On Thu, Jul 26, 2018 at 3:53 PM, Victor Yegorov <vyegorov@gmail.com> wrote:
> ср, 25 июл. 2018 г. в 8:46, Amit Kapila <amit.kapila16@gmail.com>:
>>
>> On Wed, Jul 25, 2018 at 8:43 AM, Amit Kapila <amit.kapila16@gmail.com>
>> wrote:
>> > Well spotted. I think here we can release the current page lock
>> > before calling _bt_parallel_seize as we don't need it to get the next
>> > page.
>> >
>>
>> I have written a patch on the above lines and manually verified (by
>> reproducing the issue via debugger) that it fixes the issue. Thomas,
>> Victor, is it possible for you guys to see if the attached fixes the
>> issue for you?
>
>
> First, I failed to reproduce Thomas' case on the VM with just 1 core.
> But it is reproduced just fine on an 8-core machine, make sure to warm up
> the index first though.
>
> Patch fixes the issue for me here.
>
Thanks, pushed the fix. I have not included Thomas's test as it is
timing dependent and it seems risky to include such a test.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com