BUG #15100: sequence behavior on failed insert to a partitioned table

The following bug has been logged on the website:

Bug reference: 15100
Logged by: Istvan Kassai
Email address: kassai1972(at)gmail(dot)com
PostgreSQL version: 10.3
Operating system: linux
Description:

Hi!
I started to deal with the new declarative table partitioning in 10.
I found that a failed insert (due to the field I've tried to insert is out
of partition range) also increments the sequence of the base table, even if
the insert is rejected.

A simple sample:

Column | Type | Collation | Nullable |
Default | Storage | Stats target | Description
----------+------------------------+-----------+----------+-----------------------------------------+----------+--------------+-------------
sor_azon | integer | | not null |
nextval('teszt_sor_azon_seq'::regclass) | plain | |
szoveg | character varying(100) | | |
| extended | |
Partition key: RANGE (szoveg)
Partitions: teszt_a FOR VALUES FROM ('a%') TO ('b%'),
teszt_b FOR VALUES FROM ('b%') TO ('c%')

I've inserted two rows to test if it is working well:

o=# insert into teszt (szoveg) values('aladár');
INSERT 0 1
o=# insert into teszt (szoveg) values('béla');
INSERT 0 1

then I inserted an out-of-range value:
o=# insert into teszt (szoveg) values('cicu');
ERROR: no partition of relation "teszt" found for row
DETAIL: Partition key of the failing row contains (szoveg) = (cicu).

It is working as expected.

After it I queried the content for the table:
o=# select * from teszt;
sor_azon | szoveg
----------+--------
1 | aladár
2 | béla
(2 rows)

Seems all is good. BUT!
After it I created a new partition:

o=# CREATE TABLE teszt_c PARTITION OF teszt


FOR VALUES FROM ('c%') TO ('z%');
CREATE TABLE

Then I tried to insert the last recently rejected value again:

o=# insert into teszt (szoveg) values('cicu');
INSERT 0 1

And a query again:

o=# select * from teszt;
sor_azon | szoveg
----------+--------
1 | aladár
2 | béla
4 | cicu
(3 rows)

As you can see the sor_azon isn't continous. The #3 is missing.
As I think the sequence hasn't rolled back after a failed insert.

I'm not sure is it a bug, but it is better you know about it.

bye
Istvan