Need row returned from upsert - postgresql

Question

I have a table that I need to upsert. If the row already exists then I want to update and return the row. If the row doesn't already exist then I need to insert and return the row. With the query I have below I get the row returned on insert, but not on update.
Table "main.message_account_seen"
Column | Type | Modifiers
----------------+--------------------------+-------------------------------------------------------------------
id | integer | not null default nextval('message_account_seen_id_seq'::regclass)
field_config_id | integer | not null
edit_stamp | timestamp with time zone | not null default now()
audit_stamp | timestamp with time zone |
message_id | integer | not null
account_id | integer |
Here's the sql.
with upsert as (
update message_account_seen set (message_id, account_id, field_config_id ) = (1, 60, 980)
where message_id = 1 and account_id = 60 and field_config_id = 980 returning *
)
insert into message_account_seen (message_id, account_id, field_config_id)
select 1, 60, 980
where not exists (select message_id, account_id, field_config_id from upsert) returning *;
I can't do a postgres function, it needs to be handled in a regular sql query. Also, there is no constraint on the table for uniqueness of row otherwise I would use on conflict. But I'm willing to scrap this query and go with something else if need be.
These are the results when I run the query, and then run it again. You can see that on the insert or first run I get the row returned. However on subsequent runs of the query I get 0 rows returned. I know that it's working because the edit_stamp increases in time. That's a good thing.
# with upsert as (
update message_account_seen set (message_id, account_id, field_config_id ) = (1, 60, 980)
where message_id = 1 and account_id = 60 and field_config_id = 980 returning *
)
insert into message_account_seen (message_id, account_id, field_config_id)
select 1, 60, 980
where not exists (select message_id, account_id, field_config_id from upsert) returning *;
id | field_config_id | edit_stamp | audit_stamp | message_id | account_id
--+-----------------+--------------------------------+-------------+------------+------------
38 | 980 | 09/27/2016 11:43:22.153908 MDT | | 1 | 60
(1 row)
INSERT 0 1
# with upsert as (
update message_account_seen set (message_id, account_id, field_config_id ) = (1, 60, 980)
where message_id = 1 and account_id = 60 and field_config_id = 980 returning *
)
insert into message_account_seen (message_id, account_id, field_config_id)
select 1, 60, 980
where not exists (select message_id, account_id, field_config_id from upsert) returning *;
id | field_config_id | edit_stamp | audit_stamp | message_id | account_id
----+-----------------+------------+-------------+------------+------------
(0 rows)
INSERT 0 0

Answer 1

When the update succeeds its result is not returned in your query. This does it:
with upsert as (
update message_account_seen
set (message_id, account_id, field_config_id ) = (1, 60, 980)
where (message_id, account_id, field_config_id) = (1, 60, 980)
returning *
), ins as (
insert into message_account_seen (message_id, account_id, field_config_id)
select 1, 60, 980
where not exists (select 1 from upsert)
returning *
)
select * from upsert
union all
select * from ins
;

Answer 2

The best option here is to use the new upsert that postgres 9.5 offers, but this requires a unique index on (message_id, account_id, field_config_id). It can be used like this:
INSERT INTO message_account_seen(message_id, account_id, field_config_id)
VALUES (1, 60, 980)
ON CONFLICT (message_id, account_id, field_config_id)
DO UPDATE
SET edit_stamp=now() -- adjust here
RETURNING *;
This is probably the fastest way to do this and guarantees that nothing unexpected will happen if two processes try to upsert into the same table at the same time (your approach doesn't guarantee that).