I have two tables, one containing all the hot columns and one the static ones. The static table has an unique constraint. When the conflict on the unique constraint triggers only the hot columns in the other table should be updated using the the id from the static table.
For better clarity some code:
CREATE TABLE tag (
id bigserial PRIMARY KEY
, key text
, value text
-- UNIQUE (key, value) -- ?
);
CREATE TABLE tag_hotcolumns (
id bigserial PRIMARY KEY
, hot text
, stuff text
);
with s as (
select id, "key", "value"
from tag
where key = 'key1' and value = 'value1'
), i as (
insert into tag ("key", "value")
select 'key1', 'value1'
where not exists (select 1 from s)
returning id
)
select id
from i
union all
select id
from s
The second block works fine, but I can't get the returned id into the insert statement for the tag_hotcolumns...
I tried:
insert into tag_attributes (with s as (
select id, "key", "value"
from tag
where key = 'key1' and value = 'value1'
), i as (
insert into tag ("key", "value")
select 'key1', 'value1'
where not exists (select 1 from s)
returning id
)
select id, 'hot1', 'stuff1'
from i
union all
select id
from s);
And that gives me "WITH clause containing a data-modifying statement must be at the top level
LINE 5: ), i as ("
Any help would be greatly apreciated :)
dwhitemv from stackexchange helped me solve this. The solution you can find here:
https://dbfiddle.uk/?rdbms=postgres_13&fiddle=f72cae495e6eed579d904a5c7b48f05b
Related
I am trying to insert (clone) some records in a table and need to get source ids and ids that got generated. This simplified example demonstrates my issue. After new records are created, referencing their ids in a SELECT produces no results even though records do get created and subsequent SELECT on the table shows them. It feels like insert and select are happening in different transaction scopes.
CREATE TABLE tbl_value(
id int4 NOT NULL GENERATED ALWAYS AS identity PRIMARY KEY,
some_id INTEGER NOT NULL,
value VARCHAR NOT NULL
);
INSERT INTO tbl_value(some_id, value) VALUES(1000, 'value 1'), (1000, 'value 2'), (1000, 'value 3');
with
outer_input as
(
select id, some_id, value from tbl_value where id in (1,2)
),
inner_insert as
(
INSERT INTO tbl_value(some_id, value)
select 2000, value from outer_input
returning id
)
select * from tbl_value v inner join inner_insert i on v.id = i.id;
I'd like to have column constraint based combination of 2 columns. I don't find the way to use foreign key here, because it should be conditional FK, then. Hope this basic SQL shows the problem:
CREATE TABLE performer_type (
id serial primary key,
type varchar
);
INSERT INTO performer_type ( id, type ) VALUES (1, 'singer'), ( 2, 'band');
CREATE TABLE singer (
id serial primary key,
name varchar
);
INSERT INTO singer ( id, name ) VALUES (1, 'Robert');
CREATE TABLE band (
id serial primary key,
name varchar
);
INSERT INTO band ( id, name ) VALUES (1, 'Animates'), ( 2, 'Zed Leppelin');
CREATE TABLE gig (
id serial primary key,
performer_type_id int default null, /* FK, no problem */
performer_id int default null /* want FK based on previous FK, no good solution so far */
);
INSERT INTO gig ( performer_type_id, performer_id ) VALUES ( 1,1 ), (2,1), (2,2), (1,2), (2,3);
Now, the last INSERT works, but for last 2 value pairs I'd like it fail, because there is no singer ID 2 nor band ID 3. How to set such constraint?
I already asked similar question in Mysql context and only solution was to use trigger. Problem with trigger was: you can't have dynamic list of types and table set. I'd like to add types (and related tables) on the fly.
I also found very promising pattern, but this is upside down for me, I did not figured out, how to turn it to work in my case.
What I am looking here seems to me so useful pattern, I think there must be some common way for it. Is it?
Edit.
Seems, I choose bad items in my examples, so I try make it clear: different performer tables (singer and band) have NO relation between them. gig-table just has to list tasks for different performers, without setting any relations between them.
Another example would items in stock: I may have item_type-table, which defines hundreds of item-types with related tables (for example, orange and house), and there should be table stock which enlists all appearances of items.
PostgreSQL I use is 9.6
Based on #Laurenz Albe answer I form a solution for example above. Main difference: there is parent table performer, which PK is FK/PK for specific performer-tables and is referenced also from gig table.
CREATE TABLE performer_type (
id serial primary key,
type varchar
);
INSERT INTO performer_type ( id, type ) VALUES (1, 'singer' ), ( 2, 'band' );
CREATE TABLE performer (
id serial primary key,
performer_type_id int REFERENCES performer_type(id)
);
CREATE TABLE singer (
id int primary key REFERENCES performer(id),
name varchar
);
INSERT INTO performer ( performer_type_id ) VALUES (1); -- get PK 1 for next statement
INSERT INTO singer ( id, name ) VALUES (1, 'Robert');
CREATE TABLE band (
id int primary key REFERENCES performer(id),
name varchar
);
INSERT INTO performer ( performer_type_id ) VALUES (2); -- get PK 2 for next statement
INSERT INTO singer ( id, name ) VALUES (2, 'Animates');
INSERT INTO performer ( performer_type_id ) VALUES (2); -- get PK 3 for next statement
INSERT INTO singer ( id, name ) VALUES (3, 'Zed Leppelin');
CREATE TABLE gig (
id serial primary key,
performer_id int REFERENCES performer(id)
);
INSERT INTO gig ( performer_id ) VALUES (1), (2), (3), (4);
And the last INSERT fails, as expected:
ERROR: insert or update on table "gig" violates foreign key constraint "gig_performer_id_fkey"
DETAIL: Key (performer_id)=(4) is not present in table "performer".
But
For me there is annoying problem: I have no good way to make distinction which ID is for singer and which for band etc. (in original example I had performer_type_id in gig-table for that), because any performer_id may belong any performer. So I'd like any performer type has it's own ID range, so I create dummy table for every sequence
CREATE TABLE band_id (
id int primary key,
dummy boolean default null
);
CREATE SEQUENCE band_id_seq START 1;
ALTER TABLE band_id ALTER COLUMN id SET DEFAULT nextval('band_id_seq');
CREATE TABLE singer_id (
id int primary key,
dummy boolean default null
);
CREATE SEQUENCE singer_id_seq START 2000000;
ALTER TABLE singer_id ALTER COLUMN id SET DEFAULT nextval('singer_id_seq');
Now, to insert new row into specific perfomer table I have to get next ID for it:
INSERT INTO band_id (dummy) VALUES (NULL);
Trying to figure out, is it possible to solve this process on DB level, or has something to done in App-level. It would be nice, if inserting into band table could:
before trigger inserting into band_id to genereate specific ID
before trigger inserting this new ID into performer-table
include this new ID into INSERT into band
Frist 2 points are easy, but the last point is not clear for now.
I'm trying to construct an object for use from my postgres backend. The tables in question look something like this:
We have some Things that essentially act as rows for a matrix where the columns are Field_Columns. Field_Values are filled cells.
Create Table Platform_User (
serial id PRIMARY KEY
)
Create Table Things (
serial id PRIMARY KEY,
INTEGER user_id REFERENCES Platform_User(id)
)
Create Table Field_Columns (
serial id PRIMARY KEY,
TEXT name,
)
Create Table Field_Values (
INTEGER field_column_id REFERENCES Field_Columns(id),
INTEGER thing_id REFERENCES Things(id)
TEXT content,
PRIMARY_KEY(field_column_id, thing_id)
)
This would be simple if I were trying to load just the Field_Values for a single Thing as JSON, which would look like this:
SELECT JSONB_OBJECT(
ARRAY(
SELECT name
FROM Field_Columns
ORDER BY Field_Columns.id
),
ARRAY(
SELECT Field_Values.content
FROM Fields_Columns
LEFT JOIN Field_Values ON Field_Values.field_column_id = Field_Columns.id
AND Field_Values.thing_id = Things.id
ORDER BY Field_Columns.id)
)
)
FROM Things
WHERE Thing.id = $1
however, I'd like to construct the JSON object to look like this when returned. I want to get an object of all the Fields:Field_Values objects for the Things that a user owns
{
14:
{
'first field':'asdf',
'other field':''
}
25:
{
'first field':'qwer',
'other field':'dfgdsfg'
}
43:
{
'first field':'',
'other field':''
}
}
My efforts to construct this query look like this, but I'm running into the problem where the JSONB object function doesn't want to construct an object where the value of the field is an object itself
SELECT (
JSONB_OBJECT(
ARRAY(SELECT Things.id::TEXT
FROM Things
WHERE Things.user_id = $2
ORDER BY Things.id
),
ARRAY(SELECT JSONB_OBJECT(
ARRAY(
SELECT name
FROM Field_Columns
ORDER BY Field_Columns.id),
ARRAY(
SELECT Field_Values.content
FROM Field_Columns
LEFT JOIN Field_Values ON Field_Values.field_column_Id = Field_Columns.id
AND Field_Values.thing_id = Things.id
ORDER BY Field_Columns.id)
)
FROM Things
WHERE Things.user_id = $2
ORDER BY Things.id
)
)
) AS thing_fields
The specific error I get is function jsonb_object(text[], jsonb[]) does not exist. Is there a way to do this that doesn't involve copious text conversions and nonsense like that? Or will I just need to abandon trying to sort my data in the query and do it in my code instead.
Your DDL scripts are syntactically incorrect so I created these for you:
create table platform_users (
id int8 PRIMARY KEY
);
create table things (
id int8 PRIMARY KEY,
user_id int8 REFERENCES platform_users(id)
);
create table field_columns (
id int8 PRIMARY KEY,
name text
);
create table field_values (
field_column_id int8 REFERENCES field_columns(id),
thing_id int8 REFERENCES things(id),
content text,
PRIMARY KEY(field_column_id, thing_id)
);
I also created some scripts to populate the db:
insert into platform_users(id) values (1);
insert into platform_users(id) values (2);
insert into platform_users(id) values (3);
insert into platform_users(id) values (4);
insert into platform_users(id) values (5);
insert into things(id, user_id) values(1, 1);
insert into things(id, user_id) values(2, 1);
insert into things(id, user_id) values(3, 2);
insert into things(id, user_id) values(4, 2);
insert into field_columns(id, name) values(1, 'col1');
insert into field_columns(id, name) values(2, 'col2');
insert into field_values(field_column_id, thing_id, content) values(1, 1, 'thing1 val1');
insert into field_values(field_column_id, thing_id, content) values(2, 1, 'thing1 val2');
insert into field_values(field_column_id, thing_id, content) values(1, 2, 'thing2 val1');
insert into field_values(field_column_id, thing_id, content) values(2, 2, 'thing2 val2');
Please include such scripts next time when you ask for help, and make sure that your scripts are correct. This will reduce the work needed to answer your question.
You can get your jsonb value by aggregating the key value pairs with jsonb_object_agg
select
t.id,
jsonb_object_agg(fc.name, fv.content)
from
things t inner join
field_values fv on fv.thing_id = t.id inner join
field_columns fc on fv.field_column_id = fc.id
group by 1
The results looking like this:
thing_id;jsonb_value
1;"{"col1": "thing1 val1", "col2": "thing1 val2"}"
2;"{"col1": "thing2 val1", "col2": "thing2 val2"}"
I have a simple table in PostgreSQL that has three columns:
id serial primary key
key varchar
value varchar
I have already seen this question here on SO: Insert, on duplicate update in PostgreSQL? but I'm wondering just how to get the id if it exists, instead of updating. If the standard practice is to always either "insert" or "update if exists", why is that? Is the cost of doing a SELECT (LIMIT 1) greater than doing an UPDATE?
I have the following code
INSERT INTO tag
("key", "value")
SELECT 'key1', 'value1'
WHERE
NOT EXISTS (
SELECT id,"key","value" FROM tag WHERE key = 'key1' AND value = 'value1'
);
which works in the sense that it doesn't insert if exists, but I'd like to get the id. Is there a "RETURNING id" clause or something similar that I could tap in there?
Yes there is returning
INSERT INTO tag ("key", "value")
SELECT 'key1', 'value1'
WHERE NOT EXISTS (
SELECT id, "key", "value"
FROM node_tag
WHERE key = 'key1' AND value = 'value1'
)
returning id, "key", "value"
To return the row if it already exists
with s as (
select id, "key", "value"
from tag
where key = 'key1' and value = 'value1'
), i as (
insert into tag ("key", "value")
select 'key1', 'value1'
where not exists (select 1 from s)
returning id, "key", "value"
)
select id, "key", "value"
from i
union all
select id, "key", "value"
from s
If the row does not exist it will return the inserted one else the existing one.
BTW, if the pair "key"/"value" makes it unique then it is the primary key, and there is no need for an id column. Unless one or both of the "key"/"value" pair can be null.
with vals as (
select 'key5' as key, 'value2' as value
)
insert into Test1 (key, value)
select v.key, v.value
from vals as v
where not exists (select * from Test1 as t where t.key = v.key and t.value = v.value)
returning id
sql fiddle demo
And you can store value returned to variables in form of ... RETURNING field1, field2,... INTO var1, var2,...
RETURNING will normally return a query which would return Error 'query has no destination for result data' if you call it in plpgsql without using its returned result set.
I have a simple table in PostgreSQL that has three columns:
id serial primary key
key varchar
value varchar
I have already seen this question here on SO: Insert, on duplicate update in PostgreSQL? but I'm wondering just how to get the id if it exists, instead of updating. If the standard practice is to always either "insert" or "update if exists", why is that? Is the cost of doing a SELECT (LIMIT 1) greater than doing an UPDATE?
I have the following code
INSERT INTO tag
("key", "value")
SELECT 'key1', 'value1'
WHERE
NOT EXISTS (
SELECT id,"key","value" FROM tag WHERE key = 'key1' AND value = 'value1'
);
which works in the sense that it doesn't insert if exists, but I'd like to get the id. Is there a "RETURNING id" clause or something similar that I could tap in there?
Yes there is returning
INSERT INTO tag ("key", "value")
SELECT 'key1', 'value1'
WHERE NOT EXISTS (
SELECT id, "key", "value"
FROM node_tag
WHERE key = 'key1' AND value = 'value1'
)
returning id, "key", "value"
To return the row if it already exists
with s as (
select id, "key", "value"
from tag
where key = 'key1' and value = 'value1'
), i as (
insert into tag ("key", "value")
select 'key1', 'value1'
where not exists (select 1 from s)
returning id, "key", "value"
)
select id, "key", "value"
from i
union all
select id, "key", "value"
from s
If the row does not exist it will return the inserted one else the existing one.
BTW, if the pair "key"/"value" makes it unique then it is the primary key, and there is no need for an id column. Unless one or both of the "key"/"value" pair can be null.
with vals as (
select 'key5' as key, 'value2' as value
)
insert into Test1 (key, value)
select v.key, v.value
from vals as v
where not exists (select * from Test1 as t where t.key = v.key and t.value = v.value)
returning id
sql fiddle demo
And you can store value returned to variables in form of ... RETURNING field1, field2,... INTO var1, var2,...
RETURNING will normally return a query which would return Error 'query has no destination for result data' if you call it in plpgsql without using its returned result set.