PostgreSQL:如何在JSONB字段中汇总所有属性?

时间:2016-02-01 11:18:33

标签: json postgresql postgresql-9.4 jsonb

我正在使用Postgres 9.4。我有一个JSONB字段:

     Column      │         Type         │                             Modifiers
─────────────────┼──────────────────────┼────────────────────────────────────────────────────────────────────
 id              │ integer              │ not null default
 practice_id     │ character varying(6) │ not null
 date            │ date                 │ not null
 pct_id          │ character varying(3) │
 astro_pu_items  │ double precision     │ not null
 astro_pu_cost   │ double precision     │ not null
 star_pu         │ jsonb                │

我可以很好地查询JSONB字段的原始值:

SELECT star_pu FROM mytable limit 1;
star_pu │ {"statins_cost": 16790.692924903742, "hypnotics_adq": 18523.58385328709, "laxatives_cost": 8456.98405165182, "analgesics_cost": 48271.21822239242, "oral_nsaids_cost": 9911.336052088493, "antidepressants_adq": 186715.7, "antidepressants_cost": 26885.54622478343, "bronchodilators_cost": 26646.54899847902, "cox-2_inhibitors_cost": 2063.4652015406728, "antiplatelet_drugs_cost": 4844.798321177439, "drugs_for_dementia_cost": 3390.569564110721, "antiepileptic_drugs_cost": 44990.94756286502, "oral_antibacterials_cost": 21047.048353859234, "oral_antibacterials_item": 5096.6501798218205, "ulcer_healing_drugs_cost": 15999.05326260261, "lipid-regulating_drugs_cost": 24711.589440943662, "proton_pump_inhibitors_cost": 14545.398978447573, "inhaled_corticosteroids_cost": 50759.91062192373, "calcium-channel_blockers_cost": 11571.457036131978, "omega-3_fatty_acid_compounds_adq": 2026.0, "benzodiazepine_caps_and_tabs_cost": 1800.2581325567717, "bisphosphonates_and_other_drugs_cost": 2996.912924744617, "drugs_acting_on_benzodiazepine_receptors_cost": 2993.142806352308, "drugs_affecting_the_renin_angiotensin_system_cost": 20255.500615282508, "drugs_used_in_parkinsonism_and_related_disorders_cost": 9812.457888596877}

现在我想在整个表格中使用SUM JSONB值,但我不知道如何做到这一点。理想情况下,我会找回一个字典,其中键如上所述,值是求和值。

我可以明确地对SUM一个JSONB字段执行以下操作:

    SELECT date, SUM(total_list_size) as total_list_size, 
    SUM((star_pu->>'oral_antibacterials_item')::float) AS star_pu_oral_antibac_items
    FROM mytable GROUP BY date ORDER BY date

但是如何计算JSONB字段中所有属性的总和 - 并且最好将整个字段作为字典返回?理想情况下,我会得到类似的东西:

star_pu │ {"statins_cost": very-large-number, "hypnotics_adq": very-large-number, ...

我想我可以通过明确地对每个键进行手动操作来获取每个字段,但是我拥有JSONB字段的全部原因是有很多键并且它们可能会发生变化。

可以安全地假设JSONB字段仅包含键和值,即深度为1.

3 个答案:

答案 0 :(得分:4)

查询应该完成这项工作:

select date, json_object_agg(key, val)
from (
    select date, key, sum(value::numeric) val
    from mytable t, jsonb_each_text(star_pu)
    group by date, key
    ) s
group by date;

生成的json值将按键按字母顺序排序(json_object_agg ()的副作用)。我不知道这是否重要。

答案 1 :(得分:3)

我写了Postgres extension就是这么做的。安装完成后,您可以执行以下操作:

contrib.learn

对于200万行,基准是4s,@ klin的答案需要11s

答案 2 :(得分:0)

可能有更好的方法,但至少可行:

WITH
  keys AS (SELECT DISTINCT jsonb_object_keys(star_pu) AS key FROM mytable),
  sums AS (SELECT key, sum((star_pu->>key)::float) AS total FROM keys, mytable GROUP BY key)
  SELECT json_object(array_agg(key), array_agg(total::text))::jsonb FROM sums

基本上它将jsonbs分解为行,从中获取名称,将它们相加,聚合成数组并创建jsonb结构。遗憾的是,没有jsonb_object()函数,因此我们必须将其转换为json,然后转换为jsonb。