On 04/07/2011 03:07 PM, Brendan Jurd wrote:
> On 7 April 2011 16:56, Tom Lane<tgl@sss.pgh.pa.us> wrote:
>> Brendan Jurd<direvus@gmail.com> writes:
>>> TRAP: FailedAssertion("!((data - start) == data_size)", File:
>>> "heaptuple.c", Line: 255)
>>
>> [ scratches head ... ] That implies that heap_fill_tuple came to a
>> different conclusion about a tuple's data size than the immediately
>> preceding heap_compute_data_size. Which I would sure want to believe
>> is impossible. Have you checked for flaky memory on this machine?
>>
>
> We are doing so now -- although the RAM is ECC and just a few months
> old, so flakiness seems a distant possibility. I will report back
> after we've given it a proper thrashing with memtest.
Apparently bad RAM can also mean faulty CPU (bad cache, heat problems,
etc). memtest86 seems ... rough ... at best when it comes to finding
issues; I've had some systems run it for a day yet continuously segfault
in real-world use until the RAM was re-seated or swapped out.
--
Craig Ringer