Thread: Problem with updating system indices.

Problem with updating system indices.

From
Denis Perchine
Date:
Hello,

Sometime ago I did a patch to have large objects rollbackable.
That patch was against the current (on that moment) CVS.
It worked fine.

Now I tried to port the patch to 7.0.2 and have the following problem:
Somehow index which I made is not updated.
It can be figure out quite simple.
After creating new LO I have:

webmailstation=> explain select * from pg_largeobject where loid=14035719;
NOTICE:  QUERY PLAN:

Seq Scan on pg_largeobject  (cost=0.00..0.00 rows=1 width=20)
EXPLAIN
webmailstation=> select * from pg_largeobject where loid=14035719;  loid   | pageno | data
----------+--------+------14035719 |      0 |
(1 row)
webmailstation=>  set enable_seqscan=off;
SET VARIABLE
webmailstation=> explain select * from pg_largeobject where loid=14035719;
NOTICE:  QUERY PLAN:

Index Scan using pg_largeobject_loid_pn_index on pg_largeobject  (cost=0.00..5.01 rows=1 width=20)

EXPLAIN
webmailstation=>  select * from pg_largeobject where loid=14035719;loid | pageno | data
------+--------+------
(0 rows)

As far as you can see index does not contains such value.

The code which creates a LO looks like this:

Oid LargeobjectCreate(Oid loid) {       Oid             retval;       Relation                pg_largeobject;
HeapTuple      ntup = (HeapTuple) palloc(sizeof(HeapTupleData));       Relation
idescs[Num_pg_index_indices];      Datum           values[Natts_pg_largeobject];       char
nulls[Natts_pg_largeobject];      int             i;
 
       for (i=0; i<Natts_pg_largeobject; i++) {               nulls[i] = ' ';               values[i] = (Datum)NULL;
  }
 
       i = 0;       values[i++] = ObjectIdGetDatum(loid);       values[i++] = Int32GetDatum(0);       values[i++] =
(Datum)_byteain(NULL, 0);
 
       pg_largeobject = heap_openr(LargeobjectRelationName, RowExclusiveLock);       ntup =
heap_formtuple(pg_largeobject->rd_att,values, nulls);       retval = heap_insert(pg_largeobject, ntup);
 
       if (!IsIgnoringSystemIndexes()) {               CatalogOpenIndices(Num_pg_largeobject_indices, Name_pg_largeobje
             CatalogIndexInsert(idescs, Num_pg_largeobject_indices, pg_largeo
CatalogCloseIndices(Num_pg_largeobject_indices,idescs);       }
 
       heap_close(pg_largeobject, RowExclusiveLock);       heap_freetuple(ntup);
       CommandCounterIncrement();
       return retval;
}

And all go fine and indices should be updated... What I did wrong???
I traces through this code step by step and all seems fine...

-- 
Sincerely Yours,
Denis Perchine

----------------------------------
E-Mail: dyp@perchine.com
HomePage: http://www.perchine.com/dyp/
FidoNet: 2:5000/120.5
----------------------------------


Re: Problem with updating system indices.

From
Tom Lane
Date:
Denis Perchine <dyp@perchine.com> writes:
>         Relation                idescs[Num_pg_index_indices];                                            ^^^^^^^^

Dunno whether you have Num_pg_largeobject_indices =
Num_pg_index_indices, but in any case there's a latent
stack clobber here.

>         CommandCounterIncrement();

Are you sure it's a good idea to be incrementing the CC here?
This seems like a pretty low-level routine, so there might be
people further up the stack who do not want this done.
        regards, tom lane


Re: Problem with updating system indices.

From
Denis Perchine
Date:
> >         Relation                idescs[Num_pg_index_indices];
>                                              ^^^^^^^^
> 
> Dunno whether you have Num_pg_largeobject_indices =
> Num_pg_index_indices, but in any case there's a latent
> stack clobber here.

Ough... That's bad to work too much...
> >         CommandCounterIncrement();
> 
> Are you sure it's a good idea to be incrementing the CC here?
> This seems like a pretty low-level routine, so there might be
> people further up the stack who do not want this done.

Hmmm... Problem is that if I did not do this I do not see this change
later when I do lo_create/lo_open inside transaction...
What should I use else for this purpose?

-- 
Sincerely Yours,
Denis Perchine

----------------------------------
E-Mail: dyp@perchine.com
HomePage: http://www.perchine.com/dyp/
FidoNet: 2:5000/120.5
----------------------------------