Re: Why my query not using index to sort?

When I keep the sort off, I see it is executing much more faster.

set enable_sort = off;
explain analyze select
*
from
"standard_workitems"
where
"standard_workitems"."deleted_at" is null
and "standard_workitems"."company_id" = '6fed40b7-fdd7-4efb-a163-c2b42e6486ae'
order by
item_code asc;

Index Scan using standard_workitems_partial_index_idx_1_1 on standard_workitems (cost=0.42..5802.04 rows=1697 width=763) (actual time=0.018..1.076 rows=2071 loops=1)
Index Cond: (company_id = '6fed40b7-fdd7-4efb-a163-c2b42e6486ae'::uuid)
Planning time: 0.191 ms
Execution time: 1.210 ms

I have no idea why PG thinks in memory sort will be cheaper..

Thanks,

Arup Rakshit
ar(at)zeit(dot)io

> On 29-Sep-2018, at 9:40 PM, Arup Rakshit <ar(at)zeit(dot)io> wrote:
>
> Hello Tom,
>
> I have another query, where I am expecting the sort from index, but it is in memory and takes lot of time.
>
> Query:
>
> explain analyze select
> *
> from
> "standard_workitems"
> where
> "standard_workitems"."deleted_at" is null
> and "standard_workitems"."company_id" = '6fed40b7-fdd7-4efb-a163-c2b42e6486ae'
> order by
> standard_workitems.item_code asc;
>
> Explain plan:
>
> Sort (cost=3454.03..3458.18 rows=1660 width=810) (actual time=20.302..20.502 rows=2071 loops=1)
> Sort Key: item_code
> Sort Method: quicksort Memory: 800kB
> -> Bitmap Heap Scan on standard_workitems (cost=57.29..3365.25 rows=1660 width=810) (actual time=0.297..0.781 rows=2071 loops=1)
> Recheck Cond: ((company_id = '6fed40b7-fdd7-4efb-a163-c2b42e6486ae'::uuid) AND (deleted_at IS NULL))
> Heap Blocks: exact=139
> -> Bitmap Index Scan on standard_workitems_partial_index_idx_1_1 (cost=0.00..56.87 rows=1660 width=0) (actual time=0.272..0.272 rows=2071 loops=1)
> Index Cond: (company_id = '6fed40b7-fdd7-4efb-a163-c2b42e6486ae'::uuid)
> Planning time: 0.199 ms
> Execution time: 20.688 ms
>
> Indexes I have:
>
> Indexes:
> "standard_workitems_pkey" PRIMARY KEY, btree (id)
> "index_standard_workitems_on_company_id" btree (company_id)
> "index_standard_workitems_on_deleted_at" btree (deleted_at)
> "index_standard_workitems_on_item_code" btree (item_code)
> "index_standard_workitems_on_workitem_category_id" btree (workitem_category_id)
> "standard_workitems_partial_index_idx_1_1" btree (company_id, item_code) WHERE deleted_at IS NULL
>
>
>
> Thanks,
>
> Arup Rakshit
> ar(at)zeit(dot)io <mailto:ar(at)zeit(dot)io>
>
>
>
>> On 28-Sep-2018, at 7:07 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us <mailto:tgl(at)sss(dot)pgh(dot)pa(dot)us>> wrote:
>>
>> Arup Rakshit <ar(at)zeit(dot)io <mailto:ar(at)zeit(dot)io>> writes:
>>> My query is not using name index to sort the result.
>>
>> Given the rowcounts here, I think the planner is making the right choice.
>> Sorting 70-some rows with a Sort node is probably cheaper than doing
>> random disk I/O to get them in sorted order. With more rows involved,
>> it might make the other choice.
>>
>> As a testing measure (don't do it in production!), you could set
>> enable_sort = off, which will force the planner to pick a non-Sort
>> plan if possible. Then you could see whether that's actually faster
>> or slower, and by how much.
>>
>> regards, tom lane
>