Question

在桌子上（由django模型使用）我使用jsonb列data来存储从webservice获取的任意数据：

abs=# \d data_importer_rawdata;
                                        Table "public.data_importer_rawdata"
    Column      |           Type           | Collation | Nullable |                      Default
-----------------+--------------------------+-----------+----------+---------------------------------------------------
id              | integer                  |           | not null | nextval('data_importer_rawdata_id_seq'::regclass)
created         | timestamp with time zone |           | not null |
modified        | timestamp with time zone |           | not null |
entity_id       | character varying(50)[]  |           | not null |
entity_id_key   | character varying(50)[]  |           | not null |
service         | character varying(100)   |           | not null |
data            | jsonb                    |           | not null |
data_hash       | bigint                   |           | not null |
content_type_id | integer                  |           | not null |
last_update     | timestamp with time zone |           |          |
Indexes:
    "data_importer_rawdata_pkey" PRIMARY KEY, btree (id)
    "data_importer_rawdata_entity_id_service_conten_5fcc60bd_uniq" UNIQUE CONSTRAINT, btree (entity_id, service, content_type_id)
    "data_importer_rawdata_content_type_id_63138c35" btree (content_type_id)
    "rawdata_data_idx" gin (data jsonb_path_ops)
    "rawdata_entity_id_idx" btree (entity_id)
    "rawdata_entity_id_key_idx" btree (entity_id_key)
    "rawdata_service_idx" btree (service)
Foreign-key constraints:
    "data_importer_rawdat_content_type_id_63138c35_fk_django_co" FOREIGN KEY (content_type_id) REFERENCES django_content_type(id) DEFERRABLE INITIALLY DEFERRED

记录是＆gt; 1M。

然而，尽管有各种索引策略（遵循this blog post），但性能仍然很差：

abs=# EXPLAIN ANALYZE SELECT
         "data_importer_rawdata"."id",
         "data_importer_rawdata"."created",
         "data_importer_rawdata"."modified",
         "data_importer_rawdata"."entity_id",
         "data_importer_rawdata"."entity_id_key",
         "data_importer_rawdata"."service",
         "data_importer_rawdata"."content_type_id",
         "data_importer_rawdata"."data",
         "data_importer_rawdata"."data_hash",
         "data_importer_rawdata"."last_update"
FROM "data_importer_rawdata"
WHERE ("data_importer_rawdata"."data" -> 'object_id')
      = '"b8a096da-ff83-47dc-8d22-289ddb46b1c1"';
                                                        QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Seq Scan on data_importer_rawdata  (cost=0.00..142508.65 rows=5155 width=856) (actual time=933.902..8240.465 rows=2 loops=1)
Filter: ((data -> 'object_id'::text) = '"b8a096da-ff83-47dc-8d22-289ddb46b1c1"'::jsonb)
Rows Removed by Filter: 1030908
Planning time: 0.158 ms
Execution time: 8240.493 ms

我尝试删除"rawdata_data_idx"并在单个jsonb密钥BTree上使用object_id索引，但表现几乎相同：

abs=# drop index "rawdata_data_idx";
abs=# CREATE INDEX "rawdata_data_object_ididx"
         ON "data_importer_rawdata" USING BTREE ((data->>'object_id'));

abs=# EXPLAIN ANALYZE SELECT
         "data_importer_rawdata"."id",
         "data_importer_rawdata"."created",
         "data_importer_rawdata"."modified",
         "data_importer_rawdata"."entity_id",
         "data_importer_rawdata"."entity_id_key",
         "data_importer_rawdata"."service",
         "data_importer_rawdata"."content_type_id",
         "data_importer_rawdata"."data",
         "data_importer_rawdata"."data_hash",
         "data_importer_rawdata"."last_update"
FROM "data_importer_rawdata"
WHERE ("data_importer_rawdata"."data" -> 'object_id')
      = '"b8a096da-ff83-47dc-8d22-289ddb46b1c1"';
                                                        QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------
Seq Scan on data_importer_rawdata  (cost=0.00..142508.65 rows=5155 width=856) (actual time=951.522..8318.851 rows=2 loops=1)
Filter: ((data -> 'object_id'::text) = '"b8a096da-ff83-47dc-8d22-289ddb46b1c1"'::jsonb)
Rows Removed by Filter: 1030908
Planning time: 0.311 ms
Execution time: 8318.878 ms

有关于此的任何建议吗？不确定这是此类任务的平均表现。

Answer 1

您的查询执行速度很慢，因为无法使用索引。

要使用索引，条件中的表达式必须与索引定义中的表达式相同，即

WHERE "data_importer_rawdata"."data" ->> 'object_id'
      = 'b8a096da-ff83-47dc-8d22-289ddb46b1c1'

postgres 9.6 - jsonb＆amp;索引 - 表现不佳

1 个答案: