endpoints.md

Endpoints

Spec	Method	Path
end-1	GET	/v2/
end-2	GET, HEAD	/v2/<name>/blobs/<digest> (supports Range)
end-3	GET, HEAD	/v2/<name>/manifests/<digest|tag>
end-4a	POST	/v2/<name>/blobs/uploads/
end-4b	POST	/v2/<name>/blobs/uploads/?digest=<digest>
end-5	PATCH	/v2/<name>/blobs/uploads/<ref>
end-6	PUT	/v2/<name>/blobs/uploads/<ref>?digest=<digest>
end-7a	PUT	/v2/<name>/manifests/<digest|tag>
end-7b	PUT	/v2/<name>/manifests/<digest>?tag=&tag=
end-8a	GET	/v2/<name>/tags/list
end-8b	GET	/v2/<name>/tags/list?n=&last=
end-9	DELETE	/v2/<name>/manifests/<digest|tag>
end-10	DELETE	/v2/<name>/blobs/<digest>
end-11	POST	/v2/<name>/blobs/uploads/?mount=<digest>&from=<repo>
end-12a	GET	/v2/<name>/referrers/<digest>
end-12b	GET	/v2/<name>/referrers/<digest>?artifactType=<type>
end-13	GET	/v2/<name>/blobs/uploads/<ref>

Failures that are not the request's fault

Every endpoint above may answer 503 Service Unavailable with the code TOOMANYREQUESTS and a Retry-After header, on any request that writes. It means the metastore refused the write because another writer held the lock — the request was valid, nothing is broken, and sending it again is the remedy. oras and containerd both honour it, so for those clients a contended push recovers on its own.

Any other storage failure is 500 with UNSUPPORTED, which is not retryable. The distinction is the whole point: the two used to be reported identically, and a client receiving the 500 would abandon a push that a retry would have completed.