication metrics plus
OperationFailedReplicationto Lambda, because the notification identifies the failed version and reason.
Remember
- The SLA service-credit threshold is 99.9% per Region pair per account per month.
PENDING,COMPLETED, andFAILEDdescribe source replication; a replicaA replica is an object copy that S3 creates in a replication destination bucket. normally reportsREPLICA.- RTC does not cover Batch Replication.
Trap: An alarm reports a problem; it does not repair missing bucketA bucket is a container for objects stored in S3. or KMS permissions. A retry must address the failure cause.
With replication metrics enabled, S3 sends an OperationFailedReplication notification to Lambda.
Automation examines the failure reason and submits a Batch Replication retry after the cause is addressed.
From the AWS docs (15)
- S3 RTC adds a 15-minute replication threshold to live replication rules.
- The RTC SLA uses a monthly 15-minute replication percentage of 99.9% per Region pair per account as the service-credit threshold.
- Enabling RTC also enables replication metrics; metrics can be enabled independently of RTC.
- BytesPendingReplication measures bytes of objects waiting to replicate.
- OperationsPendingReplication counts operations waiting to replicate.
- ReplicationLatency measures the maximum lag in seconds for a replication rule.
- OperationsFailedReplication counts operations that failed to replicate.
- s3:Replication:OperationMissedThreshold reports objects missing the RTC threshold.
- s3:Replication:OperationReplicatedAfterThreshold reports replication that finishes after the RTC threshold.
- With replication metrics enabled, s3:Replication:OperationFailedReplication can notify Lambda, SNS, or SQS of failures.
- A replication failure notification contains the object key, version ID, and failureReason.
- Source replication status values include PENDING, COMPLETED, and FAILED; a replica normally reports REPLICA.
- Batch Replication can retry objects filtered by FAILED replication status.
- RTC does not apply to Batch Replication.
- AWS describes consuming replication failure events and automating retries of failed objects.
Multi-Region Access Points
An S3 Multi-Region Access Point gives applications one global endpoint for bucketsA bucket is a container for objects stored in S3. in different Regions. It chooses where a request goes; separate replication rules maintain the copies at those destinations.
How it works
- Associate regional buckets with the global endpoint and set their routing states. Active Regions can receive traffic; passive Regions cannot.
- The endpoint sends each request to the closest active bucket. It does not search other buckets for the requested keyAn object key is the name used to identify an object within its bucket..
- Configure Cross-Region ReplicationCross-Region Replication copies matching objects to buckets in a different AWS Region. (CRR) between the buckets. Use two-way replicationTwo-way replication uses rules in both directions so writes to either bucket can be copied to the other. if writes can land in different Regions, and Batch Replication for data already stored.
- Use failoverFailover shifts requests away from one Region to another Region that can serve them. controls to change the Regions receiving requests while applications keep the same global endpoint.
Exam cues
- If the question says an S3 application needs one endpoint across Regions, think a Multi-Region Access Point, because it routes requests to an active regional bucket.
- If the question says S3 recovery-region writes must return to the original Region, think two-way replication, because routing alone does not synchronize objectsAn object is stored data together with metadata that describes it..
- If the question says an S3 request returns 404 although another Region has the object, think missing regional copies, because routing does not inspect object contents.
Remember
- Active-activeIn active-active routing, all configured Regions can receive requests. lets all Regions receive traffic; active-passiveIn active-passive routing, active Regions receive requests while passive Regions do not. leaves designated Regions idle for requests.
- Versioning is required for replication.
- SSE-KMSSSE-KMS is server-side object encryption using keys managed in AWS KMS. replication needs explicit configuration and key permissions.
- Backfill existing objects before relying on a recovery copy.
Trap: A global endpoint does not guarantee that its chosen bucket already contains the requested object.
The application uses one global endpoint.
Routing selects a nearby active bucket without checking which keys it holds.
Separately configured asynchronous CRR copies matching new objects to another Region.
From the AWS docs (11)
- An S3 Multi-Region Access Point provides a global endpoint for buckets in multiple Regions.
- Requests through a Multi-Region Access Point route to the closest-proximity bucket with active routing status.
- Routing does not inspect bucket contents; a GET can return HTTP 404 when the selected bucket lacks the object.
- Multi-Region Access Points require separately configured CRR to copy data between their buckets.
- Two-way replication helps synchronize buckets that accept writes through a Multi-Region Access Point.
- Active-active routing allows all configured Regions to receive traffic.
- In active-passive routing, active Regions receive traffic and passive Regions do not.
- Failover controls change which Regions receive requests while preserving the global endpoint.
- Existing objects need Batch Replication because live CRR covers new matching objects.
- Cross-Region SSE-KMS replication requires explicit encrypted-object selection, destination KMS key configuration, and source decrypt and destination encrypt permissions.
- Replication requires versioning enabled on source and destination buckets.
S3 operational troubleshooting
S3 operations depend on configuration details that affect access, encryption, event processing, retention, and replication. Check the relevant settings and API behavior when a bucketA bucket is a container for objects stored in S3. or objectAn object is stored data together with metadata that describes it. does not behave as expected.
How it works
- For cross-origin resource sharingCORS uses HTTP headers to tell browsers which cross-origin requests a server permits. (CORS), check
AllowedOrigins,AllowedMethods, andAllowedHeaders. S3 uses the first rule matching the request originAn origin is the combination of a URL protocol, host, and port., method, and headers. For requests through a VPC endpointA VPC endpoint enables private connectivity from a VPC to supported AWS services., useaws:VpcSourceIpwithaws:SourceVpcoraws:SourceVpce; private address ranges can overlap. CORS | VPC endpoints | VPCA VPC is a logically isolated virtual network in AWS. source IP - General purpose bucket names use lowercase letters, numbers, periods, and hyphens. They cannot resemble an IP address, start with
xn--, or end with-s3alias. An account regional namespace uses the account ID, Region code, and-ansuffixA suffix is the ending of an object key used to select matching object names., and only that account can create those names. Bucket naming rules | Account regional namespace - When Bucket KeysAn object key is the name used to identify an object within its bucket. are enabled, check KMS encryption-context policy conditions: Bucket Keys use the bucket ARNAn ARN is an AWS identifier for a resource. instead of the object ARN. Enabling the setting affects new objects. For existing SSE-S3SSE-S3 is server-side object encryption with keys managed by S3. or SSE-KMSSSE-KMS is server-side object encryption using keys managed in AWS KMS. objects,
UpdateObjectEncryptioncan enable a Bucket Key in place without copying data; a copy operation is another option. The update does not accept unencrypted, DSSE-KMSDSSE-KMS applies two layers of server-side encryption using KMS keys., or SSE-CSSE-C is server-side encryption using a key that the caller must supply and retain. objects. Bucket Key policies | Bucket Key configuration | Update encryption - For event consumers, compare PUT or DELETE sequencers only for the same object key, left-padding the shorter value with zeros. Decode the URL-encoded key, where a plus sign represents a space. When updating notifications, remember that
PutBucketNotificationConfigurationreplaces the whole configuration and normally rejects the entire request if a destination test fails. Event message structure | Notification configuration API - For object tags, read the current set with
GetObjectTaggingbefore adding a tag, then write the complete set withPutObjectTagging, which replaces it. Access policies and lifecycle filters can use tags. To inspect an object’s access control listAn access control list is a set of grants attached to a bucket or object. (ACL), callGetObjectAcl. Object tagging | Tag uses | GetObjectAcl - For replication troubleshooting, inspect
x-amz-replication-statusinHeadObject. Enable replication metrics for failure notifications; threshold notifications require S3 Replication Time Control (S3 RTC). Destination lifecycle rules count age from the original object’s creation time. Replication status | Replication events | ReplicaA replica is an object copy that S3 creates in a replication destination bucket. lifecycle
Exam cues
- If a new bucket rejects SSE-C uploads, check the default block. Since April 2026 it applies to new general purpose buckets and existing buckets in accounts without SSE-C objects. Explicitly enable SSE-C with
PutBucketEncryptionandBlockedEncryptionTypesset toNONE. - If an SSE-KMS upload specifies no KMS key, think
aws/s3in the uploading account, because S3 uses that AWS managed key. - If a queue rejects S3 notifications, check
MaximumMessageSize, because notifications can reach 64 KB. - If an S3 Inventory report must go to another account, use a destination bucket in the source Region with a delivery policy. Direct S3 server access log delivery requires the same account and Region as the source. Inventory destinations | Access logging
- If lifecycle actions conflict on the same day, permanent deletion takes precedence over transition, which takes precedence over creating a delete markerA delete marker is a version that makes an object appear deleted while earlier versions can remain stored.. If retaining a count of newer noncurrent versionsA noncurrent version is a saved object version that is no longer current., specify the count, an age, and a
Filterelement. - If a destination version was deleted by version IDA version ID identifies a particular saved version of an object., copy the source object in place to create a new version eligible for live replication. Batch Replication cannot restoreA restore makes archived object data readable, temporarily for Glacier Flexible Retrieval and Deep Archive. the deleted version.
Remember
- General purpose bucket names are 3 to 63 characters long.
- An object can have up to 10 tags, each with a unique key.
- Inventory runs daily or weekly in CSVCSV is a text report format whose values are separated by commas., ORCORC is a column-oriented data format supported by S3 Inventory., or ParquetParquet is a column-oriented data format supported by S3 Inventory.; the first report can take up to 48 hours.
- S3 rechecks tags when a queued tag-filtered expiration executes and skips expiration if the triggering tag is absent.
- The S3 RTC SLAA service-level agreement defines the service commitment and conditions for service credits. does not apply while the replication transfer rate exceeds the default 1 Gbps quota.
Trap: Replication failure notifications need replication metrics, but they do not require S3 RTC. Threshold notifications do require S3 RTC.
From the AWS docs (32)
- AllowedOrigins, AllowedMethods, and AllowedHeaders define the origins, methods, and headers a CORS rule permits.
- S3 uses the first CORS rule matching the request origin, method, and requested headers.
- For requests through an S3 VPC endpoint, use aws:VpcSourceIp rather than aws:SourceIp for source-address conditions.
- Pair aws:VpcSourceIp with aws:SourceVpc or aws:SourceVpce because private IP ranges can overlap between VPCs.
- General purpose bucket names are 3 to 63 characters of lowercase letters, numbers, periods and hyphens; they cannot look like an IP address, start with xn--, or end with -s3alias.
- Buckets can optionally be created in an account regional namespace, where only the owning account can create names ending in its account ID, Region code and -an, so a chosen prefix cannot be taken by another account.
- An S3 object can carry up to 10 tags with unique keys, which access policies and lifecycle rule filters can use.
- Bucket Keys use the bucket ARN as encryption context, so policies expecting an object ARN must be updated.
- From April 2026, SSE-C is blocked by default for new general purpose buckets and existing buckets in accounts without SSE-C objects.
- To enable SSE-C explicitly, use PutBucketEncryption with BlockedEncryptionTypes set to NONE.
- UpdateObjectEncryption changes an existing SSE-S3 or SSE-KMS object to SSE-KMS, or to use an S3 Bucket Key, in place without copying data; it does not accept unencrypted, DSSE-KMS or SSE-C objects.
- Turning on Bucket Keys for a bucket does not change existing objects. For eligible SSE-KMS objects, UpdateObjectEncryption enables a Bucket Key without copying object data.
- When no KMS key is given for SSE-KMS, S3 uses the AWS managed key aws/s3 of the uploading account.
- PUT and DELETE notification sequencers can order events for the same object key, but not different keys; left-pad shorter values with zeros before comparison.
- S3 notifications can be up to 64 KB; SQS rejects a notification exceeding the queue's configured MaximumMessageSize.
- PutBucketNotificationConfiguration replaces the whole configuration and normally validates destinations atomically; a failed test causes the entire request to fail.
- The object key in an S3 event notification is URL encoded, with spaces represented by plus signs.
- For direct delivery to S3, the access-log destination bucket must be in the same Region and account as the source.
- An Inventory destination bucket must be in the source Region, may be in a different account, and needs a delivery bucket policy.
- S3 Inventory reports object names and selected metadata on a daily or weekly schedule.
- Inventory output formats are CSV, Apache ORC, and Apache Parquet.
- The first Inventory report can take up to 48 hours to arrive.
- When lifecycle actions conflict on the same day, permanent deletion takes precedence over transition, which takes precedence over creating a delete marker.
- Noncurrent version expiration can retain a specified number of newer noncurrent versions; specifying that count requires a Filter element as well as an age.
- S3 rechecks tags when a queued tag-filtered expiration executes and skips the expiration if the triggering tag is absent.
- PutObjectTagging replaces the whole tag set of an object, so adding one tag means reading the set with GetObjectTagging, changing it and putting it back. An object can hold up to 10 tags.
- The GetObjectAcl API returns the access control list (ACL) of an object.
- The S3 RTC SLA does not apply when the replication transfer rate exceeds the default 1 Gbps quota.
- Batch Replication cannot restore a destination version deleted by version ID; copying the source object in place creates a new version eligible for live replication.
- HeadObject exposes source replication status as PENDING, COMPLETED or FAILED in the x-amz-replication-status header.
- Destination lifecycle rules use the original object's creation time, not the time the replica arrived.
- Replication failure notifications require replication metrics on the rule, but not S3 Replication Time Control; threshold notifications require S3 RTC.
Flashcards
Nothing to review here right now. Pick other topics or another filter.
·
All 169 cards as a list
What kind of storage is Amazon S3, and what are its two main building blocks?
Object storage. Data is stored as objects (a file plus metadata, identified by a key) inside buckets.
What durability is Amazon S3 designed for?
99.999999999% (11 nines).
How does Amazon S3 Standard protect objects against the loss of a data center?
It stores objects redundantly across at least three Availability Zones in the Region.
A new Amazon S3 bucket is created without any policy. Can anyone on the internet read its objects?
No. Buckets and objects are private until access is explicitly granted.
Which use cases fit Amazon S3 object storage?
Data lakes, static websites, mobile app content, backup and restore, archives and big data analytics.
How can a company recover an Amazon S3 object that someone overwrote by mistake?
Turn on S3 Versioning. It keeps earlier versions in the bucket, so the old version can be restored.
Users around the world upload large files to one Amazon S3 bucket and uploads are slow. Which S3 feature speeds them up?
S3 Transfer Acceleration, which routes uploads through CloudFront edge locations.
Does Amazon S3 store objects in a folder hierarchy like a file system?
No. S3 is flat; folders in the console are shared key name prefixes.
An S3 audit needs object reads recorded while a scheduled job lists bucket contents. Which features fit?
Enable CloudTrail S3 data events for object activity and S3 Inventory for the listing. Data events are not logged by default; Inventory lists objects and metadata rather than access calls.
A 3 GB upload to Amazon S3 over a flaky network keeps restarting from zero. What should the client use?
Multipart upload, so only the failed parts are retried.
An Amazon S3 bucket has Block Public Access turned off but no policy or ACL grants public reads. Does turning off the block make its objects public?
No. Public reads still need an applicable policy or ACL grant.
Objects written to one S3 bucket must appear automatically in buckets in three other Regions. What provides this without custom copy code?
S3 Replication from the source bucket to each destination bucket. One source can replicate to multiple destination buckets across Regions.
A company's data is accessed in unpredictable patterns. Which Amazon S3 storage class moves it between tiers automatically without retrieval fees?
S3 Intelligent-Tiering.
Monthly reports are read about once a month but must open in milliseconds. Which Amazon S3 storage class fits at lower cost than S3 Standard?
S3 Standard-IA, with a 30-day minimum and per-GB retrieval fees.
A company must keep compliance archives for 10 years and reads them less than once a year; retrieval within hours is fine. Which Amazon S3 storage class is cheapest?
S3 Glacier Deep Archive, with retrieval in hours and a 180-day minimum.
Medical images are rarely read, but when needed they must open in milliseconds. Which Amazon S3 archive class fits?
S3 Glacier Instant Retrieval, for data accessed about once a quarter with millisecond access.
A company stores thumbnails it can regenerate from originals. Which Amazon S3 storage class is cheapest for infrequent access?
S3 One Zone-IA. It keeps data in one AZ, which is acceptable for re-creatable data.
Is S3 One Zone-IA as resilient as S3 Standard-IA?
No. One Zone-IA keeps data in a single AZ and is not resilient to that AZ's loss; Standard-IA uses at least three AZs.
How do Amazon S3 Standard and S3 Standard-IA differ in their intended access patterns?
S3 Standard is for frequently accessed data. S3 Standard-IA is for long-lived data accessed infrequently and charges retrieval fees.
What should a company consider before using Amazon S3 One Zone-IA for infrequently accessed data?
S3 One Zone-IA stores data in one Availability Zone and is not resilient to its loss. It suits data that can be recreated.
Which Amazon S3 storage class suits a latency-sensitive application that needs fast access to objects?
S3 Express One Zone provides single-digit millisecond access in one Availability Zone.
Which Amazon S3 storage class suits data with unpredictable access patterns?
S3 Intelligent-Tiering is designed for unknown or changing access patterns. It charges monitoring and automation fees but no retrieval fees.
What happens when an S3 object is deleted without a version ID in a versioning-enabled bucket?
S3 adds a delete marker. Earlier versions remain; deleting the current marker by its version ID exposes the preceding version. A version-specific delete permanently removes that version.
Does suspending S3 bucket versioning erase existing versions?
No. Existing versions remain. New writes use the null version ID and can replace that null version. You can suspend versioning after enabling it, but cannot return to an unversioned bucket.
An S3 application overwrote a configuration file in a versioning-enabled bucket. What enables recovery?
Retrieve a previous object version by its version ID. Versioning retains separate versions rather than replacing the earlier data.
An S3 object disappeared after a delete without a version ID in a versioning-enabled bucket. How can an operator restore normal reads?
Delete the current delete marker by its version ID. This exposes the preceding object version.
An S3 team wants an MFA code required before permanently deleting object versions. Who enables this protection, and how?
The bucket owner root account enables MFA delete through the CLI or API. It also protects changes to versioning state and cannot be combined with lifecycle configurations.
What must be configured for S3 replication between buckets in different AWS accounts?
Versioning on both buckets, a source replication rule, and an IAM role S3 can assume. The destination bucket policy must grant replication access. CRR crosses Regions; SRR stays in one Region.
Does S3 replication copy deletions of object versions and delete markers?
Version-specific deletions are not replicated. V2 rules do not replicate delete markers by default; enable it for rules without tag filters. V1 rules replicate markers from user deletes. Lifecycle actions are not replicated.
Which objects or settings does S3 live replication leave out?
Pre-existing objects, replicas from another rule, archived source objects in Glacier Flexible Retrieval or Deep Archive and Intelligent-Tiering archive access tiers, lifecycle actions, and bucket configurations.
How does a team replicate SSE-KMS encrypted S3 objects across Regions?
Opt in to encrypted-object replication and specify a KMS key in the destination Region. Give the role source kms:Decrypt and destination kms:Encrypt permissions.
An S3 design replicates bucket A to B and B to C, but objects from A never reach C. What should change?
Configure A to replicate directly to the required destinations. Live replication does not copy replicas created by another rule.
An S3 team copies immutable object versions into a recovery account. What must be enabled on the destination?
Object Lock and versioning. The replication role also needs source s3:GetObjectRetention and s3:GetObjectLegalHold permissions.
An S3 replication workload needs a measurable replication time target and CloudWatch metrics. What should the team enable?
Enable S3 Replication Time Control. It replicates 99.9 percent of uploaded objects within 15 minutes and enables replication metrics.
How can a team retry failed S3 replicas or replicate objects that existed before a rule?
Run S3 Batch Replication with an existing replication configuration and a manifest.
A team replicates SSE-KMS encrypted S3 objects to another account. Which destination key should it use?
Use a customer managed KMS key owned by the destination account and grant the replication role access in the key policy.
Can a cross-account S3 replication destination use Requester Pays?
No. Cross-account destination buckets cannot be Requester Pays buckets.
An administrator cannot configure an S3 replication role for a bucket. Which permissions must the caller have?
The caller needs s3:PutReplicationConfiguration and iam:PassRole.
An SSE-S3 encrypted S3 object is replicated to a bucket whose default encryption is SSE-KMS. Which encryption type does the replica use?
SSE-S3. Replicas of encrypted objects use the source encryption type, not the destination default.
Global users need faster reads of a website; separately, a copy of all data must exist in a second Region. Which feature serves which need?
CloudFront caches content at edge locations for the reads. Cross-Region Replication keeps a full, asynchronous copy in another Region; it does not cache near users.
An Amazon S3 application writes new objects to buckets in two Regions and needs the writes copied to the other Region. What replication setup is needed?
Two-way Cross-Region Replication rules, with each bucket configured as a source for the other direction.
An Amazon S3 source uses Standard, but its recovery replicas should use a different storage class. How should replication be configured?
Specify the destination storage class in the replication rule. Without an override, replicas use the source storage class.
Several Amazon S3 buckets hold logs in one AWS Region. Which replication option can consolidate them for processing in one bucket?
Same-Region Replication, with rules from the source buckets to the common destination.
An Amazon S3 dataset serves readers in two distant geographic locations. How can the architecture keep a regional copy closer to each group?
Use Cross-Region Replication to maintain copies in Regions closer to the readers.
An Amazon S3 replication rule must copy only keys under reports/ tagged approved=yes. How should it select the objects?
Combine the reports/ key-prefix filter and approved=yes tag filter. The object must match both.
Which S3 controls move aging objects, delete old versions, and remove abandoned upload parts?
Lifecycle transition and expiration rules, separate NoncurrentVersion actions, and AbortIncompleteMultipartUpload. New or modified configurations block transitions below 128 KB by default. Minimum storage-duration charges still apply.
Can S3 Lifecycle move a Deep Archive object back to S3 Standard?
No. Restore a temporary readable copy, then copy the object with S3 Standard as the target storage class.
A versioning-enabled S3 log bucket still incurs storage costs after its current objects expire. Which rule is missing?
NoncurrentVersionExpiration. Current-version expiration adds delete markers and retains the older versions.
An S3 team wants day-zero transitions from Standard into Standard-IA. Is the old 30-day waiting rule still required?
No. S3 removed that wait in July 2026; a rule can use 0 days. Standard-IA still bills a minimum 30-day storage duration.
An S3 backup can be recreated after an Availability Zone is lost. Which infrequent-access class trades zone resilience for lower cost?
S3 One Zone-IA. It stores data in one Availability Zone; Standard-IA stores data across multiple zones.
An S3 dataset has unpredictable access patterns. Which storage class moves eligible objects between access tiers automatically?
S3 Intelligent-Tiering. It monitors access and moves eligible objects for a monitoring and automation fee. Objects smaller than 128 KB remain in Frequent Access.
An S3 team needs logs to remain immediately readable after 90 days with no access in Intelligent-Tiering. Which automatic tier fits, and what should it avoid enabling?
Archive Instant Access is automatic after 90 days without access. Do not confuse it with the optional Archive Access tier, which requires archive restoration.
A team adds an S3 Lifecycle expiration rule to a bucket containing old logs. Does it apply only to new logs?
No. Lifecycle rules apply to existing objects as well as new objects.
Can an S3 bucket policy deny prevent an S3 Lifecycle rule from deleting objects?
No. Bucket policy denies do not prevent lifecycle deletions or transitions.
Does S3 Lifecycle guarantee that an object is physically removed at its expiration time?
No. Removal can occur after the expiration date.
An S3 Lifecycle rule must expire only objects that a scanning process marked with the tag status=processed. How does the rule select them?
Use a tag filter with key status and value processed. Both must match exactly; other tags on the object do not prevent the match.
A team must add the same tag to millions of existing S3 objects listed in an inventory report. Which feature does it in one managed job?
S3 Batch Operations with a replace-tags operation over the supplied object list.
Which Amazon S3 lifecycle transition is supported: One Zone-IA to Standard-IA, Intelligent-Tiering to Standard, or Standard-IA to One Zone-IA?
Standard-IA to One Zone-IA. Transitions only go toward colder classes. One Zone-IA can only move to Glacier Flexible Retrieval or Deep Archive, and no class transitions back to Standard.
Do S3 lifecycle transitions cost anything by themselves?
Yes. Each transitioned object is billed as a transition request, so moving many tiny objects can cost more than it saves. Data uploaded from the internet, by contrast, has no transfer charge.
An Amazon S3 Lifecycle rule puts objects in Standard-IA on day 0 and schedules another transition on day 10. What prevents this schedule?
A single rule must allow the 30-day minimum storage duration in Standard-IA before transitioning again. Separate rules can move objects sooner with remaining-duration charges.
Which Amazon S3 Lifecycle destinations are available for an object in Glacier Instant Retrieval?
S3 Glacier Flexible Retrieval or S3 Glacier Deep Archive.
An Amazon S3 bucket stores several projects, but only keys beginning with logs/ should expire. How should Lifecycle select them?
Use a key-prefix filter for logs/. An empty filter would apply the rule to the entire bucket.
An Amazon S3 GET reads an Intelligent-Tiering object in Archive Instant Access. Which tier holds the object afterward?
Frequent Access. Reading an object in Archive Instant Access or Infrequent Access automatically moves it back.
An Amazon S3 workload has changing access patterns. Does the Intelligent-Tiering monitoring fee mean it always costs more than fixed storage classes?
No. Automatic moves to cheaper tiers can offset the monitoring fee; compare expected access patterns and tier savings.
New Amazon S3 objects will be rarely accessed from the start. Must they first be uploaded to Standard before using Standard-IA or One Zone-IA?
No. Select Standard-IA or One Zone-IA during upload. Standard is the default when no class is specified.
What are the normal S3 Glacier retrieval options and times?
Instant Retrieval: GET immediately. Flexible Retrieval: Expedited typically 1 to 5 minutes below 250 MB, Standard 3 to 5 hours, Bulk 5 to 12 hours. Deep Archive: Standard within 12 hours or Bulk within 48 hours; no Expedited. These are typical times.
How can an S3 team reserve Expedited retrieval capacity for Glacier Flexible Retrieval?
Purchase provisioned capacity. Each unit supports at least three Expedited retrievals per 5 minutes and up to 300 MB/s. Without it, high demand can cause rejection.
An S3 archive must be readable in milliseconds whenever requested. Which Glacier class should a team choose?
Glacier Instant Retrieval. Flexible Retrieval and Deep Archive require restoration before reads.
An S3 team must read Deep Archive data within minutes. Can provisioned Expedited capacity satisfy this requirement?
No. Deep Archive has no Expedited option. Standard typically finishes within 12 hours; choose a faster-access storage class before such a requirement arises.
An S3 team restores a Deep Archive object and needs to retain it in Standard permanently. What comes after restoration?
Copy the restored object with S3 Standard as its storage class. The restore alone creates a temporary readable copy and leaves the archive unchanged.
An Amazon S3 Glacier Flexible Retrieval archive can wait for Bulk retrieval. What retrieval charge applies to that option?
Bulk retrieval is free for S3 Glacier Flexible Retrieval.
An Amazon S3 archive design compares retrieval options for Glacier Flexible Retrieval and Deep Archive. Does the selected restore tier affect charges?
Yes. Retrieval charges depend on the selected tier; include the required restore speed when estimating costs.
What distinguishes S3 Object Lock governance and compliance retention modes?
Governance permits bypass with s3:BypassGovernanceRetention plus the bypass header. Compliance prevents even root from deleting the protected version or shortening retention. Both protect individual versions in a versioning-enabled bucket.
An S3 audit has no known end date. Which Object Lock control keeps selected versions protected until an authorized person releases them?
A legal hold. It has no expiry and requires s3:PutObjectLegalHold to remove. It is independent of a retention period, so both can protect the same version.
An S3 backup must resist deletion by a compromised administrator, including root, for a fixed retention period. What design fits?
Object Lock compliance mode with versioning and a bucket default retention for the required period. Compliance retention cannot be shortened by root.
An S3 version protected by Object Lock disappears from ordinary GET requests after a successful DELETE. Was the locked data removed?
Not necessarily. A DELETE without a version ID can add a delete marker while leaving the protected version intact.
An S3 team enabled Object Lock on an existing bucket and wants to suspend versioning. Is this supported?
No. Once Object Lock is enabled, it cannot be disabled and versioning cannot be suspended.
What encryption does S3 apply to a new upload with no encryption settings?
SSE-S3 using AES-256 and S3-managed keys. It has applied to new uploads since January 2023 at no extra charge and cannot be disabled. It does not retroactively encrypt older objects.
An Amazon S3 bucket uses SSE-KMS. What do Bucket Keys change, and how can existing eligible objects use them?
Bucket Keys reduce KMS calls, with up to 99% request-cost savings. UpdateObjectEncryption can enable them on existing eligible SSE-KMS objects without copying. The context becomes the bucket ARN; DSSE-KMS is unsupported.
An S3 bucket must reject uploads that do not explicitly request SSE-KMS. What enforces this?
A bucket-policy deny on PutObject when s3:x-amz-server-side-encryption is not aws:kms. Default encryption alone can be overridden. A strict header requirement also rejects uploads that omit the header.
An S3 team wants another account to read SSE-KMS objects. Is the AWS managed S3 key sufficient?
No. Use a customer managed KMS key with access configured for the other account.
What must a caller retain to read an S3 object encrypted with SSE-C?
The original encryption key and the mapping to its object or version. Supply the key over HTTPS on retrieval; S3 does not keep it, and losing it loses access.
S3 requests on an SSE-KMS bucket start failing with throttling at high request rates. Why?
Each SSE-KMS object request calls KMS and counts against the KMS request quota. S3 Bucket Keys or SSE-S3 cut those calls.
Data must be encrypted with a company-specific algorithm and keys before AWS ever receives it. Which S3 encryption option fits?
Client-side encryption. The client encrypts the data before upload, so S3 stores only ciphertext. SSE-S3, SSE-KMS and SSE-C all encrypt on the S3 side with AES-256.
With SSE-S3, does every object in a bucket share one encryption key?
No. Each object gets its own unique key, and S3 encrypts those keys with a root key that it rotates regularly.
S3 administrators must manage a bucket of SSE-KMS objects without being able to read the data. How?
Give them S3 permissions but no kms:Decrypt on the KMS key. Reading an SSE-KMS object needs both S3 access and kms:Decrypt.
A bucket in eu-west-1 must use SSE-KMS with a customer managed key that lives in us-east-1. Will that work?
No. The KMS key must be in the same Region as the bucket. Use a KMS key in eu-west-1.
An Amazon S3 audit sees fewer AWS KMS CloudTrail records after Bucket Keys are enabled for SSE-KMS. What explains the change?
Bucket Keys reduce calls to KMS, so fewer corresponding KMS events appear in CloudTrail.
An Amazon S3 application needs SSE-KMS within one account without creating a customer managed key. Which key can it use?
The AWS managed KMS key aws/s3.
An Amazon S3 bucket changes its default encryption to SSE-KMS. Does that change objects already stored there?
No. Existing objects keep their encryption settings; changing the default does not update them.
How can an S3 administrator block public access even when bucket-level permissions grant it?
Enable Block Public Access at the account level, or enforce it through an organization policy. The most restrictive applicable organization, account, bucket, and access point settings apply.
What is the default S3 Object Ownership setting for a new bucket?
Bucket owner enforced. ACLs are disabled, the bucket owner owns all objects, and policies control access.
A legacy S3 object uses authenticated-read. Does this restrict readers to the bucket owner organization?
No. Its AuthenticatedUsers grant includes any authenticated AWS account, subject to other applicable controls.
An S3 client can upload without an ACL but fails with AccessControlListNotSupported when sending a custom ACL. What setting explains this?
Bucket owner enforced Object Ownership. It accepts uploads without ACLs or with bucket-owner-full-control, and rejects other ACLs.
An S3 security requirement says to reject unencrypted network transport from clients. What policy condition fits?
An explicit deny when aws:SecureTransport is false. Exclude AWS service principals with aws:PrincipalIsAWSService false when service-to-service network context could be redacted.
An S3 static website must accept HTTPS at its website endpoint. Is that endpoint sufficient?
No. S3 website endpoints support HTTP and publicly readable content. Use CloudFront for HTTPS delivery.
A security team wants a list of S3 buckets that are public or shared with other AWS accounts, reviewed from the S3 console. Which feature fits?
IAM Access Analyzer for S3. It lists buckets with public or cross-account access and can block public access on a bucket in one step.
What problem do S3 access points solve for teams sharing a bucket?
They provide named endpoints with distinct access policies and network controls. A point can be VPC-only and has its own Block Public Access settings; its policy works with the bucket policy.
Does enabling CORS grant a browser access to a private S3 object?
No. CORS controls browser cross-origin behavior. Existing ACLs and policies still determine authorization.
An S3 presigned URL was given a long lifetime but fails when its signing role session expires. Why?
Its lifetime cannot exceed the signing credentials. Expiration or revocation of those credentials ends access even if the URL expiration is later.
An S3 application must let a recipient upload one object without receiving AWS credentials. What feature fits?
A presigned PUT URL. Its permissions are limited by the signer and its validity by the URL and signing credentials; protect it as a bearer token.
An existing S3 Object Lambda customer needs a redacted GET response without editing the source object. Which mechanism fits?
An Object Lambda Access Point invokes Lambda to transform the response through a supporting access point. Since November 2025, availability is limited to existing customers and selected partner solutions.
Thousands of corporate directory users each need access to their own S3 prefixes, beyond what bucket policies scale to. Which S3 feature fits?
S3 Access Grants, which maps prefix-level permissions to IAM principals and to directory users and groups.
An S3 log bucket must trigger file processing on arrival. Which notification choices fit?
Native notifications to Lambda, standard SNS, or standard SQS in the same Region, or EventBridge routing. Grant destination permissions; SQS can retain messages for later processing.
When should an S3 team choose EventBridge over native object event notifications?
Use EventBridge for rules over the event payload and additional targets such as SQS FIFO. Native notifications support prefix and suffix filtering and deliver at least once, with possible duplication or reordering.
Can S3 send a native event notification directly to an SQS FIFO queue?
No. Enable EventBridge on the bucket and route the event to the FIFO queue through an EventBridge rule.
An S3-triggered Lambda writes processed results back to the input prefix and keeps invoking itself. What should change?
Write outputs to a separate bucket or use a trigger limited to a distinct input prefix. This breaks the notification loop.
An S3 consumer receives the same object event twice. What application behavior is required?
Idempotent processing: applying the same event again should not repeat its side effects. Track processed events rather than assuming exactly-once delivery.
An application receives an S3 ObjectRemoved:Delete notification. Does that mean the bucket is empty?
No. It signals an object deletion, not that all objects in the bucket have been removed.
An S3 bucket sends notifications to an SNS topic in another account. What should the topic policy allow?
Allow the S3 service principal SNS:Publish, restricted by the source bucket ARN and account.
S3 cannot publish notifications to a customer-managed KMS encrypted SQS queue despite an allowing queue policy. What else is needed?
Grant the S3 service principal kms:GenerateDataKey and kms:Decrypt in the KMS key policy.
Will an S3 ObjectRemoved notification alert a consumer when S3 Lifecycle automatically deletes an object?
No. Subscribe to lifecycle expiration events for automatic lifecycle deletions.
A security team wants a Lambda function to run whenever someone changes an object's ACL in an S3 bucket. Which S3 event type does it subscribe to?
s3:ObjectAcl:Put, which fires when an ACL is put on an object or changed.
An app must start processing as soon as a Glacier Deep Archive restore finishes. Which S3 event type fits?
s3:ObjectRestore:Completed, sent to SNS, SQS, Lambda or EventBridge.
What S3 settings matter when CodePipeline reads a source archive and stores encrypted build artifacts?
Enable versioning on the S3 source bucket. Package ZIP input for actions that expect it. CodePipeline supports symmetric KMS artifact encryption; lifecycle rules can remove aging objects and noncurrent versions.
New ALB or CloudTrail files in S3 must be parsed automatically and aged into cheaper storage before deletion. Which controls fit?
Configure service-specific delivery permissions, S3 arrival notifications to Lambda, and lifecycle transition and expiration rules. Lambda can parse and index records into OpenSearch. ALB delivery supports SSE-S3 only.
A distant S3 client repeatedly restarts a large artifact upload after network errors. What upload mechanism reduces repeated work?
Multipart upload. Upload parts in parallel and retry interrupted parts independently; complete or abort the upload to stop charges for unfinished parts.
Worldwide clients upload artifacts into one S3 Region and network distance is a bottleneck. Which bucket feature can help?
Transfer Acceleration. Enable it and use the s3-accelerate endpoint so uploads enter through CloudFront edge locations and an optimized path to S3.
An S3 team compares a multipart upload ETag with a local full-file MD5 and sees a mismatch. Is that proof of corruption?
No. A multipart ETag is not the full-object MD5. Use S3 checksum validation and request stored checksums during download.
A security policy requires SSE-KMS for every S3 log destination, but ALB log delivery must continue. What compatibility issue matters?
ALB access log delivery supports SSE-S3 only. Its destination encryption requirement must account for this service-specific limitation.
How can custom processing forward parsed S3 log records into CloudWatch Logs?
Read and parse the arriving S3 object in Lambda, then use PutLogEvents to upload batches into a destination log stream with the required access. This is custom forwarding code.
What are the stages of an S3 multipart upload?
Initiate the upload, upload the parts, then complete the upload to assemble the object. An upload ID identifies the operation; unfinished parts incur storage charges until completion or abort.
An S3 team needs a weekly list of every version with encryption and replication metadata for SQL analysis. What should it configure?
S3 Inventory with all versions and the required metadata, in CSV, ORC, or Parquet format. Athena can query the reports.
An S3 incident investigation found a missing server access-log record. Does server access logging guarantee every request is recorded promptly?
No. Delivery is best effort; records can be delayed, missing, or duplicated. Enable CloudTrail data events when object API activity needs auditing.
An S3 auditor sees no GetObject calls in CloudTrail Event history. Where should object activity be configured?
Enable S3 data events on a trail for the required buckets. Data events are disabled by default and are not shown in Event history.
A company must audit the ACLs of millions of objects on a schedule without calling an API per object. Which S3 feature fits?
S3 Inventory with the Object ACL field in its reports.
Server access logging is on, but no logs reach the destination bucket. What permission is often missing?
A destination bucket policy that allows s3:PutObject for the logging.s3.amazonaws.com service principal.
An Amazon S3 auditor needs a weekly object listing limited to keys beginning with finance/. How can Inventory select them?
Set the Inventory source prefix to finance/; the report can cover a prefix rather than the whole bucket.
An Amazon S3 investigation needs requester identity, time, operation, and response status for bucket requests. Which records contain those fields?
S3 server access logs contain those request fields.
Some S3 objects have FAILED replication status after a permissions error. How can a team retry them after fixing access?
Run S3 Batch Replication with the FAILED status filter. Review the completion report for remaining failures.
Does a new S3 live replication rule copy objects already stored in the source bucket?
No. Use Batch Replication to backfill existing objects. Keep the source replication configuration in place; RTC does not apply to the batch job.
An S3 team adds another recovery destination for objects already copied elsewhere. Which Batch Replication filter helps backfill it?
COMPLETED. It selects source objects that S3 has already replicated.
A versioned S3 bucket needs selected historical versions copied through Batch Replication. What must its custom CSV manifest identify?
Object keys and version IDs. Only the selected versions are replicated.
An S3 recovery replica was deleted by version ID. Can Batch Replication simply recreate that same replica?
No. Copy the source object in place to create a new source version and initiate live replication.
What does S3 Replication Time Control add to live replication?
A 15-minute replication threshold, replication metrics, and events for missed and later-completed replication. The SLA service-credit threshold is 99.9% per Region pair per account per month; it is not a guarantee for every object.
How can an S3 operator distinguish replication backlog, lag, and failures?
Use CloudWatch BytesPendingReplication, OperationsPendingReplication, ReplicationLatency, and OperationsFailedReplication. Inspect per-object replication status and failure events for diagnosis.
An S3 team needs predictable cross-Region replication timing and notifications when copying is late. What should it enable?
S3 RTC on the CRR rule and notifications for s3:Replication:OperationMissedThreshold. RTC enables replication metrics but does not cover Batch Replication.
An S3 team wants failed replication attempts to trigger automated retries. What event and retry mechanism fit?
Enable replication metrics and send s3:Replication:OperationFailedReplication events to Lambda. Automation can inspect failureReason, correct or escalate the cause, and submit Batch Replication retries for failed objects.
Does S3 RTC promise every new object will reach its destination in 15 minutes?
No. It has a monthly SLA percentage, and missed-threshold events exist for late objects. Batch Replication is outside RTC.
An S3 team needs SSE-KMS objects, including existing objects, in another Region behind one endpoint. What is required?
Use a Multi-Region Access Point for routing. Enable versioning and CRR, opt in to encrypted-object replication, and grant source decrypt and destination encrypt access. Use Batch Replication for existing data.
Does an S3 Multi-Region Access Point replicate objects or find the bucket that already holds a requested key?
No. It routes to a nearby active bucket without checking contents. Configure replication separately; a routed GET can return 404 if that bucket lacks the object.
An application writes through an S3 Multi-Region Access Point and needs existing data available before recovery-region failover. What copying setup fits?
Two-way CRR for new writes and Batch Replication for existing objects. The global endpoint handles routing and does not copy data.
An S3 team wants a recovery Region to receive no normal traffic through its Multi-Region Access Point. Which routing arrangement fits?
Active-passive. Keep the recovery Region passive, then use failover controls to change routing when needed.
An S3 Multi-Region Access Point returns 404 although another associated bucket has the object. What explains it?
Routing chose a nearby active bucket without checking object placement. Configure CRR and backfill existing objects before relying on regional copies.
A browser application on another origin cannot make an authorized S3 request. What CORS settings should be checked?
AllowedOrigins, AllowedMethods, and AllowedHeaders. S3 uses the first rule that matches all required request attributes.
An S3 bucket policy restricts endpoint requests by private IP address. Which conditions prevent matching another VPC with the same private range?
Use aws:VpcSourceIp together with aws:SourceVpc or aws:SourceVpce. Do not use aws:SourceIp for requests through the endpoint.
Is MyBucket_2026 a valid S3 general purpose bucket name?
No. Names use only lowercase letters, numbers, periods and hyphens, 3 to 63 characters, so uppercase letters and underscores are rejected.
A team wants bucket names that no other AWS account can ever claim first. Which S3 option gives that?
The account regional namespace: names carry the account ID, Region code and -an suffix, and only that account can create them.
How many tags can one S3 object have?
Up to 10, each with a unique key.
An S3 team enabled Bucket Keys and KMS policies scoped to object ARNs stopped working. What should it inspect?
Encryption-context conditions. Bucket Keys use the bucket ARN, so policies that require an object ARN must be updated.
A new S3 bucket rejects SSE-C uploads although the caller supplies the correct key. Which dated default should be checked?
SSE-C is blocked by default from April 2026 for new general purpose buckets. Explicitly enable it through PutBucketEncryption with BlockedEncryptionTypes set to NONE if required.
Existing SSE-S3 objects must move to SSE-KMS with a Bucket Key without rewriting them. Which operation does that?
UpdateObjectEncryption, which updates the encryption in place with no data copy. CopyObject also works but rewrites each object.
An object is uploaded with SSE-KMS but no key ID, and the bucket default names no key. Which key encrypts it?
The AWS managed key aws/s3 of the uploading account.
An S3 event consumer maintains an index and receives out-of-order PUT events for one key. How can it identify the later event?
Compare sequencer values for that key after left-padding the shorter value with zeros. Sequencers cannot order different keys.
An S3 notification is rejected by an SQS queue with a small message limit. What limit should the team check?
Set the queue MaximumMessageSize to at least 64 KB. Notifications can be up to that size.
A PutBucketNotificationConfiguration request adds an SQS target but an existing SNS destination fails validation. Is the new configuration saved?
No. Normal destination validation is atomic and failure rejects the whole request. The operation replaces the whole configuration.
An S3 event consumer sees a plus sign in an object key that contains a space. How should it interpret the key?
Decode it as a URL-encoded key; a plus sign represents a space.
An S3 team sends server access logs directly to a different account log bucket. What destination requirement prevents this?
Direct S3 server access-log delivery requires the same account and Region as the source bucket. Inventory has a different rule and can use another account in the same Region.
Which report schedules and file formats does S3 Inventory support?
Daily or weekly reports in CSV, ORC, or Parquet. The first report can take up to 48 hours.
An S3 object qualifies for permanent deletion and a lifecycle transition on the same day. Which action wins?
Permanent deletion takes precedence over transition.
An S3 lifecycle rule must retain a number of newer noncurrent versions before deleting older versions. What configuration is required?
Specify the noncurrent age, the number of newer versions to retain, and a Filter element.
A queued S3 Lifecycle expiration uses a tag filter, but the triggering tag is removed before execution. What happens?
S3 rechecks the tags and skips expiration if the triggering tag is absent at execution.
A Lambda function adds a tag to S3 objects that already carry other tags, and the old tags disappear. Why?
PutObjectTagging replaces the whole tag set. Read the current set with GetObjectTagging, add the new tag and put the full set back.
Which S3 API call returns the ACL of a single object?
GetObjectAcl.
An S3 Replication Time Control workload exceeds its default transfer rate quota. Does the SLA apply during that period?
No. The SLA does not apply when the replication transfer rate exceeds the default 1 Gbps quota.
A destination S3 object version was deleted by version ID. Can S3 Batch Replication recreate that version?
No. Copy the source object in place to create a new source version eligible for live replication.
How can an operator distinguish pending from failed S3 object replication?
Use HeadObject and inspect the x-amz-replication-status header.
When does an S3 destination lifecycle rule start counting the age of a replicated object?
From the original object's creation time, not its arrival at the destination.
A team wants S3 replication failure notifications without a replication time target. Must it enable S3 Replication Time Control?
No. Enable replication metrics on the rule and configure the failure notification. S3 RTC is required for threshold notifications.