Re: Should we add GUCs to allow partition pruning to be disabled?

On 2019/03/11 11:00, David Rowley wrote:
> On Mon, 11 Mar 2019 at 14:33, Amit Langote
> <Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>> PG 11 moved the needle a bit for SELECT queries:
>>
>> Excluding unnecessary partitions is slow for UPDATE and DELETE queries,
>
> With those words I expect the user might be surprised that it's still
> slow after doing SET enable_partition_pruning = off;
>
> I'm not really talking about constraint exclusion or partition
> pruning. The memory growth problem the user was experiencing was down
> to the fact that we plan once per partition and each of the
> PlannerInfos used for each planner run has a RangeTblEntry for all
> partitions. This means if you add one more partition and you get N
> partitions more RangeTblEntry items in memory. This is the quadratic
> memory growth that I mentioned in the -general post.

Yeah, I get it. As I said in my email, all we have ever mentioned in the
documentation as the reason for queries on partitioned tables being slow
is that partition exclusion is slow and nothing else. Can we put
quadratic memory growth during planning as the reason for performance
degradation into the documentation? Maybe we could, but every time I
tried it, it didn't read like user-facing documentation to me. Do you
have something in mind that we could add?

Thanks,
Amit