Amazon S3 for developers: DVA-C03 study notes and practice questions

What to know for the exam

S3 developer questions describe an upload or download that fails, costs too much, or behaves oddly. Match the symptom to the feature, then check the rule that causes it.

Presigned URLs

• A presigned URL grants temporary access to one object, for GET or PUT, without giving the client credentials.
• It stops working at the set expiry (at most 7 days with Signature Version 4), or earlier if the credentials that signed it expire first. A URL signed with a role’s temporary credentials can end sooner than the time you set.
• The URL is only as good as its signature. Changing the method, key or signed headers breaks it.
• Browser uploads also need a CORS rule on the bucket that allows the page’s origin and the PUT method. Without it, the browser fails before sending any data.
• Caching a presigned URL and reusing it after it expires gives “access denied” errors.

Large uploads

• A single PUT can upload up to 5 GB; AWS recommends multipart upload for objects over about 100 MB, and it is required above 5 GB, up to the maximum object size. Uploads can resume by part, and a failed part can be retried without restarting.
• Add a lifecycle rule that deletes incomplete multipart uploads. Unfinished parts are stored and billed until they are removed or aborted.

Event notifications

• An S3 event that invokes Lambda, and a function that writes output into the same bucket, can loop forever. Limit the notification to an input prefix, and write output to a different prefix or bucket that the notification doesn’t match.

Versions and overwrites

• With versioning on, a lifecycle rule can expire noncurrent versions after a number of days. A rule that expires current objects is a different action.
• To stop anyone replacing an existing key, use a conditional write (If-None-Match: *). The second write then fails instead of overwriting.

Archived objects in S3 Glacier

DVA-C03 adds Amazon S3 Glacier to its in-scope list, so expect a question where your code reads an object that has been archived.

• There are three Glacier storage classes. Glacier Instant Retrieval returns objects in milliseconds, like S3 Standard. Glacier Flexible Retrieval and Glacier Deep Archive are archives: you can’t read an object until you restore it.
• A GET on an archived object fails with an InvalidObjectState error. Call RestoreObject with the number of days the temporary copy should stay available, then read it once the restore finishes.
• Restore tiers trade speed for cost. Flexible Retrieval offers Expedited (usually 1-5 minutes), Standard (3-5 hours) and Bulk (5-12 hours). Deep Archive offers Standard (within 12 hours) and Bulk (within 48 hours).
• Don’t poll for a restore. The s3:ObjectRestore:Completed event notification can invoke a function or send a message when the copy is ready.
• Lifecycle rules move objects into Glacier classes after a number of days. Each class has a minimum storage charge (90 days for Instant and Flexible Retrieval, 180 days for Deep Archive), so archiving short-lived data costs more, not less.

How to choose

Common traps

• Reusing an expired presigned URL.
• Writing output to the bucket that triggers the function.
• Expiring current objects when the goal is cleaning old versions.
• Assuming a presigned URL needs a public bucket. It does not.
• Thinking a single PUT suits a multi-gigabyte file. Use multipart.
• Reading an archived object directly. Flexible Retrieval and Deep Archive objects must be restored first.

Try the 1 practice question ↓