Maybe it was not the best use of the word "crashes". It gives me the error I showed you.
I don't really understand how PostgreSQL handles indexes, but would it be possible that the INDEX is being used/updated at that moment and so the INSERT doesn't know it exists? Can concurrency also be a problem?
Peter Geoghegan <pg@bowt.ie> escreveu no dia terça, 7/02/2017 às 23:36:
On Tue, Feb 7, 2017 at 3:21 PM, Tiago Babo <tiago.babo@gmail.com> wrote: > The only difference (so it seems), is that when it crashes, the person was created in the seconds before creating the account. When it previously exists in the persons table, it never gives the error.
It crashes?
Anyway, it's very hard to see how that could be, since the error in question is thrown from within the planner. I don't recall the exact details of how inference will do offhand, but I am suspicious of the cast that appears in the partial index predicate.