Re: DELETE query and indexes again

I should have included this: when I use the first two cols from the PK
(gid, guid) it uses the index. When I use the first col only (gid), it does
not. Here's the index delete.

Delete on play_statistics ps (cost=0.69..472003.08 rows=28920 width=12)
(actual time=3534.409..3534.409 rows=0 loops=1)
-> Nested Loop (cost=0.69..472003.08 rows=28920 width=12) (actual
time=16.703..1857.140 rows=2269013 loops=1)
-> Seq Scan on stats s (cost=0.00..2253.24 rows=54624 width=26)
(actual time=16.565..60.615 rows=21063 loops=1)
-> Index Scan using play_statistics_pkey on play_statistics ps
(cost=0.69..8.59 rows=1 width=26) (actual time=0.008..0.065 rows=108
loops=21063)
Index Cond: ((gid = s.gid) AND (guid = s.guid))
Planning Time: 1.494 ms
JIT:
Functions: 9
Options: Inlining false, Optimization false, Expressions true, Deforming
true
Timing: Generation 2.500 ms, Inlining 0.000 ms, Optimization 0.873 ms,
Emission 15.309 ms, Total 18.682 ms
Execution Time: 3569.225 ms
(11 rows)

On Thu, Jul 23, 2020 at 2:11 PM Wells Oliver <wells(dot)oliver(at)gmail(dot)com> wrote:

> This play_statistics table has 96,577,179 rows. Its PK is gid integer,
> guid uuid, cat text, name text, target integer.
>
> Doing a EXPLAIN ANALYZE DELETE WHERE gid IN (SELECT DISTINCT gid FROM
> temptable) where there is one row in the temptable, I get this:
>
> Delete on play_statistics (cost=1984.14..8037586.36 rows=3734075 width=34)
> (actual time=407093.859..407093.859 rows=0 loops=1)
> -> Hash Join (cost=1984.14..8037586.36 rows=3734075 width=34) (actual
> time=407093.857..407093.857 rows=0 loops=1)
> Hash Cond: (play_statistics.gid = "ANY_subquery".gid)
> -> Seq Scan on play_statistics (cost=0.00..7781818.16 rows=96600516
> width=10) (actual time=201.521..396052.101 rows=96556116 loops=1)
> -> Hash (cost=1981.64..1981.64 rows=200 width=32) (actual
> time=13.020..13.020 rows=1 loops=1)
> Buckets: 1024 Batches: 1 Memory Usage: 9kB
> -> Subquery Scan on "ANY_subquery" (cost=1977.64..1981.64 rows=200
> width=32) (actual time=13.012..13.013 rows=1 loops=1)
> -> HashAggregate (cost=1977.64..1979.64 rows=200 width=4) (actual
> time=13.001..13.002 rows=1 loops=1)
> Group Key: stats.gid
> -> Seq Scan on stats (cost=0.00..1924.91 rows=21091 width=4) (actual
> time=0.028..9.244 rows=21091 loops=1)
> Planning Time: 0.657 ms
> JIT:
> Functions: 15
> Options: Inlining true, Optimization true, Expressions true, Deforming true
> Timing: Generation 1.816 ms, Inlining 53.472 ms, Optimization 89.014 ms,
> Emission 58.759 ms, Total 203.060 ms
> Execution Time: 407112.908 ms
> (16 rows)
>
> It takes 7 minutes to delete 21063 rows for one given gid value, which
> seems excessive given I think it should be using the PK index.
>
> Anything I'm missing here?
>
> --
> Wells Oliver
> wells(dot)oliver(at)gmail(dot)com
>

--
Wells Oliver
wells(dot)oliver(at)gmail(dot)com <wellsoliver(at)gmail(dot)com>