Protocol does not handle race against a resource

[mmatuska]

The protocol does not tell us what to do if a PATCH, DELETE or HEAD request is submitted against the same resource as an already running PATCH upload. How should the server tell the client that the resource is busy?

[Acconut]

That's true, the protocol does not specify this. Most implementations (e.g. tusd) do not allow concurrent access and will return a 423 Locked status if they detect parallel requests to the same upload resource.

However, since an implementation may also allow concurrent requests (e.g. a HEAD request while a PATCH request is running), the protocol does not currently force locking upon implementations.

[mitar]

One alternative to locks is something called idempotency keys. It is very useful when primarily source of conflicts comes from the fact that the client itself retried, but the server has not yet noticed that the previous attempt failed.