if the write call completes, then its safe to mimick on the other data source; it'll be guaranteed for the relationship to exist/not exist depending on the semantics

Yes, the use case here is as follows: - I already have a relation

group:A member user:B

- I start a transaction in my client, make a call to SpiceDB to

TOUCH group:A member user:B

and write some data to my database as well - Something happens in my client, I need to rollback the transaction - In this case calling

DELETE group:A member user:B

would actually leave SpiceDB in a different state to what it was before, because the relationship already existed, therefore the cleanup I do in my rollback handler should be a no-op instead of calling SpiceDB to delete

if you already have the relationship, why are you calling TOUCH?

It's a case where I might be receiving an incoming request, I want to dual write to SpiceDB and my own DB, I write to SpiceDB first to have the ZedToken, only then I try to write to my own DB and might realise that data already exists

this is what preconditions are designed to handle

How would that work in a case where my input to

WriteRelationships

might include 5 updates for example, and one of the relationships might already exist but the other 4 still need to be written to SpiceDB?

right now it would cancel the entire transaction

I see, so I could then correlate which precondition failed with which relationship I was trying to write (if let's say I add one precondition per relationship) and retry the call without that one as I'd know it would be a no-op?

Pretty much

If multiple preconditions aren't satisfied, do I get the full list of the failed ones, or does it stop after the first failed one?

We could look into adding info on which touch or deletes impacted but it would slow down writes

Yeah it might be good to have an optional way of specifying that you'd like info back in the

WriteRelationshipsResponse

File an issue and we’ll look into it; there will be a performance penalty in certain data stores though

Thanks, created the issue: https://github.com/authzed/spicedb/issues/1903