BUG #17068: Incorrect ordering of a particular row.

Started by PG Bug reporting formalmost 5 years ago7 messagesbugs
Jump to latest
#1PG Bug reporting form
noreply@postgresql.org

The following bug has been logged on the website:

Bug reference: 17068
Logged by: ganesh mahesh
Email address: ganeshmmahesh@gmail.com
PostgreSQL version: 10.15
Operating system: Ubuntu
Description:

Version:
version

--------------------------------------------------------------------------------------------------------------------------------------
PostgreSQL 10.15 (Ubuntu 10.15-0ubuntu0.18.04.1) on x86_64-pc-linux-gnu,
compiled by gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0, 64-bit

Query in question:
SELECT ALL nullnamesb.id, alias3.* FROM nullnamesb RIGHT JOIN (SELECT ALL
alias1.yearsTenured, alias1.firstName FROM nullnames alias1 ORDER BY
alias1.firstName, alias1.yearsTenured) AS alias1 ON
((nullnamesb.yearsTenured <= alias1.yearsTenured) OR (nullnamesb.id =
alias1.yearsTenured)) LEFT JOIN (SELECT DISTINCT alias3.lastName FROM
nullnamesb alias3 ORDER BY alias3.lastName) AS alias3 ON
(((nullnamesb.lastName != alias3.lastName)) OR NOT (nullnamesb.salary <= ANY
(SELECT DISTINCT alias4.salary FROM nullnames alias4 ORDER BY alias4.salary
LIMIT 1))) WHERE nullnamesb.exempt = FALSE ORDER BY alias3.*,
nullnamesb.id;

Info on the tables itself:
\d+ nullnames:

Table "public.nullnames"
Column | Type | Collation | Nullable | Default |
Storage | Stats target | Description
--------------+-----------------------+-----------+----------+---------+----------+--------------+-------------
id | integer | | not null | |
plain | |
firstname | character varying(30) | | | |
extended | |
lastname | character varying(30) | | | |
extended | |
salary | numeric | | not null | |
main | |
exempt | boolean | | not null | |
plain | |
yearstenured | integer | | | |
plain | |
Indexes:
"nullnames_pkey" PRIMARY KEY, btree (id)

\d+ nullnamesb:

Table "public.nullnamesb"
Column | Type | Collation | Nullable | Default |
Storage | Stats target | Description
--------------+-----------------------+-----------+----------+---------+----------+--------------+-------------
id | integer | | not null | |
plain | |
firstname | character varying(30) | | | |
extended | |
lastname | character varying(30) | | | |
extended | |
salary | numeric | | not null | |
main | |
exempt | boolean | | not null | |
plain | |
yearstenured | integer | | not null | |
plain | |
Indexes:
"nullnamesb_pkey" PRIMARY KEY, btree (id)

Data in the table:
select * from nullnames;

id | firstname | lastname | salary | exempt | yearstenured
----+-----------+----------+------------+--------+--------------
0 | Zero | Cool | 25000.01 | t | 10
1 | Acid | Burn | 62530.56 | f | 5
2 | Cereal | Killer | 0 | f |
3 | Lord | Nikon | 2000567.49 | t | 2
4 | Joey | | 0 | f |
5 | Zero | Cool | 25000.01 | t | 10
(6 rows)

select * from nullnamesb;
nullnames=> select * from nullnamesb;
id | firstname | lastname | salary | exempt | yearstenured
----+-----------+----------+------------+--------+--------------
0 | Zero | Cool | 25000.01 | f | 20
1 | Acid | Burn | 62530.56 | f | 5
2 | Cereal | Killer | 0 | t | 0
3 | Lord | Nikon | 2000567.49 | f | 2
4 | Joey | | 0 | f | 0
5 | Zero | Cool | 25000.01 | f | 20
(6 rows)

Partial Query result:
```
.
.
5 | Nikon
1 |
1 |
1 |
3 |
3 |
3 |
3 |
5 |
4 |
4 |
4 |
4 |
(44 rows)
```

Problem:
`5|` ordering is incorrect.

Result expected:
`5|` row should be the last row in the output.

#2David Rowley
dgrowleyml@gmail.com
In reply to: PG Bug reporting form (#1)
Re: BUG #17068: Incorrect ordering of a particular row.

On Tue, 22 Jun 2021 at 21:39, PG Bug reporting form
<noreply@postgresql.org> wrote:

5 | Nikon
1 |
1 |
1 |
3 |
3 |
3 |
3 |
5 |
4 |
4 |
4 |
4 |
(44 rows)
```

Problem:
`5|` ordering is incorrect.

Result expected:
`5|` row should be the last row in the output.

That seems very bug-like to me. Thanks for reporting it.

I've attached an SQL file to make it easier to reproduce.

The top-level sort does appear to contain all the correct columns and
as far as I can tell the sort operation is using all the correct
functions during the comparison. record_cmp() seems to be correctly
returning 0 when the final columns being compared are both NULL.
Putting some debug inside btint4fastcmp(), it seems to receive:

NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 5, b = 3, compare = 1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 5, b = 3, compare = 1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 1, compare = 1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 5, b = 3, compare = 1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 3, b = 5, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 5, b = 3, compare = 1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 4, b = 4, compare = 0
NOTICE: a = 4, b = 4, compare = 0
NOTICE: a = 4, b = 4, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 3, b = 5, compare = -1
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 5, b = 3, compare = 1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 3, compare = -1
NOTICE: a = 3, b = 3, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0
NOTICE: a = 1, b = 1, compare = 0

It seems a bit fishy that the only 4s that appear here are being
compared to 4. How does qsort figure out that 5 > 4 or 4 < 5?

David

Attachments:

bug17068.sqlapplication/octet-stream; name=bug17068.sqlDownload
#3Thomas Munro
thomas.munro@gmail.com
In reply to: David Rowley (#2)
Re: BUG #17068: Incorrect ordering of a particular row.

On Wed, Jun 23, 2021 at 1:07 AM David Rowley <dgrowleyml@gmail.com> wrote:

I've attached an SQL file to make it easier to reproduce.

If you change the SELECT list to output alias3::text, there's a
different value there:

id | alias3
----+----------
1 | (Burn)
1 | (Burn)
1 | (Burn)
3 | (Burn)
3 | (Burn)
3 | (Burn)
3 | (Burn)
5 | (Burn)
1 | (Cool)
1 | (Cool)
1 | (Cool)
3 | (Cool)
3 | (Cool)
3 | (Cool)
3 | (Cool)
5 | (Cool)
1 | (Killer)
1 | (Killer)
1 | (Killer)
3 | (Killer)
3 | (Killer)
3 | (Killer)
3 | (Killer)
5 | (Killer)
1 | (Nikon)
1 | (Nikon)
1 | (Nikon)
3 | (Nikon)
3 | (Nikon)
3 | (Nikon)
3 | (Nikon)
5 | (Nikon)
1 | ()
1 | ()
1 | ()
3 | ()
3 | ()
3 | ()
3 | ()
5 | ()
4 |
4 |
4 |
4 |
(44 rows)

#4David Rowley
dgrowleyml@gmail.com
In reply to: David Rowley (#2)
Re: BUG #17068: Incorrect ordering of a particular row.

On Wed, 23 Jun 2021 at 01:06, David Rowley <dgrowleyml@gmail.com> wrote:

It seems a bit fishy that the only 4s that appear here are being
compared to 4. How does qsort figure out that 5 > 4 or 4 < 5?

Even if I change the < 7 qsort optimization in sort_template.h
(working in master here) to be < 7000 so we bubble sort everything
here, I still don't see 4 being compared to any other value. Something
else must be going on.

David

#5David Rowley
dgrowleyml@gmail.com
In reply to: Thomas Munro (#3)
Re: BUG #17068: Incorrect ordering of a particular row.

On Wed, 23 Jun 2021 at 01:26, Thomas Munro <thomas.munro@gmail.com> wrote:

If you change the SELECT list to output alias3::text, there's a
different value there:

Yeah, if I change it to ORDER BY alias3.lastname, nullnamesb.id;
instead of ORDER BY alias3.*, nullnamesb.id; then the tie break using
btint4fastcmp() is called more often. It's also getting called with
the missing 4s which I mentioned upthread.

See the two outputs. That points me towards something weird going on
in record_cmp().

David

Attachments:

order_by_alias3_lastname.txttext/plain; charset=US-ASCII; name=order_by_alias3_lastname.txtDownload
order_by_alias3_star.txttext/plain; charset=US-ASCII; name=order_by_alias3_star.txtDownload
#6Tom Lane
tgl@sss.pgh.pa.us
In reply to: David Rowley (#5)
Re: BUG #17068: Incorrect ordering of a particular row.

David Rowley <dgrowleyml@gmail.com> writes:

On Wed, 23 Jun 2021 at 01:26, Thomas Munro <thomas.munro@gmail.com> wrote:

If you change the SELECT list to output alias3::text, there's a
different value there:

Yeah, if I change it to ORDER BY alias3.lastname, nullnamesb.id;
instead of ORDER BY alias3.*, nullnamesb.id; then the tie break using
btint4fastcmp() is called more often. It's also getting called with
the missing 4s which I mentioned upthread.
See the two outputs. That points me towards something weird going on
in record_cmp().

I believe Munro's point is that in some rows alias3.* is a NULL composite
value, while in other rows it is a composite containing one NULL, and they
don't sort the same. Presumably the former are from left-join extension
while the latter come from actual table rows having NULL in that column.

(I'd suspected something of the kind, but being caffeine-deprived I'd
first added "alias3.* IS NULL" to the query, which of course fails to
expose the difference. Thanks SQL.)

In short, I see no bug here. It is kind of obscure though.

regards, tom lane

#7David Rowley
dgrowleyml@gmail.com
In reply to: Tom Lane (#6)
Re: BUG #17068: Incorrect ordering of a particular row.

On Wed, 23 Jun 2021 at 02:00, Tom Lane <tgl@sss.pgh.pa.us> wrote:

(I'd suspected something of the kind, but being caffeine-deprived I'd
first added "alias3.* IS NULL" to the query, which of course fails to
expose the difference. Thanks SQL.)

In short, I see no bug here. It is kind of obscure though.

Oh right. Thanks for looking.

David