Re: regression, deadlock in high frequency single-row UPDATE

On 13/12/14 10:57, Alvaro Herrera wrote:
> Alvaro Herrera wrote:
>> Alvaro Herrera wrote:
>>
>>> I'm going to experiment with that idea and see if it leads to a
>>> solution.  I tried the other idea yesterday (to keep the HW tuple lock
>>> we acquire in heap_lock_tuple until heap_update is done) but aside from
>>> being very complicated and bug-prone, it doesn't solve the problem
>>> anyway.
>>
>> Here's a preliminary patch.
>
> Here's a finished version of this patch, which I messed a bit with and
> so needs some extra testing.  I want to push this shortly and backpatch
> to 9.4 and 9.3.  I don't want to disrupt the 9.4.0 release next week,
> but it'd be nice not to ship it with this bug.
>

This patch seems good too. I'm possibly seeing it performing slightly
slower than the previous patch (approx 68 tps vs 63 on average -
completely untuned 9.5 running on a single SATA drive, but that could
just be natural variation/noise).

Regards

Mark