postgres-xc-developers Mailing List for Postgres-XC

Hi all,

Related to this old bug, here is a summary of the ideas to make sequence
creation and deletion consistent in the cluster.
During sequence create, activate a hook that will drop sequence if an error
occurred in transaction while processing.
During sequence drop, activate a hook to effectively drop the sequence at
the transaction commit instead of dropping it at the statement step.

With such a solution, the same hook could be used to maintain sequence
consistency at gtm level, and error handling of drop sequence would be easy.
In there already in Postgres code some ways to activate hook mechanisms at
transaction abort or commit?
We may also face a certain challenge with 2PC transactions. But I was
thinking of launching the hook at the prepare step.

Regards,
-- 
Michael Paquier
http://michael.otacoo.com

Hi all,

I am having a look at bug 3141171, with such a test case:
create table a(a int);
create table b(b int) inherits(a);
insert into a values(1);
insert into b values(2,2);
select a.* from a,pg_class where a.tableoid=pg_class.oid;
postgres=# explain verbose select a.* from a,pg_class where
a.tableoid=pg_class.oid;
                                  QUERY PLAN

-------------------------------------------------------------------------------
 Hash Join  (cost=0.03..12.39 rows=2 width=4)
   Output: public.a.a
   Hash Cond: (pg_class.oid = public.a.tableoid)
   ->  Seq Scan on pg_catalog.pg_class  (cost=0.00..10.90 rows=290 width=4)
         Output: pg_class.oid
   ->  Hash  (cost=0.00..0.00 rows=2000 width=8)
         Output: public.a.a, public.a.tableoid
         ->  Append  (cost=0.00..0.00 rows=2000 width=8)
               ->  Data Node Scan on a  (cost=0.00..0.00 rows=1000 width=8)
                     Output: public.a.a, public.a.tableoid
                     Node/s: dn1, dn2
                     Remote query: SELECT a, tableoid FROM ONLY a WHERE true
               ->  Data Node Scan on a  (cost=0.00..0.00 rows=1000 width=8)
                     Output: public.a.a, public.a.tableoid
                     Node/s: dn1, dn2
                     Remote query: SELECT a, tableoid FROM ONLY b a WHERE
true
(16 rows)
In this case we fetch from the datanodes the table oid, which is not
consistent among nodes.
So, even if we should output 2 tuples as a result, nothing is obtained.

I have been thinking about 2 methods to fix that,but globally I think we
should apply oid conditions as coordinator quals.
1) Replace conditions on oid by a condition on the node name casted on
regclass.
postgres=# explain verbose select a.* from a,pg_class where
'a'::regclass=pg_class.oid;

                                 QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Nested Loop  (cost=0.00..8.29 rows=2 width=4)
   Output: public.a.a
   ->  Index Scan using pg_class_oid_index on pg_catalog.pg_class
 (cost=0.00..8.27 rows=1 width=0)
         Output: pg_class.relname, pg_class.relnamespace, pg_class.reltype,
pg_class.reloftype, pg_class.relowner, pg_class.relam,
pg_class.relfilenode, pg_class.reltab
lespace, pg_class.relpages, pg_class.reltuples, pg_class.reltoastrelid,
pg_class.reltoastidxid, pg_class.relhasindex, pg_class.relisshared,
pg_class.relpersistence, pg_
class.relkind, pg_class.relnatts, pg_class.relchecks, pg_class.relhasoids,
pg_class.relhaspkey, pg_class.relhasrules, pg_class.relhastriggers,
pg_class.relhassubclass,
pg_class.relfrozenxid, pg_class.relacl, pg_class.reloptions
         Index Cond: (16386::oid = pg_class.oid)
   ->  Append  (cost=0.00..0.00 rows=2000 width=4)
         ->  Data Node Scan on a  (cost=0.00..0.00 rows=1000 width=4)
               Output: public.a.a
               Node/s: dn1, dn2
               Remote query: SELECT a FROM ONLY a WHERE true
         ->  Data Node Scan on a  (cost=0.00..0.00 rows=1000 width=4)
               Output: public.a.a
               Node/s: dn1, dn2
               Remote query: SELECT a FROM ONLY b a WHERE true

2) Scan the quals of query and add them as coordinator quals directly at
planning phase.
I may need to touch a bit FQS to do that, but looks cleaner.

Suggestions?
-- 
Michael Paquier
http://michael.otacoo.com

I haven't seen any additional failures and it looks clean.
Please go ahead.

On Wed, Mar 14, 2012 at 11:34 AM, Ashutosh Bapat <
ash...@en...> wrote:

> Hi All,
> PFA the patch for support for bound parameters and cursor options in FQS
> planner.
>
> The regression does not show new failure.
>
> --
> Best Wishes,
> Ashutosh Bapat
> EntepriseDB Corporation
> The Enterprise Postgres Company
>
>
>
> ------------------------------------------------------------------------------
> Virtualization & Cloud Management Using Capacity Planning
> Cloud computing makes use of virtualization - but cloud computing
> also focuses on allowing computing to be delivered as a service.
> http://www.accelacomm.com/jaw/sfnl/114/51521223/
> _______________________________________________
> Postgres-xc-developers mailing list
> Pos...@li...
> https://lists.sourceforge.net/lists/listinfo/postgres-xc-developers
>
>


-- 
Michael Paquier
http://michael.otacoo.com

Hi All,
PFA the patch for support for bound parameters and cursor options in FQS
planner.

The regression does not show new failure.

-- 
Best Wishes,
Ashutosh Bapat
EntepriseDB Corporation
The Enterprise Postgres Company