Storage locations
Where a bucket's objects are stored (a zone chosen at creation, or on request storage of your organization's own) and how each behaves when something fails.
Zones
A bucket's objects are stored in one zone, chosen when the bucket is created and fixed after that:
| Zone | Location |
|---|---|
de1a |
Falkenstein (the default) |
de1b |
Nuremberg |
fi1a |
Helsinki, Finland: in the EU, outside Germany, only when allowed (below) |
fi1a is offered only when an administrator has turned on Allow EU locations outside Germany
under Settings → Data location. Without it, creating a bucket there is refused with 403,
and a replication rule that would copy from Germany into a Helsinki bucket with 400. The console
marks such a bucket outside Germany · Finland.
Every bucket is reached at the same endpoint whatever its zone; GetBucketLocation answers the
zone. Put a bucket in the zone of the instances that read it. To keep data through the loss of a
whole zone, replicate the bucket to one in the other zone (see
Replicate a bucket).
Own storage
On request, Deutsche Cloud puts an organization's new buckets on storage of its own instead of a zone. The bucket list shows which in Location:
| Storage | Location label | Copies |
|---|---|---|
| A Deutsche Cloud storage cluster | e.g. "2 copies in 2 buildings · storage cluster" | every object in more than one building |
| Your organization's own hosts | "Own hosts · single copy" | one copy |
The API, endpoint, keys, encryption and console are the same. Such a bucket has no zone to choose
(a zone in the create is refused with 400) and no replication. It is billed per GB-month held at
the price set for that storage; without a price it is counted, not billed. It is limited by the
s3.stored-bytes quota (see Quotas): past it, new data answers 403 QuotaExceeded;
reading and deleting always work.
When it fills up
- A storage cluster stops taking new data when its fullest building reaches its fill limit, with a
warning before: 45 % (warning at 35 %) for a cluster in two buildings, 70 % (warning at 60 %) from
three. New data then answers
507 InsufficientStorage. Your organization is told when the warning level is reached. - Your own hosts' store refuses new data from 95 % full (
507 InsufficientStorage).
Reads and deletes keep working.
When a part fails
- One server of a storage cluster down: its buckets keep working from the other copies.
- A whole building down, or the buildings cut off from each other: reads go on, and every write
answers
503 ServiceUnavailablewithRetry-After: 10at once instead of waiting. Retry the write. - No server of the cluster can read: every request answers
503 ServiceUnavailable. - The whole cluster unavailable: see Keep a remote copy.
Versioning and lifecycle on a storage cluster
A bucket on a storage cluster created without object lock starts in shared storage. Turning on
versioning or lifecycle rules is accepted while the bucket is empty (it is then moved to storage
of its own, invisibly), and refused with 409 once it holds objects: turn them on right after
creating the bucket, as Terraform and most tools do. Listing versions of such a bucket before
versioning was ever on answers 501 NotImplemented; ListObjectsV2 lists its objects.