Hi Jesper,
On 2019/03/22 22:01, Jesper Pedersen wrote:
> Hi Alvaro,
>
> On 3/21/19 6:18 PM, Alvaro Herrera wrote:
>> On 2019-Mar-21, Jesper Pedersen wrote:
>>> pgbench -M prepared -f select.sql ....
>>>
>>> I'm seeing 82.64% spent in GetCachedPlan(). plan_cache_mode is auto.
>>
>> Hmm, I can't reproduce this at all ... I don't even see GetCachedPlan
>> in the profile. Do you maybe have some additional patch in your tree?
>>
>
> No, with 7df159a62 and v7 compiled with "-O0 -fno-omit-frame-pointer" I
> still see it.
>
> plan_cache_mode = auto
> 2394 TPS w/ GetCachePlan() @ 81.79%
>
> plan_cache_mode = force_generic_plan
> 10984 TPS w/ GetCachePlan() @ 23.52%
Wouldn't you get the same numbers on HEAD too? IOW, I'm not sure how the
patch here, which seems mostly about getting DDL in order to support
foreign keys on partitioned tables, would have affected the result of this
benchmark. Can you clarify your intention of running this benchmark
against these patches?
Thanks,
Amit