make coffer_blobserver restful

[benoitc]

It came to myself that the blobserver should be restful. By restful I mean resource aware and each storage should be considered as a resouce receiving messages on it.

A storage should only receive and send messages, A general abstraction on top of it should be able to forward it to another protocol if needed.

Rignt now a blob strage return an anonymous function and s state associated, but it shows its limit when you want to distribute the messages between the Erlang nodes or simply when you want to handle a remote node via HTTP, which handles by nature the messages asynchronously.

A storage server would receive the following:

Storage requests

Message:

{storage_request, From, ReplyAs, Request}

• From: Pid of the process making the request
• ReplyAs: Reference
• Request: a request

Requests on a storage for now can be:

• capabilities : get all the capabilties
• {enumerate, Options}: get the lists of blobs in this storage
• {setopts, Options}: replace the storage' options for the properties given
• getopts: get the storage' options

If an index is associated to the storage we can also have

• {search, Args}

blob request:

Message:

{blob_request, From, ReplyAs, Request}

• From: Pid of the process making the request
• ReplyAs: Reference
• Request: a request

Request can be:

• {fetch, BlobRef} : return {ok, ReaderStream} | {error, not_found} . ReaderStream is a reference to an stream opened by the storage handler to fetch the blob content. remotely or not.
• {receive, BlobRef}: return {ok, WriterStream} | {error, already_exists} | {error, term()} . WriterStream os a stream opened by the storage handler to store the content remotely or not.
• {delete, BlobRef}: : return ok | {error, not_found} | {error, term()} . To delete a blob.

A ReaderStream send messages to the requesting process:

• {Ref, {ok, Data}} : binary data
• {Ref, {ok, Data, Ranges}} : return only some part of the blob to support range requests. A range is a list of tuple {Start, End} where {Start, End} are positions in the blob (in bytes).
• {Ref, eob} : end of the blob
• {error, term()}: an error happened

A WriterStream receive messages from the requesting process:

• {Ref, {ok, Data}} : binary data
• {Ref, eob} : end of the blob
• {error, term()}: an error happened

Note:

• A writer should always make sure that a blob with the same reference is not actually uploaded. If an error happen, it should remove any stored content.
• A storage should efficiently store blobs so he can respond to the ranges requests