On Mon, Jan 27, 2014 at 6:08 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> "Timed just right"? I could see it possibly happening due to queryid
> collisions, but I'm not seeing how it would happen absent such a hash
> collision.
Consider what happens when there is a pg_stat_statements_reset() call
query after another query's parse analysis, but before its execution
finishes. That's one obvious way. But you don't even need a reset - a
badly timed entry_dealloc() could do it too.
I don't see what hash collisions have to do with it, though.
--
Peter Geoghegan