Batch Request does not take dependsOn into account about restier

[MitchKuijpers]

I think sending a singlechangset should never be async?
However for splitting into operations it should be fine...

[jspuij]

This is an interesting issue, because I think it highlights some delicate formulations in the OData spec. (I used the formal OData 4.0 spec to cross reference).

Firstly, requests within a batch have to be processed in order. Requests within a change-set MAY be executed in any order (but it's not mandatory).

https://docs.oasis-open.org/odata/odata/v4.01/odata-v4.01-part1-protocol.html#sec_ProcessingaMultipartBatchRequest

However, within a change-set previously inserted entities CAN be referenced by a content-id:

https://docs.oasis-open.org/odata/odata/v4.01/odata-v4.01-part1-protocol.html#sec_ReferencingNewEntities

This of course implies, as @MartinTopfstedtSinc noted, that any request on a relational database that has foreign key constraints enabled cannot be executed in parallel.

@robertmclaws I think this is something we have to fix. Now there are multiple ways to go about this:

• Disable parallelisation entirely. This of course is the slowest option.
• Disable parallelisation of the change-set whenever a dependency is detected. This enables parallelisation of change-sets whenever they contain no dependencies at all.
• Resolve the dependency graph. This requires building a dependency graph before processing and subsequently solving it before executing the change-set. The neatest, but also the most time-consuming solution.

Any thoughts?

[robertmclaws]

My plan was to detect if dependencies were involved and if so, disable parallelization.