If I get one of the other 2 options it is like the query never happened. So far this only seems to happen when I send a query in dgraph cloud but I have not tested a lot with my client yet.
This is quite annoying. Why did this break all of the sudden?
There was a single replica in the backend that had some disk corruption for only this specific entry, which turned out to be your GraphQL schema. It went undetected which then reflected in the error messages you were seeing. Replacing the replica fixed the issue. The other replica servers were OK which is why your error message was intermittent (the error message came only from the bad server).
We’ve added additional monitoring to detect this kind of disk issue.