Re: [provision] RE: Batch as a Logical Unit of Work (was"Re: [provision] Comment s on Draft 8 of the Spec...")

Bohren, Jeffrey wrote:

>Yes, a batch does not imply a logical unit of work. But that is not the same
>as saying a batch can not be a logical unit of work. The spec needs to be
>more neutral on the issue than it is now.
>
>Another way to say it is that an RA can not request that a batch be treated
>as a logical unit, but a PSP can still do so anyway. A PSP may choose to
>roll back the effects of a batch and report all the requests as failures.
>That is outside the scope of the spec but the PSP is free to do so.
>
>  
>
You are right. How is this?

*No transactional semantics*. Using a batch operation to combine 
individual requests does not imply atomicity (i.e., “all-or-nothing” 
semantics) for the group of batched requests. A requestor must not 
assume that the failure of a nested request will undo a nested request 
that has already completed. (See the section entitled “Transactional 
Semantics <#_Transactional_Semantics>”.)

Note that this does not /preclude/ a batch operation having 
transactional semantics—this is unspecified. A provider (or some 
higher-level service) with the ability to undo specific operations could 
support rolling back an entire batch if an operation nested within the 
batch fails.