double opens

[Charset iso-8859-1 unsupported, filtering to ASCII...]

Hi all,

There is a TODO item
* Overhaul mdmgr/smgr to fix double unlinking and double opens, cleanup

In Windows-NT,we could see the following error reported by
yutaka tanida [yutaka@marin.or.jp].

version
------------------------------------------------------------
PostgreSQL 6.5.1 on i686-pc-cygwin, compiled by gcc gcc-2.95
(1 row)

template1=> create table table1 ( i int,j int);
CREATE
template1=> create view view1 as select * from table1;
CREATE
template1=> drop view view1;
DROP
template1=> create view view1 as select * from table1;
ERROR: cannot create view1

"drop view" couldn't unlink the base file of target view because
it is doubly opened and so "create view" coundn't create the view.

After applying the following patch on trial,"drop view" was able to
unlink the base file and "create view" was able to create the view
again.

I think base files should be closed at the time of cache invalidation.
RelationFlushRelation() invalidates the entry of relation cache but
doesn't close the base file of target relation.
Is there any reason ?

Or why doesn't RelationCacheDelete() close the base file of
target relation ?

Regards.

Hiroshi Inoue
Inoue@tpf.co.jp

*** utils/cache/relcache.c.orig	Wed May 26 16:05:38 1999
--- utils/cache/relcache.c	Wed Jul 28 13:23:49 1999
***************
*** 1282,1287 ****
--- 1282,1288 ----
oldcxt = MemoryContextSwitchTo((MemoryContext) CacheCxt);

RelationCacheDelete(relation);
+ smgrclose(DEFAULT_SMGR, relation);

FreeTupleDesc(relation->rd_att);
SystemCacheRelationFlushed(RelationGetRelid(relation));

Basically, I thought the close was done already in the drop table code.
Is it strange to do the close inside the cache? The cache does the
opens, right?

-- 
  Bruce Momjian                        |  http://www.op.net/~candle
  maillist@candle.pha.pa.us            |  (610) 853-3000
  +  If your life is a hard drive,     |  830 Blythe Avenue
  +  Christ can be your backup.        |  Drexel Hill, Pennsylvania 19026

[Charset iso-8859-1 unsupported, filtering to ASCII...]

Hi all,

There is a TODO item
* Overhaul mdmgr/smgr to fix double unlinking and double opens, cleanup

In Windows-NT,we could see the following error reported by
yutaka tanida [yutaka@marin.or.jp].

version
------------------------------------------------------------
PostgreSQL 6.5.1 on i686-pc-cygwin, compiled by gcc gcc-2.95
(1 row)

template1=> create table table1 ( i int,j int);
CREATE
template1=> create view view1 as select * from table1;
CREATE
template1=> drop view view1;
DROP
template1=> create view view1 as select * from table1;
ERROR: cannot create view1

"drop view" couldn't unlink the base file of target view because
it is doubly opened and so "create view" coundn't create the view.

[snip]

*** utils/cache/relcache.c.orig	Wed May 26 16:05:38 1999
--- utils/cache/relcache.c	Wed Jul 28 13:23:49 1999
***************
*** 1282,1287 ****
--- 1282,1288 ----
oldcxt = MemoryContextSwitchTo((MemoryContext) CacheCxt);

RelationCacheDelete(relation);
+ smgrclose(DEFAULT_SMGR, relation);

FreeTupleDesc(relation->rd_att);
SystemCacheRelationFlushed(RelationGetRelid(relation));

Basically, I thought the close was done already in the drop table code.
Is it strange to do the close inside the cache? The cache does the
opens, right?

No,relcache stuff doesn't do the opens.

First,my patch is not only for "drop view" case.
It's for cases such that
A backend registers an information to invalidate a relcache
entry and another backend removes the relcache entry trig-
gered by the information.

"drop view" plays both of the part alone and doubly opens
as follows.

RemoveView()
RemoveRewriteRule()
prs2_deleteFromRelation()
heap_open(relid of view) ---- opens a new file descriptor
for the base file of the view
setRelhasrulesInRelations()
heap_replace(tuple of "pg_class") ---- registers an informat-
ion to invalidate the view's
relcache entry.
heap_close() ---- doesn't close the file descriptor
CommandCounterIncrement()
.................
RelationFlushRelation() --- removes the relcache entry
of the view
heap_destroy_with_catalog()
heap_openr(viewName) --- opens another file descriptor
for the same view because
heap_openr() couldn't find the
relcache entry

Regards.

Hiroshi Inoue
Inoue@tpf.co.jp