postgres chooses an aweful query plan , how can that be fixed - postgresql

Question

I'm trying to optimize this query :
EXPLAIN ANALYZE
select
dtt.matching_protein_seq_ids
from detected_transcript_translation dtt
join peptide_spectrum_match psm
on psm.detected_transcript_translation_id =
dtt.detected_transcript_translation_id
join peptide_spectrum_match_sequence psms
on psm.peptide_spectrum_match_sequence_id =
psms.peptide_spectrum_match_sequence_id
WHERE
dtt.matching_protein_seq_ids && ARRAY[654819, 294711]
;
When seq_scan are allowed (set enable_seqscan = on), the optimizer chooses a pretty awful plan that runs in 49.85 seconds :
https://explain.depesz.com/s/WKbew
With set enable_seqscan = off, the plan chosen uses proper indexes and the query runs instantely.
https://explain.depesz.com/s/ISHV
note that I did run a ANALYZE on all three tables...

Answer 1

Your problem is that PostgreSQL cannot estimate the WHERE condition well, so it estimates it as a certain percentage of the estimated total rows, which is way too much.
If you know that there will always few result rows for a query like this, you could cheat by defining a function
CREATE OR REPLACE FUNCTION matching_transcript_translations(integer[])
RETURNS SETOF detected_transcript_translation
LANGUAGE SQL
STABLE STRICT
ROWS 2 /* pretend there are always exactly two matching rows */
AS
'SELECT * FROM detected_transcript_translation
WHERE matching_protein_seq_ids && $1';
You could use that like
select
dtt.matching_protein_seq_ids
from matching_transcript_translations(ARRAY[654819, 294711]) dtt
join peptide_spectrum_match psm
on psm.detected_transcript_translation_id =
dtt.detected_transcript_translation_id
join peptide_spectrum_match_sequence psms
on psm.peptide_spectrum_match_sequence_id =
psms.peptide_spectrum_match_sequence_id;
Then PostgreSQL should be cheated into thinking that there will be exactly one matching row.
However, if there are a lot of matching rows, the resulting plan will be even worse than your current plan is…