Thread: Different plan chosen when in lateral subquery

Weird, when I deleted an erroneous index it started picking a reasonable plan. This now works as expected, for posterity here is the bad plan:

 Nested Loop  (cost=21281.50..21323812.82 rows=5621000 width=47) (actual time=171.648..7233.298 rows=85615 loops=1)

   ->  Function Scan on generate_series dates  (cost=0.00..3.00 rows=1000 width=8) (actual time=0.031..0.252 rows=267 loops=1)

   ->  Unique  (cost=21281.50..21290.08 rows=5621 width=39) (actual time=25.730..27.050 rows=321 loops=267)

         ->  Sort  (cost=21281.50..21284.36 rows=5724 width=39) (actual time=25.728..26.242 rows=6713 loops=267)

               Sort Key: alloc.note_id, alloc.series_id

               Sort Method: quicksort  Memory: 2220kB

               ->  Nested Loop  (cost=10775.92..21210.05 rows=5724 width=39) (actual time=1.663..21.938 rows=6713 loops=267)

                     ->  Hash Join  (cost=10775.83..20355.61 rows=5724 width=52) (actual time=1.657..5.980 rows=6713 loops=267)

                           Hash Cond: (alloc.note_id = contrib.id)

                           ->  Bitmap Heap Scan on portfolio_allocations alloc  (cost=69.82..9628.13 rows=5724 width=39) (actual time=1.010..2.278 rows=6713 loops=267)

                                 Recheck Cond: ((entity_id = '\x5787f132f50f7b03002cf835'::bytea) AND (allocated_on <= date(dates.dates)))

                                 Heap Blocks: exact=118074

                                 ->  Bitmap Index Scan on portfolio_allocations_entity_id_allocated_on_idx  (cost=0.00..69.53 rows=5724 width=0) (actual time=0.956..0.956 rows=6713 lo

                                       Index Cond: ((entity_id = '\x5787f132f50f7b03002cf835'::bytea) AND (allocated_on <= date(dates.dates)))

                           ->  Hash  (cost=9464.85..9464.85 rows=354617 width=26) (actual time=169.792..169.792 rows=354617 loops=1)

                                 Buckets: 524288  Batches: 1  Memory Usage: 24296kB

                                 ->  Seq Scan on contributions contrib  (cost=0.00..9464.85 rows=354617 width=26) (actual time=0.007..83.246 rows=354617 loops=1)

                     ->  Index Only Scan using investments_pkey on investments inv  (cost=0.08..0.15 rows=1 width=13) (actual time=0.002..0.002 rows=1 loops=1792457)

                           Index Cond: (id = contrib.investment_id)

                           Heap Fetches: 1792457

 Planning time: 0.721 ms

 Execution time: 7236.507 ms

On Tue, Dec 5, 2017 at 10:04 AM Alex Reece <awreece@gmail.com> wrote:

I get very different plan chosen when my query is in a lateral subquery vs standalone -- it doesn't use a key when joining on a table, instead opting to do a hash join. Here is the query:

select distinct on (sub.entity_id, sub.note_id, sub.series_id)

       entity_id, note_id, series_id

from

(

select alloc.entity_id, alloc.note_id, alloc.series_id, alloc.amount, inv.name

from public.portfolio_allocations alloc

JOIN contributions contrib on contrib.id = alloc.note_id

JOIN investments inv on inv.id = contrib.investment_id

where entity_id = '\x5787f132f50f7b03002cf835' and 

alloc.allocated_on <= dates.date

) sub

And wrapped inside the lateral:

        explain analyze

        select *

        from generate_series('2017-03-14 20:59:59.999'::TIMESTAMPTZ,            current_timestamp::TIMESTAMP + INTERVAL '1 day', '24 hours') dates,

        LATERAL (

         ... <SUB QUERY HERE> ...

        ) lat

Run by itself injecting a hard coded value for dates.date, I get the expected plan which uses a key index on contributions:

      Unique  (cost=14.54..14.54 rows=2 width=39) (actual time=0.052..0.053 rows=2 loops=1)

         ->  Sort  (cost=14.54..14.54 rows=2 width=39) (actual time=0.052..0.052 rows=2 loops=1)

               Sort Key: alloc.note_id, alloc.series_id

               Sort Method: quicksort  Memory: 25kB

               ->  Nested Loop  (cost=0.25..14.53 rows=2 width=39) (actual time=0.030..0.042 rows=2      loops=1)

                     ->  Nested Loop  (cost=0.17..14.23 rows=2 width=52) (actual time=0.022..0.028       rows=2 loops=1)

                           ->  Index Scan using portfolio_allocations_entity_id_allocated_on_idx on      portfolio_allocations alloc  (cost=0.09..6.05 rows=2 width=39) (actual     time=0.012..0.014 

                                 Index Cond: ((entity_id = '\x5787f132f50f7b03002cf835'::bytea) AND      (allocated_on <= '2017-03-14 20:59:59.999+00'::timestamp with time   zone))

                           ->  Index Scan using contributions_id_accrue_from_idx on contributions     contrib  (cost=0.08..4.09 rows=1 width=26) (actual time=0.005..0.005    rows=1 loops=2)

                                 Index Cond: (id = alloc.note_id)

                     ->  Index Only Scan using investments_pkey on investments inv  (     cost=0.08..0.15 rows=1 width=13) (actual time=0.005..0.006 rows=1 loops=2)

                           Index Cond: (id = contrib.investment_id)

                           Heap Fetches: 2

       Planning time: 0.617 ms

       Execution time: 0.100 ms

      (15 rows)

But run in the lateral, it doesn't use the index:

       Nested Loop  (cost=14.54..24.55 rows=2000 width=47) (actual time=0.085..0.219 rows=534      loops=1)

         ->  Function Scan on generate_series dates  (cost=0.00..3.00 rows=1000 width=8) (actual      time=0.031..0.043 rows=267 loops=1)

         ->  Materialize  (cost=14.54..14.55 rows=2 width=39) (actual time=0.000..0.000 rows=2     loops=267)

               ->  Unique  (cost=14.54..14.54 rows=2 width=39) (actual time=0.052..0.053 rows=2       loops=1)

                     ->  Sort  (cost=14.54..14.54 rows=2 width=39) (actual time=0.051..0.052 rows=2      loops=1)

                           Sort Key: alloc.note_id, alloc.series_id

                           Sort Method: quicksort  Memory: 25kB

                           ->  Nested Loop  (cost=0.25..14.53 rows=2 width=39) (actual       time=0.029..0.041 rows=2 loops=1)

                                 ->  Nested Loop  (cost=0.17..14.23 rows=2 width=52) (actual       time=0.021..0.027 rows=2 loops=1)

                                       ->  Index Scan using       portfolio_allocations_entity_id_allocated_on_idx on      portfolio_allocations alloc  (cost=0.09..6.05 rows=2     width=39) (actual time=0

                                             Index Cond: ((entity_id =     '\x5787f132f50f7b03002cf835'::bytea) AND (allocated_on      <= '2017-03-14 20:59:59.999+00'::timestamp with time    zone))

                                       ->  Index Scan using contributions_id_accrue_from_idx on       contributions contrib  (cost=0.08..4.09 rows=1 width=26) (     actual time=0.005..0.005 rows=1 loo

                                             Index Cond: (id = alloc.note_id)

                                 ->  Index Only Scan using investments_pkey on investments inv  (     cost=0.08..0.15 rows=1 width=13) (actual time=0.005..0.006 rows=1    loops=2)

                                       Index Cond: (id = contrib.investment_id)

                                       Heap Fetches: 2

       Planning time: 0.718 ms

       Execution time: 0.296 ms

      (18 rows)

For reference, here are the indexes on the relevant tables:

Indexes:

    "portfolio_allocations_entity_id_allocated_on_idx" btree (entity_id, allocated_on DESC)

    "portfolio_allocations_note_id_allocated_on_idx" btree (note_id, allocated_on DESC)

    "portfolio_allocations_pnsa" btree (entity_id, note_id, series_id, allocated_on DESC)

Indexes:

    "contributions_pkey" PRIMARY KEY, btree (id)

    "contributions_id_accrue_from_idx" btree (id, events_earnings_accrue_from)

I have a few questions here:

  - Why doesn't it use the primary key index in either case?

  - Why isn't it choosing portfolio_allocations_pnsa, which seems like it would prevent it from having to sort?

Best,

~Alex