The API client

The api groups are named to match the surface the pydo SDK presented, and every response is a plain dict:

from ccai.restclient.digitalocean.digitaloceanrestclient import (
    DigitalOceanRestClient
)

client = DigitalOceanRestClient(token=token)

created = client.droplets.create(body={"name": "srv-a", ...})
droplet = created.get("droplet", {})

page = client.droplets.list(per_page=200, page=1, tag_name="run:abc")
for entry in page.get("droplets", []):
    ...

client.droplets.destroy(droplet_id=droplet["id"])

Groups: vpcs, firewalls, droplets, volumes, volume_actions, ssh_keys, domains.

list returns ONE page, exactly as the SDK did, so paging stays visible to the caller rather than hidden in the transport. Follow links.pages.next to exhaust a collection.

The Spaces client

from ccai.restclient.digitalocean.spacesclient import SpacesClient

spaces = SpacesClient(access_key=key, secret_key=secret, region="nyc3")

spaces.create_bucket(Bucket="my-bucket")
spaces.head_bucket(Bucket="my-bucket")
spaces.delete_bucket(Bucket="my-bucket")

The method and keyword names mirror the S3 surface — including the capitalised Bucket — so call sites written against an S3 SDK keep their shape. Signing comes from ccai.restclient.base.awssigv4, shared with AWS and any other S3-compatible store.

Design notes

Requests are path-style. A bucket created moments ago is immediately addressable at https://{region}.digitaloceanspaces.com/{bucket}, while its virtual-host name has to wait on DNS. An apply that creates a bucket and then heads it would intermittently fail under virtual-host addressing.

Spaces errors carry the S3 code as an attribute. SpacesRequestError exposes status, code and body. Callers distinguish “bucket absent” (recover) from “auth or throttle failed” (do not recover) on that code, so it is first-class rather than something to dig out of a message. A failed HEAD has no body at all, so the status is used as the code in that case — it must never come back empty for a 404.

Error XML is matched on local names. S3 responses are namespaced, and whether an endpoint declares the namespace varies, so tags are matched with the namespace stripped.

Errors are thin. Translating status codes into a typed hierarchy is the consumer’s policy, not the transport’s.