Ignore conflicting concurrent storage writes of the same data with version="*"

[ftkg]

Upsert with DO NOTHING will ignore insert on conflicting key, instead of failing. The intended upside is that transactions with conflicting inserts will not have to be retried anymore, saving unnecessary database processing.

Results do not change, as the new query will still return no rows on conflict, hence the batch insert will be aborted with ErrStorageRejectedVersion.

[ftkg]

New query should return success for concurrent writes of identical objects. Had to use ON CONFLICT UPDATE to return a row for this scenario, that's an extra step but it should be better than retrying (additionally, the previous WHERE NOT EXISTS clause will avoid unnecessary updates for non-concurrent duplicate writes and keep the semantics of rejecting them).

[redbaron]

UNION ALL LIMIT 1 variant doesn't unnecessary overwrite row just to return it and can be made to work by checking returned row on the Go side: If returned row is the same which is written, then write succeeds.

I see what you mean by "concurrent safe"

[redbaron]

This is a behaviour change which needs to be prominently mentioned in release notes and docs. Previously it was possible to use this operation to create distributed lock/ leader election , now it is still possible but written value must be unique for each node

[zyro]

After internal discussion, we determined the current behaviour is actually desirable: "*" must fail on any existing value even if fully identical. This allows Nakama to support use cases such as using a storage object as a lock or signal if needed.

We may however introduce a new OCC operator at some point that allows the write to succeed if the object already existed but the value was identical.