Improve error handling when sending a duplicate entry via GraphQL publishEntry

[adzialocha]

Currently it returns a "DUPLICATE" error coming directly from SQL which makes sense, but I expect the invalid entry to be catched before it hits the database (by checking the log integrity for example).

[cafca]

The error coming from the db is not necessarily a bad thing I think. We don't want to hit the db another time just to check something that a unique index can also catch. But returning a nice and friendly error message that doesn't depend would of course be nice.

[adzialocha]

The error coming from the db is not necessarily a bad thing I think. We don't want to hit the db another time just to check something that a unique index can also catch. But returning a nice and friendly error message that doesn't depend would of course be nice.

I agree, this is a totally fine pattern! 👍

Still I wonder, shouldn't the bamboo integrity validation throw before? Of course, two identical entries are valid if they are not appended yet, but if one already exists, the seq num and backlink is wrong?

[sandreae]

Of course, two identical entries are valid if they are not appended yet, but if one already exists, the seq num and backlink is wrong?

I haven't checked this myself, so this may be wrong, but is it because we are seeing this always with the first entry in a log? Then I could imagine the first error we hit is "DUPLICATE" for the log id.

[sandreae]

Looking at the verification code, we could probably catch this case here if we make some changes:

https://github.com/p2panda/p2panda/blob/fa7d916dfaee356a97d8e833174dfd8ba8aac9aa/p2panda-rs/src/storage_provider/traits/storage_provider.rs#L150-L162

[sandreae]

I think find_document_log_id() is a bit too smart, it returns either the log_id for the passed document_id, or the next available log id if there is no existing document. So we don't know if it's a new or existing log which is coming back.

We want to do: if existing_log_id & operation.is_create() { "Ouch!" }