oldest_xmin missing xmin from replication_slots

[Krysztophe]

With such primary slots:

[local]:10021 postgres@pgbench=# select * from pg_replication_slots ;
 slot_name | plugin | slot_type | datoid | database | temporary | active | active_pid |  xmin   | catalog_xmin | restart_lsn | confirmed_flush_lsn 
-----------+--------+-----------+--------+----------+-----------+--------+------------+---------+--------------+-------------+---------------------
 abeille4  | ¤      | physical  |      ¤ | ¤        | f         | t      |    2454458 | 7995805 |            ¤ | 8/AC000F60  | ¤
 abeille2  | ¤      | physical  |      ¤ | ¤        | f         | t      |    2454173 | 7995805 |            ¤ | 8/AC000F60  | ¤
 abeille3  | ¤      | physical  |      ¤ | ¤        | f         | f      |          ¤ |       ¤ |            ¤ | 8/A25AD5B0  | ¤

the oldest_xmin query

check_pgactivity/check_pgactivity

Line 5891 in 2595f39

$PG_VERSION_94 => q{

returns an empty result:

  datname  |    kind     | age | xmin 
-----------+-------------+-----+------
 postgres  | repslot     |   ¤ |    ¤
 postgres  | active_xact |   ¤ |    ¤
 pgbench   | query       |   ¤ |    ¤
 template1 | 2pc         |   ¤ |    ¤
 template1 | idle_xact   |   ¤ |    ¤
 pgbench   | repslot     |   ¤ |    ¤
 pgbench   | active_xact |   ¤ |    ¤
 postgres  | query       |   ¤ |    ¤
 template1 | walsender   |   ¤ |    ¤
 template1 | repslot     |   ¤ |    ¤
 pgbench   | idle_xact   |   ¤ |    ¤
 template1 | active_xact |   ¤ |    ¤
 pgbench   | walsender   |   ¤ |    ¤
 postgres  | 2pc         |   ¤ |    ¤
 postgres  | idle_xact   |   ¤ |    ¤
 postgres  | walsender   |   ¤ |    ¤
 template1 | query       |   ¤ |    ¤
 pgbench   | 2pc         |   ¤ |    ¤

although the part of the query that queries it is okay:

  SELECT 'repslot' AS kind,
pgbench-#             database AS datname,
pgbench-#             xmin AS xmin
pgbench-#             FROM  pg_replication_slots;

  kind   | datname |  xmin   
---------+---------+---------
 repslot | ¤       | 7995805
 repslot | ¤       | 7995805
 repslot | ¤       |       ¤

The datname is empty and it is filtered in the RIGHT JOIN at the end.

When replacing it with a FULL OUTER JOIN and adding 2 COALESCE, I obtain this, which is much more useful:

  datname  |    kind     | age |  xmin   
-----------+-------------+-----+---------
 ¤         | active_xact |   ¤ |       ¤
 ¤         | repslot     |   0 | 7995805
 ¤         | walsender   |   ¤ |       ¤
 postgres  | repslot     |   ¤ |       ¤
 postgres  | active_xact |   ¤ |       ¤
 pgbench   | query       |   ¤ |       ¤
 template1 | 2pc         |   ¤ |       ¤
 template1 | idle_xact   |   ¤ |       ¤
 pgbench   | repslot     |   ¤ |       ¤
 pgbench   | active_xact |   ¤ |       ¤
 postgres  | query       |   ¤ |       ¤
 template1 | walsender   |   ¤ |       ¤
 template1 | repslot     |   ¤ |       ¤
 pgbench   | idle_xact   |   ¤ |       ¤
 template1 | active_xact |   ¤ |       ¤
 pgbench   | walsender   |   ¤ |       ¤
 postgres  | 2pc         |   ¤ |       ¤
 postgres  | idle_xact   |   ¤ |       ¤
 postgres  | walsender   |   ¤ |       ¤
 template1 | query       |   ¤ |       ¤
 pgbench   | 2pc         |   ¤ |       ¤

I'm wondering it this should not have been a LEFT JOIN from the beginning.
So many empty values pollute the perfdata.

Or did I miss something?