db:Postgresql-14. 这将是一个罕见的转变,并且我在寻找建议的/改进,可以做所以我可以学习/磨练我的postgres/json技能(和速度/优化这一非常缓慢的查询)。
我们收到的可变尺寸/结构星从外部api。
每个json目的是调查的响应。 每个嵌套的"问答"的对象可以有一个完全不同的结构。 在总共大约有-5的已知结构。
响应的对象存储在一个jsonb列jsonb_ops杜松子酒的索引。
表具有约500,000行。 每行的jsonb列的对象都具有约200套价值观。
我们的目标是提取所有的嵌套问答的答复到另一个表中的问题,答案。 在目的地表,我们将做广泛的查询与FTS和卦,是瞄准为模式的简单性。 这就是为什么我提取到一个简单的表格而不是做什么更异域与jsonb查询。 还有一个很大的元数据令人讨厌在这些对象,我不需要。 所以我也希望能够节省一些空间归档源表(这是5GB+索引).
具体地说,我爱学习一个更优雅的方式,穿越提取和计的目标表。
而且我一直无法找出一种方法投的结果,实际sql文本,而不是援引jsontext(通常我会用->>,::文字,或_text版本的jsonb功能)
这是一个非常简化的版本启目,以便于运行这一点。
谢谢你提前!
create table test_survey_processing(
id integer generated always as identity constraint test_survey_processing_pkey primary key,
json_data jsonb
);
insert into test_survey_processing (json_data)
values ('{"survey_data": {"2": {"answer": "Option 1", "question": "radiobuttonquesiton"}, "3": {"options": {"10003": {"answer": "Option 1"}, "10004": {"answer": "Option 2"}}, "question": "checkboxquestion"}, "5": {"answer": "Column 2", "question": "Row 1"}, "6": {"answer": "Column 2", "question": "Row 2"}, "7": {"question": "checkboxGRIDquesiton", "subquestions": {"8": {"10007": {"answer": "Column 1", "question": "Row 1 : Column 1"}, "10008": {"answer": "Column 2", "question": "Row 1 : Column 2"}}, "9": {"10007": {"answer": "Column 1", "question": "Row 2 : Column 1"}, "10008": {"answer": "Column 2", "question": "Row 2 : Column 2"}}}}, "11": {"answer": "Option 1", "question": "Row 1"}, "12": {"answer": "Option 2", "question": "Row 2"}, "13": {"options": {"10011": {"answer": "Et molestias est opt", "option": "Option 1"}, "10012": {"answer": "Similique magnam min", "option": "Option 2"}}, "question": "textboxlist"}, "14": {"question": "textboxgridquesiton", "subquestions": {"15": {"10013": {"answer": "Qui error magna omni", "question": "Row 1 : Column 1"}, "10014": {"answer": "Est qui dolore dele", "question": "Row 1 : Column 2"}}, "16": {"10013": {"answer": "vident mol", "question": "Row 2 : Column 1"}, "10014": {"answer": "Consectetur dolor co", "question": "Row 2 : Column 2"}}}}, "17": {"question": "contactformquestion", "subquestions": {"18": {"answer": "Rafael", "question": "First Name"}, "19": {"answer": "Adams", "question": "Last Name"}}}, "33": {"question": "customgroupquestion", "subquestions": {"34": {"answer": "Sed magnam enim non", "question": "customgroupTEXTbox"}, "36": {"answer": "Option 2", "question": "customgroupradiobutton"}, "37": {"options": {"10021": {"answer": "Option 1", "option": "customgroupCHEC KBOX question : Option 1"}, "10022": {"answer": "Option 2", "option": "customgroupCHEC KBOX question : Option 2"}}, "question": "customgroupCHEC KBOX question"}}}, "38": {"question": "customTABLEquestion", "subquestions": {"10001": {"answer": "Option 1", "question": "customTABLEquestioncolumnRADIO"}, "10002": {"answer": "Option 2", "question": "customTABLEquestioncolumnRADIO"}, "10003": {"options": {"10029": {"answer": "OPTION1"}, "10030": {"answer": "OPTION2"}}, "question": "customTABLEquestioncolumnCHECKBOX"}, "10004": {"options": {"10029": {"answer": "OPTION1"}, "10030": {"answer": "OPTION2"}}, "question": "customTABLEquestioncolumnCHECKBOX"}, "10005": {"answer": "Aperiam itaque dolor", "question": "customTABLEquestioncolumnTEXTBOX"}, "10006": {"answer": "Hic qui numquam inci", "question": "customTABLEquestioncolumnTEXTBOX"}}}}}');
create index test_survey_processing_gin_index on test_survey_processing using gin (json_data);
-- the query I'm using (it works, but it is unmanageably slow)
-- EXPLAIN (ANALYZE, VERBOSE, BUFFERS, FORMAT JSON)
select level1.value['question'] question, level1.value['answer'] as answer ,tgsr.json_data['survey_data']
from test_survey_processing tgsr,
jsonb_each(tgsr.json_data['survey_data']::jsonb) level1
-- where survey_id = 6633968 and id = 4
union
select level1.value['question'] question, jsonb_path_query(level1.value, '$.answer')::jsonb as answer ,tgsr.json_data['survey_data']
from test_survey_processing tgsr,
jsonb_each(tgsr.json_data['survey_data']::jsonb) level1
-- where survey_id = 6633968 and id = 4
union
select level1.value['question'] question, jsonb_path_query(level1.value, '$.options.*.answer')::jsonb as answer ,tgsr.json_data['survey_data']
from test_survey_processing tgsr,
jsonb_each(tgsr.json_data['survey_data']::jsonb) level1
-- where survey_id = 6633968 and id = 4
union
select level1.value['question'] question, jsonb_path_query(level1.value, '$.subquestions.*.*.answer')::jsonb as answer ,tgsr.json_data['survey_data']
from test_survey_processing tgsr,
jsonb_each(tgsr.json_data['survey_data']::jsonb) level1
-- where survey_id = 6633968 and id = 4
后续编辑后提炼和得到的结果,我需要
这是查询我结束了在运行。 它采取了11min处理和插入34million记录。 它是好的,因为它是一个时间的操作。
一些评论对我所做的更改
-我->和->>,而不是[n]因为我读,即使在pg14,下标不使用索引(未确定,如果该事项在从)
-对 "to_json(...) #>> '{}'"
是我转变id string到一个没有引用的字符串基于这样的: 堆溢出答案
create table respondent_questions_answers as
select tgsr.id,tgsr.survey_id,level1.value ->> 'question' question, '' as sub_question,
to_json(jsonb_path_query(level1.value, '$.answer')) #>> '{}' as answer
from test_survey_processing tgsr, jsonb_each(tgsr.json -> 'survey_data') level1
union
select tgsr.id,tgsr.survey_id,level1.value ->> 'question' question,
to_json(jsonb_path_query(level1.value, '$.options.*.option')) #>> '{}' as sub_question,
to_json(jsonb_path_query(level1.value, '$.options.*.answer')) #>> '{}' as answer
from test_survey_processing tgsr, jsonb_each(tgsr.json -> 'survey_data') level1
union
select tgsr.id,tgsr.survey_id,level1.value ->> 'question' question,
to_json(jsonb_path_query(level1.value, '$.subquestions.*.*.question')) #>> '{}' as sub_question,
to_json(jsonb_path_query(level1.value, '$.subquestions.*.*.answer')) #>> '{}' as answer
from test_survey_processing tgsr, jsonb_each(tgsr.json -> 'survey_data') level1
union
select tgsr.id,tgsr.survey_id,level1.value ->> 'question' question,
to_json(jsonb_path_query(level1.value, '$.subquestions.*.question')) #>> '{}' as sub_question,
to_json(jsonb_path_query(level1.value, '$.subquestions.*.answer')) #>> '{}' as answer
from test_survey_processing tgsr, jsonb_each(tgsr.json -> 'survey_data') level1;
最后编辑后接受下面的回答,因为解决方案
谢谢是@爱德华*H.回答,并更好地了解如何正确使用jsonb_path_query,我能够消除所有的 UNION SELECT
,发现一些价值观已经失踪,并且除需要对该to_json丐。 虽 CROSS JOIN LATERAL
是隐含的结果的方式的功能,它是最好的形式包括 JOIN
而不是逗号,因为它们更紧密结合,并更易于阅读。 下面是最后的查询我的使用。
SELECT concat_ws(' ',
qu.value::jsonb->>'question'
, an.answer::jsonb->>'question'
, an.answer::jsonb->>'option') AS question
, an.answer::jsonb->>'answer' AS answer
-- , tgsr.json_data->>'survey_data'
FROM test_survey_processing tgsr
CROSS JOIN LATERAL jsonb_each(tgsr.json_data->'survey_data') AS qu
CROSS JOIN LATERAL jsonb_path_query(qu.value::jsonb, '$.** ? (exists(@.answer))') AS an(answer)