behaviour change - default_tablesapce + partition table

classic Classic list List threaded Threaded
11 messages Options
Reply | Threaded
Open this post in threaded view
|

behaviour change - default_tablesapce + partition table

Rushabh Lathia
Hi,

Consider the below test:

create tablespace mytbs location '/home/rushabh/mywork/workspace/pg/';
create table test ( a int , b int ) partition by list (a);

set default_tablespace to mytbs;
create table test_p1 partition of test for values in (1);

In the above test, after the setting the default_tablespace I am creating
a partition table and expecting that to get created under "mytbs" tablespace.

But that is not the case:

postgres@66247=#select relname, reltablespace from pg_class where relname = 'test_p1';
 relname | reltablespace
---------+---------------
 test_p1 |             0
(1 row)

I noticed the behaviour change for default_tablespace with partition table
with below commit.

commit 87259588d0ab0b8e742e30596afa7ae25caadb18
Author: Alvaro Herrera <[hidden email]>
Date:   Thu Apr 25 10:20:23 2019 -0400

    Fix tablespace inheritance for partitioned rels

    Commit ca4103025dfe left a few loose ends.  The most important one
    (broken pg_dump output) is already fixed by virtue of commit
    3b23552ad8bb, but some things remained:

I don't think that new behaviour is intended and if it's an intended change
than need to fix pg_dump as well - other wise lets say,
1) create the above test on v11
2) take a dump
3) restore on v12

will end up partition into wrong tablesapce.

Looking at the commit changes, it seems like at condition when no other
tablespace is specified, we default the tablespace to the parent partitioned
table's.

    else if (stmt->partbound)
    {
        /*
         * For partitions, when no other tablespace is specified, we default
         * the tablespace to the parent partitioned table's.
         */
        Assert(list_length(inheritOids) == 1);
        tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
    }

But here it doesn't consider the default_tablespace if the parent partitioned
tablespace is an InvalidOid (which was the care before this commit).

PFA patch to fix the same.

Thanks,

--
Rushabh Lathia

default_tablespace_for_partition.patch (804 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Amit Langote-2
Agree that this behavior change seems unintentional.

On 2019/05/17 12:40, Rushabh Lathia wrote:

> Looking at the commit changes, it seems like at condition when no other
> tablespace is specified, we default the tablespace to the parent partitioned
> table's.
>
>     else if (stmt->partbound)
>     {
>         /*
>          * For partitions, when no other tablespace is specified, we default
>          * the tablespace to the parent partitioned table's.
>          */
>         Assert(list_length(inheritOids) == 1);
>         tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
>     }
>
> But here it doesn't consider the default_tablespace if the parent
> partitioned
> tablespace is an InvalidOid (which was the care before this commit).
>
> PFA patch to fix the same.

+
+ if (!OidIsValid(tablespaceId))
+ tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,
partitioned);
  }
  else
  tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,

Why not change it like this instead:

@@ -681,7 +681,8 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid
ownerId,
         Assert(list_length(inheritOids) == 1);
         tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
     }
-    else
+
+    if (!OidIsValid(tablespaceId))
         tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,
                                             partitioned);

Thanks,
Amit



Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Amit Langote-2
In reply to this post by Rushabh Lathia
On 2019/05/17 12:40, Rushabh Lathia wrote:

> Hi,
>
> Consider the below test:
>
> create tablespace mytbs location '/home/rushabh/mywork/workspace/pg/';
> create table test ( a int , b int ) partition by list (a);
>
> set default_tablespace to mytbs;
> create table test_p1 partition of test for values in (1);
>
> In the above test, after the setting the default_tablespace I am creating
> a partition table and expecting that to get created under "mytbs"
> tablespace.
>
> But that is not the case:
>
> postgres@66247=#select relname, reltablespace from pg_class where relname =
> 'test_p1';
>  relname | reltablespace
> ---------+---------------
>  test_p1 |             0
> (1 row)
>
> I noticed the behaviour change for default_tablespace with partition table
> with below commit.
>
> commit 87259588d0ab0b8e742e30596afa7ae25caadb18
> Author: Alvaro Herrera <[hidden email]>
> Date:   Thu Apr 25 10:20:23 2019 -0400
>
>     Fix tablespace inheritance for partitioned rels
>
>     Commit ca4103025dfe left a few loose ends.  The most important one
>     (broken pg_dump output) is already fixed by virtue of commit
>     3b23552ad8bb, but some things remained:
>
> I don't think that new behaviour is intended

Should we add this to open items?

Thanks,
Amit



Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Rushabh Lathia
In reply to this post by Amit Langote-2


On Fri, May 17, 2019 at 10:30 AM Amit Langote <[hidden email]> wrote:
Agree that this behavior change seems unintentional.

On 2019/05/17 12:40, Rushabh Lathia wrote:
> Looking at the commit changes, it seems like at condition when no other
> tablespace is specified, we default the tablespace to the parent partitioned
> table's.
>
>     else if (stmt->partbound)
>     {
>         /*
>          * For partitions, when no other tablespace is specified, we default
>          * the tablespace to the parent partitioned table's.
>          */
>         Assert(list_length(inheritOids) == 1);
>         tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
>     }
>
> But here it doesn't consider the default_tablespace if the parent
> partitioned
> tablespace is an InvalidOid (which was the care before this commit).
>
> PFA patch to fix the same.

+
+               if (!OidIsValid(tablespaceId))
+                       tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,
partitioned);
        }
        else
                tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,

Why not change it like this instead:

@@ -681,7 +681,8 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid
ownerId,
         Assert(list_length(inheritOids) == 1);
         tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
     }
-    else
+
+    if (!OidIsValid(tablespaceId))
         tablespaceId = GetDefaultTablespace(stmt->relation->relpersistence,
                                             partitioned);


Yes, sure we can do that. Here is the patch for the same.


--
Rushabh Lathia

default_tablespace_for_partition_v2.patch (1K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Amit Langote-2
On 2019/05/20 13:42, Rushabh Lathia wrote:

> On Fri, May 17, 2019 at 10:30 AM Amit Langote <[hidden email]>
>> Why not change it like this instead:
>>
>> @@ -681,7 +681,8 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid
>> ownerId,
>>          Assert(list_length(inheritOids) == 1);
>>          tablespaceId = get_rel_tablespace(linitial_oid(inheritOids));
>>      }
>> -    else
>> +
>> +    if (!OidIsValid(tablespaceId))
>>          tablespaceId =
>> GetDefaultTablespace(stmt->relation->relpersistence,
>>                                              partitioned);
>
> Yes, sure we can do that. Here is the patch for the same.

Thanks Rushabh.

Regards,
Amit



Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Alvaro Herrera-9
In reply to this post by Amit Langote-2
On 2019-May-20, Amit Langote wrote:

> Should we add this to open items?

Yeah.  I'm AFK this week, but can handle it afterwards.  The fix already
missed beta1, so I don't think there's a problem with taking a little
bit longer.

--
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services


Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Amit Langote-2
On 2019/05/21 11:29, Alvaro Herrera wrote:
> On 2019-May-20, Amit Langote wrote:
>
>> Should we add this to open items?
>
> Yeah.  I'm AFK this week, but can handle it afterwards.  The fix already
> missed beta1, so I don't think there's a problem with taking a little
> bit longer.

OK, added.

Thanks,
Amit



Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Alvaro Herrera-9
In reply to this post by Alvaro Herrera-9
On 2019-May-20, Alvaro Herrera wrote:

> On 2019-May-20, Amit Langote wrote:
>
> > Should we add this to open items?
>
> Yeah.  I'm AFK this week, but can handle it afterwards.  The fix already
> missed beta1, so I don't think there's a problem with taking a little
> bit longer.

Here's my proposed patch.  I changed/expanded the tests a little bit to
ensure more complete coverage.

--
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

tablespace-partitioned.patch (6K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Alvaro Herrera-9
On 2019-Jun-06, Alvaro Herrera wrote:

> Here's my proposed patch.  I changed/expanded the tests a little bit to
> ensure more complete coverage.

Well, revise the comments a little bit.

--
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

comments.patch (956 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Amit Langote
Hi Alvaro,

On Fri, Jun 7, 2019 at 8:02 AM Alvaro Herrera <[hidden email]> wrote:
> On 2019-Jun-06, Alvaro Herrera wrote:
>
> > Here's my proposed patch.  I changed/expanded the tests a little bit to
> > ensure more complete coverage.
>
> Well, revise the comments a little bit.

Thanks for adding the tests.  Looks good.

Regards,
Amit


Reply | Threaded
Open this post in threaded view
|

Re: behaviour change - default_tablesapce + partition table

Alvaro Herrera-9
Hi Amit, Rushabh,

On 2019-Jun-07, Amit Langote wrote:

> Thanks for adding the tests.  Looks good.

Thanks Amit, pushed now.  Marking open item as done.

--
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services