What is S3-compatible storage and what does it guarantee?
S3-compatible means a service largely speaks the S3 API; which features really work is something you need to test yourself.
S3-compatible storage is object storage that implements the S3 API, so existing tools and libraries work. The label does not tell you which features work fully. Test versioning, lifecycle rules, object lock, replication and multipart upload before you make a production choice.
Who is it for: Developers and DevOps teams who are choosing object storage or want to migrate without rewriting their code.
What does S3-compatible mean?
S3 is an API for object storage: a set of HTTP requests to create, read, delete and manage buckets (containers) and objects. Because a lot of software speaks this API, other storage systems have adopted it as well. This lets you work with different services using the same SDK or command-line tool.
The API is not formally defined as an independent standard with an official test suite. Providers choose for themselves which part they implement. Compatible therefore means: the core operations work. It does not mean that every behaviour, every error code or every extra feature is identical.
What does compatibility guarantee, and what not?
Common basic operations such as putting, retrieving and deleting an object and requesting a list can be used with most services. Authentication with access keys and signed requests usually works too, although support for the newest signing methods differs.
Less self-evident are advanced features, policy rules, notifications, access models and behaviour in failure situations. Consistency after a write, the maximum object size and limits on the number of requests also differ. Ask a provider for a list of supported API calls and compare it with what your application uses.
Which features do you test before you choose?
Build a short test set that mimics your own workload and run it against each candidate. Focus on the features that matter most during a migration or an incident.
- Versioning: are earlier versions kept after overwriting and deleting, and can you restore them?
- Lifecycle rules: are objects moved or deleted automatically according to rules on prefix or age?
- Object lock: can you make objects immutable for a period, and does that also hold against administrators?
- Replication: can data be copied automatically to a second bucket or location, and how do you see the backlog?
- Multipart upload: can large files be sent in parts and resumed, even after an interruption?
How do you run the test?
Use the same SDK versions and tools as in production, not just a generic client. Do not compare only the success path: deliberately make requests fail and check whether error messages and status codes match what your code expects.
- Create a test bucket and an access key with limited permissions.
- Run your existing integration tests against the candidate’s endpoint.
- Carry out the features from the list above and record what differs.
- Measure latency and throughput with realistic object sizes.
- Check how you can export your data again, so you are not locked in.
Testing compatibility
| Feature | What do you check | Risk if it differs |
|---|---|---|
| Versioning | Recovery of overwritten or deleted objects | A mistaken deletion cannot be undone |
| Lifecycle | Rules on prefix, age and versions | Storage grows unnoticed or data disappears too early |
| Object lock | Retention modes and behaviour against administrators | Backups remain vulnerable to misuse or ransomware |
| Replication | Direction, delay and error handling | Recovery plan rests on incorrect assumptions |
| Multipart upload | Resuming, maximum parts and cleaning up aborted uploads | Failed large uploads cost time and storage |
Who is this less suitable for?
- If you depend on additional services around the storage, such as linked compute functions or event processing, API compatibility alone is not enough.
- If you have strict regulations around immutable storage, have the basis for object lock assessed and do not rely on a compatibility label.
- For a one-off, small storage need, extensive testing is often more work than it is worth.
Frequently asked questions
Can I switch without changing any code?
Often changing the endpoint, region and keys is enough. You then only adjust what differs, such as the bucket addressing model, signature versions or missing features. Test this with a copy first.
What is path-style and virtual-hosted addressing?
These are two ways of naming a bucket in the URL: as a path or as part of the hostname. Not every service supports both, which has consequences for DNS and certificates.
Does a compatibility claim say anything about performance?
No. The same API says nothing about latency, throughput or limits. Measure these with your own object sizes and concurrency.
How do I avoid vendor lock-in?
Stick to the standard operations in your code, keep configuration outside the application and periodically test whether you can copy your data to another service.
