Peter Geoghegan <pg@heroku.com> writes:
> A customer is on 9.6.1, and complains of a segfault observed at least
> 3 times. In all cases, the logs look like this:
> ...
> I can use GDB to get details of the instruction pointer that appeared
> in the kernel trap error, which shows a function from the expanded
> value representation infrastructure:
It would help to know the data types of the columns involved in this
query; but just eyeballing it, it doesn't look like it involves any
array operations, so it's pretty hard to believe that the expanded-object
code could have gotten invoked intentionally. (The mere presence of
an array column somewhere in the vicinity would not do that; you'd
need to invoke an array-ish operation, or at least pass the array into
a plpgsql function.)
If I had to bet on the basis of this much info, I would bet that the
parallel-query infrastructure is dropping the ball somewhere and
transmitting a corrupted datum that accidentally looks like it is
an expanded-object reference.
If $customer wants a quick fix, I'd suggest seeing whether disabling
parallel query makes the problem go away. That might be a good first
step anyway, just to narrow down where the problem lies.
regards, tom lane