$batch used to return a 202 accepted to indicate it accepted the $batch request even though it might not have read, let alone processed, all the requests in the batch request. This 202 accepted response conflicted with the use of 202 for asynchronous requests; this has been dealt with in ODATA-233.

So now we can distinguish between batch requests that are handled synchronously and asynchronously but that brings up the question of, in asynchronous mode, when to return results from a batch request?

Some of the requests/change sets in the batch might take almost no time to process while others might take minutes if not longer. Our proposal for asynchronous requests wraps the result of such an asynchronous request in a message/http response. Would we collect all the results and only return once the complete batch is processed? Or would we start returning results as soon as the result of the first request is available turning the asynchronous request into a synchronous request from that point onwards?

At a minimum we would have to describe the expected behavior of an asynchronous batch request but we might need to consider adding to what's there already and make it possible to return the results of a batch in chunks (not to confuse with chuck encoding).

A proposal could be to allow the return of a application/http at any time with a 202 accepted and a Location header indicating that the remainder of the response to the batch request is being accepted and dealt with and will be returned later by querying the URL provided in the location just as any other async request. Using the example in section 10.5.3 of the OData Core Part-1 document the response, after handling the first request (and including the new 200 OK response on the $batch), would look like:

HTTP/1.1 200 OK
DataServiceVersion: 4.0
Content-Length: ####
Content-Type: multipart/mixed; boundary=batch_36522ad7-fc75-4b56-8c71-56071383e77b

--batch_36522ad7-fc75-4b56-8c71-56071383e77b
Content-Type: application/http
Content-Transfer-Encoding: binary

HTTP/1.1 200 Ok
Content-Type: application/atom+xml;type=entry
Content-Length: ###

<AtomPub representation of the Customer entity with EntityKey ALFKI>

--batch_36522ad7-fc75-4b56-8c71-56071383e77b
Content-Type: application/http

HTTP/1.1 202 Accepted
Retry-After: ###
Location: https://services.odata.org/monitor/12345

After which the async pattern continues until the processing has progressed and the remainder of the response is available which in this example presumably would look like:

HTTP/1.1 200 OK
DataServiceVersion: 4.0
Content-Length: ####
Content-Type: multipart/mixed; boundary=batch_36522ad7-fc75-4b56-8c71-56071383e77b

--batch_36522ad7-fc75-4b56-8c71-56071383e77b
Content-Type: multipart/mixed; boundary=changeset_77162fcd-b8da-41ac-a9f8-9357efbbd621
Content-Length: ###

--changeset_77162fcd-b8da-41ac-a9f8-9357efbbd621
Content-Type: application/http
Content-Transfer-Encoding: binary

HTTP/1.1 201 Created
Content-Type: application/atom+xml;type=entry
Location: http://host/service.svc/Customer('POIUY')
Content-Length: ###

<AtomPub representation of a new Customer entity>

--changeset_77162fcd-b8da-41ac-a9f8-9357efbbd621
Content-Type: application/http
Content-Transfer-Encoding: binary

HTTP/1.1 204 No Content
Host: host

-changeset_77162fcd-b8da-41ac-a9f8-9357efbbd621-

--batch_36522ad7-fc75-4b56-8c71-56071383e77b
Content-Type: application/http
Content-Transfer-Encoding: binary

HTTP/1.1 404 Not Found
Content-Type: application/xml
Content-Length: ###

<Error message>

Note that I'd propose that the results of a change-set would never be split up as a result of asynchronous processing of a $batch request (probably don't need to as all request in such a set have to succeed or fail 'together' anyway).