Cannot modify PostgreSql query, how to speed it up - postgresql

Question

I'm looking for speeding up a query (PostgreSql 9.5), but I cannot change it, because is executed by an application I cannot modify.
So, I captured the query from the PostgreSql logs, here it is:
SELECT Count(*)
FROM (SELECT ti.idturnosistemaexterno,
ti.historiaclinica_hp,
p.patientname,
CASE
WHEN ( ti.creationdate :: VARCHAR IS NOT NULL ) THEN
Date_trunc('SECOND', ti.creationdate) :: VARCHAR
ELSE 'NO EXISTE' :: VARCHAR
END AS creationdate,
CASE
WHEN ( st.idstudy :: VARCHAR IS NOT NULL ) THEN 'SI' :: VARCHAR
ELSE 'NO' :: VARCHAR
END AS idstudy,
st.institutionname,
CASE
WHEN ( st.created_time :: VARCHAR IS NOT NULL ) THEN
Date_trunc('SECOND', st.created_time) :: VARCHAR
ELSE 'NO EXISTE' :: VARCHAR
END AS created_time,
ti.enviado,
st.accessionnumber,
st.modality
FROM study st
right join turnointegracion ti
ON st.accessionnumber = ti.idturnosistemaexterno
left join patient p
ON st.idpatient = p.idpatient
ORDER BY ti.creationdate DESC) AS foo;
The explain analyze output is this:
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------------------
Aggregate (cost=231136.16..231136.17 rows=1 width=0) (actual time=32765.883..32765.883 rows=1 loops=1)
-> Sort (cost=230150.04..230314.39 rows=65741 width=8) (actual time=32754.992..32761.780 rows=64751 loops=1)
Sort Key: ti.creationdate DESC
Sort Method: external merge Disk: 1648kB
-> Hash Right Join (cost=219856.39..224889.28 rows=65741 width=8) (actual time=26653.007..32714.961 rows=64751 loops=1)
Hash Cond: ((st.accessionnumber)::text = (ti.idturnosistemaexterno)::text)
-> Seq Scan on study st (cost=0.00..4086.26 rows=77126 width=12) (actual time=12.983..6032.251 rows=77106 loops=1)
-> Hash (cost=219048.95..219048.95 rows=64595 width=16) (actual time=26639.722..26639.722 rows=64601 loops=1)
Buckets: 65536 Batches: 1 Memory Usage: 3602kB
-> Seq Scan on turnointegracion ti (cost=0.00..219048.95 rows=64595 width=16) (actual time=17.259..26611.806 rows=64601 loops=1)
Planning time: 25.519 ms
Execution time: 32766.710 ms
(12 rows)
Here are the related table definitions:
Table "public.turnointegracion"
Column | Type | Modifiers
---------------------------+-----------------------------+--------------------------------------------------------------------
idturnosistemaexterno | character varying(50) |
historiaclinica_hp | integer |
matriculaprofrealiza | character varying(10) |
matriculaprofinforma | character varying(10) |
obrasocial | character varying(20) |
planobrasocial | character varying(20) |
nroafiliado | character varying(20) |
nroautorizacion | character varying(20) |
matriculaprofprescribe | character varying(10) |
codigonomenclador | character varying(10) |
cantidadcodigonomenclador | integer |
importeunitariohonorarios | money |
importeunitarioderechos | money |
nrodefacturacion | character varying(15) |
informe | bytea |
titulodelestudio | character varying(250) |
fechaprescripcion | timestamp without time zone |
fechahora | timestamp without time zone |
enviado | boolean | not null default false
enviadofechahora | timestamp without time zone |
procesado_hp | timestamp without time zone |
modalidad | character varying(6) |
orden | integer | not null default nextval('turnointegracion_orden_seq'::regclass)
idturno | integer | not null default nextval('seq_turnointegracion_idturno'::regclass)
creationdate | timestamp without time zone | default now()
informetxt | text |
informedisponible | timestamp without time zone |
informeprocesado | timestamp without time zone |
Indexes:
"turnointegracion_pkey" PRIMARY KEY, btree (idturno)
"idx_fechahora" btree (fechahora)
"idx_historiaclinicahp" btree (historiaclinica_hp)
"idx_idturnosistemaexterno" btree (idturnosistemaexterno)
"idx_informedisponible" btree (informedisponible)
"idx_turnointegracion_creationdate" btree (creationdate DESC)
"idx_turnointegracion_idturnosistext_text" btree ((idturnosistemaexterno::text))
Table "public.study"
Column | Type | Modifiers
------------------------------+-----------------------------+---------------------------------------------------------
idstudy | integer | not null default nextval('study_idstudy_seq'::regclass)
studydate | date |
studytime | time without time zone |
studyid | character varying(20) |
studydescription | character varying(255) |
modality | character varying(2) |
modalityaetitle | character varying(50) |
nameofphysiciansreadingstudy | character varying(255) |
accessionnumber | character varying(20) |
performingphysiciansname | character varying(255) |
referringphysiciansname | character varying(255) |
studyinstanceuid | character varying(255) |
status | status_ |
institutionname | character varying(100) |
idpatient | integer |
created_time | timestamp without time zone |
Indexes:
"study_pkey" PRIMARY KEY, btree (idstudy)
"study_studyinstanceuid_key" UNIQUE CONSTRAINT, btree (studyinstanceuid)
"idx_study_accession_text" btree ((accessionnumber::text))
"idx_study_accessionnumber" btree (accessionnumber)
"idx_study_idstudy" btree (idstudy)
Foreign-key constraints:
"study_idpatient_fkey" FOREIGN KEY (idpatient) REFERENCES patient(idpatient)
Referenced by:
TABLE "series" CONSTRAINT "series_idstudy_fkey" FOREIGN KEY (idstudy) REFERENCES study(idstudy)
As you can see, I've added indexes on the affected columns but the planner is still doing sequential scans. Is there a way to improve this?.

Answer 1

There is no WHERE condition, due to this join:
right join turnointegracion ti
ON st.accessionnumber = ti.idturnosistemaexterno
you're reading all records from turnointegracion, adding an index for `creationdate' you can accelerate the sort function, but again all records are returned.
Filtering by creationdate can reduce the final time.