Re: [BUGS] Urgent - SQL left join bug? - Mailing list pgsql-bugs

On Wed, Jun 21, 2017 at 9:06 AM, Kaijiang Chen <chenkaijiang@gmail.com> wrote:

Hi, I'm running PostgreSQL 9.4.10 on CentOS 6.5. It looks like that I found a bug with left join. It is very URGENT since it is running in the production servers.

### Conditions: ###

I have 2 tables:

TABLE 1: (2171209 records)

\d prescription_herbs

                                           Table "public.prescription_herbs"

       Column        |              Type              |                            Modifiers                            

---------------------+--------------------------------+-----------------------------------------------------------------

 id                  | integer                        | not null default nextval('prescription_herbs_id_seq'::regclass)

 prescription_id     | integer                        | not null

 herb_id             | integer                        | not null

 weight              | integer                        | not null

 created_at          | timestamp(0) without time zone | not null

 updated_at          | timestamp(0) without time zone | not null

 deleted_at          | timestamp(0) without time zone | 

 price               | numeric(10,5)                  | 

 special_manufacture | character varying(255)         | 

 cost                | numeric(10,5)                  | 

 pharmacy_id         | integer                        | 

Indexes:

    "prescription_herbs_pkey" PRIMARY KEY, btree (id)

    "prescription_herbs_hid" btree (herb_id)

    "prescription_herbs_prid" btree (prescription_id)

TABLE 2: (4406 records)

\d pharmacy_herbs

                                       Table "public.pharmacy_herbs"

   Column    |              Type              |                          Modifiers                          

-------------+--------------------------------+-------------------------------------------------------------

 id          | integer                        | not null default nextval('pharmacy_herbs_id_seq'::regclass)

 pharmacy_id | integer                        | 

 herb_id     | integer                        | 

 cost        | numeric(10,5)                  | 

 price       | numeric(10,5)                  | 

 no          | character varying(255)         | 

 deleted_at  | timestamp(0) without time zone | 

 created_at  | timestamp(0) without time zone | not null

 updated_at  | timestamp(0) without time zone | not null

 name        | character varying(255)         | 

Indexes:

    "pharmacy_herbs_pkey" PRIMARY KEY, btree (id)

    "pharmacy_herbs_herb_id" btree (herb_id)

### BUG: ###

I ran a SQL: 

select pha.id,ph.herb_id,pha.name,ph.weight 

from prescription_herbs as ph left join pharmacy_herbs as pha on ph.herb_id=pha.herb_id 

where ph.prescription_id=116285 and ph.deleted_at is null and pha.deleted_at is null and pha.pharmacy_id=22 order by ph.herb_id;

Expected:

It should have 10 rows because the SQL "select * from prescription_herbs as ph where ph.prescription_id=116285 and ph.deleted_at is null" returned 10 rows and I'm using LEFT JOIN in the above SQL.

Actual Result:

It returned only 9 rows and the result is the same as JOIN (not LEFT JOIN).

### More info: ###

I explain the SQL:

explain select pha.id,ph.herb_id,pha.name,ph.weight 

from prescription_herbs as ph left join pharmacy_herbs as pha on ph.herb_id=pha.herb_id 

where ph.prescription_id=116285 and ph.deleted_at is null and pha.deleted_at is null and pha.pharmacy_id=22 order by ph.herb_id;

I got:

 Sort  (cost=131.73..131.76 rows=10 width=24)

   Sort Key: ph.herb_id

   ->  Hash Join  (cost=122.02..131.56 rows=10 width=24)

         Hash Cond: (ph.herb_id = pha.herb_id)

         ->  Index Scan using prescription_herbs_prid on prescription_herbs ph  (cost=0.43..9.68 rows=23 width=8)

               Index Cond: (prescription_id = 116285)

               Filter: (deleted_at IS NULL)

         ->  Hash  (cost=113.08..113.08 rows=681 width=20)

               ->  Seq Scan on pharmacy_herbs pha  (cost=0.00..113.08 rows=681 width=20)

                     Filter: ((deleted_at IS NULL) AND (pharmacy_id = 22))

I think the above "Hash Join" SHOULD BE "Hash Left Join", right?

I tried to explain another SQL: 

explain select * from doctors d left join prescriptions p on d.id=p.doctor_id;

I got:

 Hash Right Join  (cost=2159.33..31453.58 rows=130330 width=2936)

   Hash Cond: (p.doctor_id = d.id)

   ->  Seq Scan on prescriptions p  (cost=0.00..9273.30 rows=130330 width=495)

   ->  Hash  (cost=576.37..576.37 rows=5037 width=2441)

         ->  Seq Scan on doctors d  (cost=0.00..576.37 rows=5037 width=2441)

The "Hash Right Join" is the correct node.

Any help is very appreciated! WAITING...

Thanks,

Kaijiang

This is not a bug. The conditions of the "right" table pha - basically (pha.pharmacy_id=22) and secondary ((pha.deleted_at is null) )- that you have put in the WHERE clause make the query act as if it was an INNER join.

Move them to the ON and you'll get your 10 rows.

select ...

from prescription_herbs as ph left join pharmacy_herbs as pha
  on ph.herb_id=pha.herb_id and pha.deleted_at is null and pha.pharmacy_id=22

where ph.prescription_id=116285 and ph.deleted_at is null
order by ...

The (pha.deleted_at is null) part may give different results depending on where it is placed (ON vs WHERE) but my guess - since you want all the 10 rows of the left table - is that it should be in ON, too

Pantelis