Repo API: Bulk read/write

[enikao]

Ask the model repository for some subset of the stored model, and update it.

Assumptions for time being

• No federation/references to external repositories (Repository federation #23)
• No versioning (Repo API: Versioning / Collaboration #26)
• No access control (Repo API: Access control #27)
• No partitioning (Repo API: Do we need model partitions? #29)

Assumptions

• Reference nodes by id (Repo API: Node IDs #31)

Conceptual API

Independent of implementation (will most probably be REST, but could also use Protobuf etc.)

retrieve part of model

in: (list of) node IDs
in: mode {node, subtree, closure}
out: (list of) subgraph(s)

store part of model

in: (list of) subgraph(s)
out: success

Give me a range of IDs I can use

TODO:
in: Number of ids required
out: list of reserved ids

[enikao]

Do we care about Command Query Responsibility Segregation (https://www.martinfowler.com/bliki/CQRS.html), i.e. different APIs for read and write access? Seems to be popular especially with highly scalable systems.

[dslmeinte]

I think it's unavoidable if we want to enable collaborative editing. Even when using CRDTs it's useful to separate the intent of performing some edit action from the determination of that edit action being consistent with persisted state and (possibly) other edit actions.

[enikao]

I have no experience with web-based systems. Would "different API" mean something like "use POST instead of GET", "have another fragment in my URL" or more like "completely different URL, possibly another server"?

[enikao]

How do we report back unknown ids?
Example: I sent a list of 5 ids to the server in mode node (i.e. only asking for each node, neither subtree nor closure).
3 of the ids can be resolved, 2 not. How is this reported back to me?

(Relates to discussion about resolve info)

[joswarmer]

@enikao I think that we need to decide how to report errors in general. And of course we need the list of potential error that may be reported back.

[ftomassetti]

Inspired by @joswarmer comment I created this issue: #74

[enikao]

relates to #127

[joswarmer]

How do we report back unknown ids? Example: I sent a list of 5 ids to the server in mode node (i.e. only asking for each node, neither subtree nor closure). 3 of the ids can be resolved, 2 not. How is this reported back to me?

(Relates to discussion about resolve info)

Asking for nodes that do not exist in the repository is not an error, but a normal thing. Consider the following use cases:

• The model at the client contains an unresolved reference
• The client thinks a node exists, but in the mean time another client has changed the repository such that the node does not exist anymore.

Therefore, I think we should not explicitly report back unknown nodes. The client can simply check whether a requested node is in the returned chunk or not.

[enikao]

Spec in #167