我不明白为什么这个查询在我的生产数据库上异常缓慢。我看到两次连续扫描 - 第一次来自 dockets_table.content_type_id = 2 上的过滤器，第二次我什至无法弄清楚它与什么有关。此查询在生产机器上需要几分钟，最多需要几秒钟。
查询：
SELECT COUNT(*)
FROM (
SELECT DISTINCT
ON (rank, "dockets_document"."id") "dockets_document"."id" AS Col1,
"dockets_document"."title" AS Col2,
"dockets_document"."url" AS Col3,
"dockets_document"."content" AS Col4,
"dockets_document"."num_tables" AS Col5,
"dockets_document"."num_pages" AS Col6,
"dockets_document"."tables_processed" AS Col7,
"dockets_document"."summary" AS Col8,
"dockets_document"."importance" AS Col9,
"dockets_document"."title_vector" AS Col10,
"dockets_document"."search_vector" AS Col11,
"dockets_document"."datetime_created" AS Col12,
"dockets_document"."datetime_modified" AS Col13,
"dockets_document"."docket_id" AS Col14,
"dockets_document"."docket_number" AS Col15,
"dockets_document"."date_filed" AS Col16,
"dockets_document"."pdf" AS Col17,
((1 * ts_rank("dockets_document"."search_vector", plainto_tsquery('public.bb'::regconfig, 'knighthead'))) + (0.8 * ts_rank("dockets_attachment"."search_vector", plainto_tsquery('public.bb'::regconfig, 'knighthead')))) AS "rank",
ts_headline("dockets_document"."title", plainto_tsquery('public.bb'::regconfig, 'knighthead'), 'StartSel=''<mark>'', StopSel=''</mark>'', HighlightAll=true') AS "title_snippet",
ts_headline(CONCAT(array_to_string("dockets_document"."content", ' ', ''), CONCAT(array_to_string("dockets_attachment"."content", ' ', ''), CONCAT(array_to_string("dockets_table"."content", ' ', ''), array_to_string(T5."content", ' ', '')))), plainto_tsquery('public.bb'::regconfig, 'knighthead'), 'StartSel=''<mark>'', StopSel=''</mark>'', MaxFragments=5') AS "content_snippet"
FROM "dockets_document"
LEFT OUTER JOIN "dockets_attachment"
ON ("dockets_document"."id" = "dockets_attachment"."main_document_id")
LEFT OUTER JOIN "dockets_table"
ON ("dockets_document"."id" = "dockets_table"."object_id" AND ("dockets_table"."content_type_id" = 8))
LEFT OUTER JOIN "dockets_table" T5
ON ("dockets_attachment"."id" = T5."object_id" AND (T5."content_type_id" = 24))
WHERE ("dockets_document"."docket_id" = 'p_hertz' AND NOT ("dockets_document"."title_vector" @@ ((((((plainto_tsquery('pro hac') || plainto_tsquery('certificate of mailing')) || plainto_tsquery('request for service')) || plainto_tsquery('certification of counsel')) || plainto_tsquery('receipt of filing fee')) || plainto_tsquery('affidavit of service')) || plainto_tsquery('Transfer/Assignment of Claim')) AND "dockets_document"."title_vector" IS NOT NULL) AND ("dockets_document"."search_vector" @@ plainto_tsquery('public.bb'::regconfig, 'knighthead') OR "dockets_attachment"."search_vector" @@ plainto_tsquery('public.bb'::regconfig, 'knighthead') OR "dockets_table"."search_vector" @@ plainto_tsquery('public.bb'::regconfig, 'knighthead') OR T5."search_vector" @@ plainto_tsquery('public.bb'::regconfig, 'knighthead')))
ORDER BY "rank" DESC, "dockets_document"."id" ASC
) subquery
解释分析输出：
Aggregate (cost=140427.33..140427.34 rows=1 width=8) (actual time=1758.440..1758.635 rows=1 loops=1)
-> Unique (cost=140425.97..140426.48 rows=68 width=1041) (actual time=1758.122..1758.625 rows=103 loops=1)
-> Sort (cost=140425.97..140426.14 rows=68 width=1041) (actual time=1758.120..1758.397 rows=1734 loops=1)
Sort Key: ((('1'::double precision * ts_rank(dockets_document.search_vector, '''knighthead'''::tsquery)) + ('0.8'::double precision * ts_rank(dockets_attachment.search_vector, '''knighthead'''::tsquery)))) DESC, dockets_document.id
Sort Method: quicksort Memory: 184kB
-> Hash Right Join (cost=137938.16..140423.90 rows=68 width=1041) (actual time=221.737..1756.835 rows=1734 loops=1)
Hash Cond: ((t5.object_id)::text = (dockets_attachment.id)::text)
Filter: ((dockets_document.search_vector @@ '''knighthead'''::tsquery) OR (dockets_attachment.search_vector @@ '''knighthead'''::tsquery) OR (dockets_table.search_vector @@ '''knighthead'''::tsquery) OR (t5.search_vector @@ '''knighthead'''::tsquery))
Rows Removed by Filter: 45943
-> Index Scan using dockets_table_content_type_id_c1e999b5 on dockets_table t5 (cost=0.43..2414.76 rows=16933 width=46) (actual time=0.040..8.865 rows=20302 loops=1)
Index Cond: (content_type_id = 24)
-> Hash (cost=137852.10..137852.10 rows=6851 width=306) (actual time=214.841..215.034 rows=16360 loops=1)
Buckets: 16384 (originally 8192) Batches: 2 (originally 1) Memory Usage: 3969kB
-> Gather (cost=1243.24..137852.10 rows=6851 width=306) (actual time=2.297..201.903 rows=16360 loops=1)
Workers Planned: 1
Workers Launched: 1
-> Nested Loop Left Join (cost=243.24..136167.00 rows=4030 width=306) (actual time=1.154..195.086 rows=8180 loops=2)
-> Nested Loop Left Join (cost=242.81..43560.75 rows=1667 width=274) (actual time=1.134..147.731 rows=2794 loops=2)
-> Parallel Bitmap Heap Scan on dockets_document (cost=242.39..13706.77 rows=1667 width=116) (actual time=1.113..113.645 rows=2401 loops=2)
Recheck Cond: ((docket_id)::text = 'p_hertz'::text)
Filter: ((NOT (title_vector @@ ((((((plainto_tsquery('pro hac'::text) || plainto_tsquery('certificate of mailing'::text)) || plainto_tsquery('request for service'::text)) || plainto_tsquery('certification of counsel'::text)) || plainto_tsquery('receipt of filing fee'::text)) || plainto_tsquery('affidavit of service'::text)) || plainto_tsquery('Transfer/Assignment of Claim'::text)))) OR (title_vector IS NULL))
Heap Blocks: exact=2244
-> Bitmap Index Scan on dockets_document_company_id_6b3b6f6d (cost=0.00..241.68 rows=2834 width=0) (actual time=1.105..1.105 rows=4943 loops=1)
Index Cond: ((docket_id)::text = 'p_hertz'::text)
-> Index Scan using dockets_attachment_main_document_id_1f6c050c on dockets_attachment (cost=0.42..17.87 rows=4 width=171) (actual time=0.013..0.013 rows=0 loops=4802)
Index Cond: ((dockets_document.id)::text = (main_document_id)::text)
-> Index Scan using dockets_table_object_id_da06f22d on dockets_table (cost=0.43..54.83 rows=72 width=46) (actual time=0.014..0.015 rows=2 loops=5587)
Index Cond: ((dockets_document.id)::text = (object_id)::text)
Filter: (content_type_id = 8)
Planning time: 3.283 ms
Execution time: 1759.062 ms
如果有帮助，这里是构建查询的 Django ORM：
def search(self, search_term):
def myStringAgg(field: str):
return Func(
F(field),
Value(" "),
Value(""),
function="array_to_string",
output_field=models.TextField(),
)
query = SearchQuery(search_term, config="public.bb")
rank = (
1*SearchRank(F("search_vector"), query)
+ 0.8*SearchRank(F("attachments__search_vector"), query)
)
qs = (
self.filter(
Q(search_vector=query)
| Q(attachments__search_vector=query)
| Q(tables__search_vector=query)
| Q(attachments__tables__search_vector=query)
)
.annotate(rank=rank)
.order_by("-rank", "pk")
.annotate(
title_snippet=SearchHeadline(
"title",
query,
highlight_all=True,
start_sel="<mark>",
stop_sel="</mark>",
)
)
.annotate(
content_snippet=SearchHeadline(
Concat(
myStringAgg("content"),
myStringAgg("attachments__content"),
myStringAgg("tables__content"),
myStringAgg("attachments__tables__content"),
),
query,
max_fragments=5,
start_sel="<mark>",
stop_sel="</mark>",
)
)
.distinct("rank", "pk")
)
return qs