-
-
Save MarkusH/4c03d0867ba37b6b89379bced12cb3ae to your computer and use it in GitHub Desktop.
WITH tab AS ( | |
SELECT | |
a.attrelid::regclass::text AS t, | |
a.attname AS c, | |
pg_get_serial_sequence(a.attrelid::regclass::text, a.attname) AS s, | |
nextval(pg_get_serial_sequence(a.attrelid::regclass::text, a.attname)) AS v | |
FROM pg_attribute a | |
JOIN pg_class c ON c.oid = a.attrelid | |
WHERE | |
a.attrelid::regclass::text LIKE '%' | |
AND c.relkind IN ('r', 'p') /* regular and partitioned tables */ | |
AND a.attnum > 0 | |
AND NOT a.attisdropped | |
AND a.atttypid = ANY ('{int,int8,int2}'::regtype[]) | |
AND EXISTS ( | |
SELECT FROM pg_attrdef ad | |
WHERE | |
ad.adrelid = a.attrelid | |
AND ad.adnum = a.attnum | |
AND ( | |
pg_get_expr(ad.adbin, ad.adrelid) | |
= | |
'nextval(''' | |
|| ( | |
pg_get_serial_sequence(a.attrelid::regclass::text, a.attname) | |
)::regclass | |
|| '''::regclass)' | |
) | |
) | |
ORDER BY a.attnum | |
) | |
SELECT | |
'BEGIN; ' | |
|| 'LOCK TABLE ' || quote_ident(t) || ' IN ACCESS EXCLUSIVE MODE; ' | |
|| 'ALTER TABLE ' || quote_ident(t) || ' ALTER COLUMN ' || quote_ident(c) || ' DROP DEFAULT; ' | |
|| 'DROP SEQUENCE ' || s || '; ' | |
|| 'ALTER TABLE ' || quote_ident(t) || ' ALTER ' || quote_ident(c) || ' ADD GENERATED BY DEFAULT AS IDENTITY (RESTART ' || v || '); ' | |
|| 'COMMIT;' | |
FROM tab; |
Does any know if any changes need to be made to the migrations of Django after runing the SQL manually to change the columns? Like if I ran the migrations to recreate my database, would it create the tables using the old
serial
type of columns?
If you recreate the database with Django 4.1 or higher, you do not need to rerun the SQL.
Unless I'm mistaken, this appears to get the nextval
of the old sequence at the the SQL is generated, rather than at the time the generated SQL itself it run. That means that if any rows are inserted in-between the time the SQL is generated and when it's run, the new sequence value will be incorrect. Shouldn't the call to nextval
be part of the generated SQL (and inside the exclusive table lock)? That'd also mean re-ordering commands so that the DROP SEQUENCE
statement is after the last ALTER TABLE
.
Update: Here's an updated version of this script that fixes the issue w/ nextval timing.
WITH tab AS (
SELECT
a.attrelid::regclass::text AS table_name,
a.attname AS pk_col,
pg_get_serial_sequence(a.attrelid::regclass::text, a.attname) AS seq_name
FROM pg_attribute a
JOIN pg_class c ON c.oid = a.attrelid
WHERE
a.attrelid::regclass::text LIKE '%'
AND c.relkind IN ('r', 'p') /* regular and partitioned tables */
AND a.attnum > 0
AND NOT a.attisdropped
AND a.atttypid = ANY ('{int,int8,int2}'::regtype[])
AND EXISTS (
SELECT FROM pg_attrdef ad
WHERE
ad.adrelid = a.attrelid
AND ad.adnum = a.attnum
AND (
pg_get_expr(ad.adbin, ad.adrelid)
=
'nextval('''
|| (
pg_get_serial_sequence(a.attrelid::regclass::text, a.attname)
)::regclass
|| '''::regclass)'
)
)
ORDER BY a.attnum
)
SELECT
'BEGIN; '
|| 'DO $$ DECLARE nextSeqVal bigint; '
|| 'BEGIN '
|| 'LOCK TABLE ' || quote_ident(table_name) || ' IN ACCESS EXCLUSIVE MODE; '
|| 'ALTER TABLE ' || quote_ident(table_name) || ' ALTER COLUMN ' || quote_ident(pk_col) || ' DROP DEFAULT; '
|| 'nextSeqVal := nextval(''' || seq_name || '''); '
|| 'DROP SEQUENCE ' || seq_name || '; '
|| 'EXECUTE format(''ALTER TABLE ' || quote_ident(table_name) || ' ALTER ' || quote_ident(pk_col) || ' ADD GENERATED BY DEFAULT AS IDENTITY (RESTART %s);'', nextSeqVal); '
|| 'END $$; '
|| 'COMMIT;'
FROM tab;
As well, here's an example bash script of how you can generate and run the migration code on a batch of DBs:
#!/usr/bin/env bash
set -euxo pipefail
DB_URL="postgres://foo:[email protected]:5432"
for DB in foo bar baz
do
psql \
-d "${DB_URL}/${DB}" \
-f serial-to-identity.sql \
| grep 'BEGIN;' \
| awk '{$1=$1};1' \
> "migrate-${DB}.sql";
psql -d "${DB_URL}/${DB}" \
-f "migrate-${DB}.sql";
echo "Finished ${DB}";
done
This should execute directly
DO $$
DECLARE
rec record;
table_ident text;
seq_ident text;
max_val bigint;
BEGIN
FOR rec IN
SELECT
a.attrelid::regclass::text AS t,
a.attname AS c,
pg_get_serial_sequence(a.attrelid::regclass::text, a.attname) AS s
FROM pg_attribute a
JOIN pg_class c ON c.oid = a.attrelid
WHERE
a.attrelid::regclass::text LIKE '%'
AND c.relkind IN ('r', 'p') -- regular and partitioned tables
AND a.attnum > 0
AND NOT a.attisdropped
AND a.atttypid = ANY ('{int,int8,int2}'::regtype[])
AND EXISTS (
SELECT 1
FROM pg_attrdef ad
WHERE
ad.adrelid = a.attrelid
AND ad.adnum = a.attnum
AND (
pg_get_expr(ad.adbin, ad.adrelid)
=
'nextval('''
|| (pg_get_serial_sequence(a.attrelid::regclass::text, a.attname))::regclass
|| '''::regclass)'
)
)
ORDER BY a.attnum
LOOP
-- Handle schema-qualified table names
table_ident := CASE
WHEN position('.' in rec.t) > 0
THEN quote_ident(split_part(rec.t, '.', 1)) || '.' || quote_ident(split_part(rec.t, '.', 2))
ELSE quote_ident(rec.t)
END;
-- Handle schema-qualified sequence names
seq_ident := CASE
WHEN position('.' in rec.s) > 0
THEN quote_ident(split_part(rec.s, '.', 1)) || '.' || quote_ident(split_part(rec.s, '.', 2))
ELSE quote_ident(rec.s)
END;
-- Lock the table to prevent concurrent modifications
EXECUTE 'LOCK TABLE ' || table_ident || ' IN ACCESS EXCLUSIVE MODE';
-- Get the max value of the column to set the new identity restart value
EXECUTE 'SELECT COALESCE(MAX(' || quote_ident(rec.c) || '), 0) + 1 FROM ' || table_ident INTO max_val;
-- Drop the default and sequence
EXECUTE 'ALTER TABLE ' || table_ident || ' ALTER COLUMN ' || quote_ident(rec.c) || ' DROP DEFAULT';
EXECUTE 'DROP SEQUENCE ' || seq_ident;
-- Convert the column to GENERATED ALWAYS AS IDENTITY
EXECUTE 'ALTER TABLE ' || table_ident || ' ALTER COLUMN ' || quote_ident(rec.c)
|| ' ADD GENERATED ALWAYS AS IDENTITY (RESTART WITH ' || max_val || ')';
END LOOP;
END $$;
Cool! Thank you for posting this! I just had to remove the single quotes around table names (or use double quotes) in the generated sql, then it worked perfect!
I found this gist from a great article that helped me understand this change in Django 4.1
Does any know if any changes need to be made to the migrations of Django after runing the SQL manually to change the columns? Like if I ran the migrations to recreate my database, would it create the tables using the old
serial
type of columns?