33

For example when I'm using query which returns record ids

INSERT INTO projects(name)
VALUES (name1), (name2), (name3) returning id;

Which produce output:

1
2
3

Will this ids point to corresponding inserted values?

1 -> name1
2 -> name2
3 -> name3
Erwin Brandstetter
  • 185,527
  • 28
  • 463
  • 633
Sergey
  • 433
  • 1
  • 4
  • 4

2 Answers2

37

The answer for this simple case is: "Yes". Rows are inserted in the provided order in the VALUES expression. And if your id column is a serial type, values from the underlying sequence will be fetched in that order.

But this is an implementation detail and there are no guarantees. In particular, the order is not necessarily maintained in more complex queries with WHERE conditions or joins.

You might also get gaps or other rows mixed in if you have concurrent transactions writing to the same table at the same time. Unlikely, but possible.

There is no "natural" order in a database table. While the physical order of rows (which is reflected in the system column ctid) will correspond to their inserted order initially, that may change any time. UPDATE, DELETE, VACUUM and other commands can change the physical order of rows. To SELECT rows in any particular order, you must add an ORDER BY clause. Values for id, once generated, are stable and decoupled from any physical order, of course.

Erwin Brandstetter
  • 185,527
  • 28
  • 463
  • 633
8

Erwin Brandstetter's answer may not be correct in a certain case.

We've done an INSERT INTO ... SELECT bar,baz FROM foo ORDER BY bar and we see that the SELECT ctid,* FROM foo shows that the physical ordering of the rows in the table doesn't match the the insert order exactly, it seems scrambled up a little bit. Note that our table has a jsonb column with highly variable data size. Experimentally truncating the jsonb data during the insert caused the insert order to be correct.

Community
  • 1
user2052675
  • 81
  • 1
  • 1