Обсуждение: SIGSEGV in sebselect.

Hi All,

I'm getting a SIGSEGV with the following subselect:-

select * from disks where diskid in
 ( select distinct diskid from tracks where artist = 'Led Zeppelin');

The source tree is a cvsup from 22nd Feb (AM GMT)
Platform is Linux/SPARC

I'm afraid my debugging skills are limited to generating a backtrace
so here goes....

Keith.

Program received signal SIGSEGV, Segmentation fault.
0xe0104ee4 in strlen ()
(gdb) bt
#0  0xe0104ee4 in strlen ()
#1  0xd1264 in pstrdup (string=0x0) at palloc.c:115
#2  0x696c0 in _copyUnique (from=0x1b0b90) at copyfuncs.c:552
#3  0x6abdc in copyObject (from=0x1b0b90) at copyfuncs.c:1687
#4  0x69754 in _copySubPlan (from=0x1a83d0) at copyfuncs.c:594
#5  0x6abfc in copyObject (from=0x1a83d0) at copyfuncs.c:1693
#6  0x698b0 in _copyExpr (from=0x1b1f90) at copyfuncs.c:679
#7  0x6ac2c in copyObject (from=0x1b1f90) at copyfuncs.c:1706
#8  0x6a304 in _copyCInfo (from=0x1b1710) at copyfuncs.c:1334
#9  0x6ad5c in copyObject (from=0x1b1710) at copyfuncs.c:1767
#10 0x6ae54 in copyObject (from=0x1b2d10) at copyfuncs.c:1829
#11 0x7e8b8 in create_seqscan_path (rel=0x1b0a90) at pathnode.c:228
#12 0x737c4 in find_rel_paths (root=0x15d390, rels=0x1b2f70) at allpaths.c:118
#13 0x73778 in find_paths (root=0x15d390, rels=0x1b2f70) at allpaths.c:72
#14 0x78ae4 in subplanner (root=0x15d390, flat_tlist=0x1ae330, qual=0x1ae3f0) at
planmain.c:271
#15 0x789fc in query_planner (root=0x15d390, command_type=1, tlist=0x1aa910,
qual=0x1ae3f0) at planmain.c:181
#16 0x79040 in union_planner (parse=0x15d390) at planner.c:138
#17 0x78f2c in planner (parse=0x15d390) at planner.c:72
#18 0xa80f4 in pg_parse_and_plan (
    query_string=0xefffdc38 "select * from disks where diskid in ( select
distinct diskid from tracks where artist = 'Led Zeppelin');", typev=0x0,
nargs=0, queryListP=0xefffdb64, dest=Remote) at postgres.c:534
#19 0xa81e0 in pg_exec_query_dest (
    query_string=0xefffdc38 "select * from disks where diskid in ( select
distinct diskid from tracks where artist = 'Led Zeppelin');", argv=0x0,
typev=0x0, nargs=0, dest=Remote) at postgres.c:615
#20 0xa81bc in pg_exec_query (
    query_string=0xefffdc38 "select * from disks where diskid in ( select
distinct diskid from tracks where artist = 'Led Zeppelin');", argv=0x0,
typev=0x0, nargs=0) at postgres.c:597
#21 0xa90c4 in PostgresMain (argc=1122304, argv=0xeffffc38) at postgres.c:1373
#22 0x68788 in main (argc=11, argv=0xeffffd04) at main.c:97

This is a bug.  I can reproduct it here with:

    select * from pguser where usesysid in (select distinct usesysid
from pg_user);

The Unique node copy has a NULL uniqueAttr value.

Vadim is this related to subselect distinct handling, or is there some
larger bug here?

>
> Hi All,
>
> I'm getting a SIGSEGV with the following subselect:-
>
> select * from disks where diskid in
>  ( select distinct diskid from tracks where artist = 'Led Zeppelin');
>
> The source tree is a cvsup from 22nd Feb (AM GMT)
> Platform is Linux/SPARC
>
> I'm afraid my debugging skills are limited to generating a backtrace
> so here goes....
>
> Keith.
>
> Program received signal SIGSEGV, Segmentation fault.
> 0xe0104ee4 in strlen ()
> (gdb) bt
> #0  0xe0104ee4 in strlen ()
> #1  0xd1264 in pstrdup (string=0x0) at palloc.c:115
> #2  0x696c0 in _copyUnique (from=0x1b0b90) at copyfuncs.c:552
> #3  0x6abdc in copyObject (from=0x1b0b90) at copyfuncs.c:1687
> #4  0x69754 in _copySubPlan (from=0x1a83d0) at copyfuncs.c:594
> #5  0x6abfc in copyObject (from=0x1a83d0) at copyfuncs.c:1693
> #6  0x698b0 in _copyExpr (from=0x1b1f90) at copyfuncs.c:679
> #7  0x6ac2c in copyObject (from=0x1b1f90) at copyfuncs.c:1706
> #8  0x6a304 in _copyCInfo (from=0x1b1710) at copyfuncs.c:1334
> #9  0x6ad5c in copyObject (from=0x1b1710) at copyfuncs.c:1767
> #10 0x6ae54 in copyObject (from=0x1b2d10) at copyfuncs.c:1829
> #11 0x7e8b8 in create_seqscan_path (rel=0x1b0a90) at pathnode.c:228
> #12 0x737c4 in find_rel_paths (root=0x15d390, rels=0x1b2f70) at allpaths.c:118
> #13 0x73778 in find_paths (root=0x15d390, rels=0x1b2f70) at allpaths.c:72
> #14 0x78ae4 in subplanner (root=0x15d390, flat_tlist=0x1ae330, qual=0x1ae3f0) at
> planmain.c:271
> #15 0x789fc in query_planner (root=0x15d390, command_type=1, tlist=0x1aa910,
> qual=0x1ae3f0) at planmain.c:181
> #16 0x79040 in union_planner (parse=0x15d390) at planner.c:138
> #17 0x78f2c in planner (parse=0x15d390) at planner.c:72
> #18 0xa80f4 in pg_parse_and_plan (
>     query_string=0xefffdc38 "select * from disks where diskid in ( select
> distinct diskid from tracks where artist = 'Led Zeppelin');", typev=0x0,
> nargs=0, queryListP=0xefffdb64, dest=Remote) at postgres.c:534
> #19 0xa81e0 in pg_exec_query_dest (
>     query_string=0xefffdc38 "select * from disks where diskid in ( select
> distinct diskid from tracks where artist = 'Led Zeppelin');", argv=0x0,
> typev=0x0, nargs=0, dest=Remote) at postgres.c:615
> #20 0xa81bc in pg_exec_query (
>     query_string=0xefffdc38 "select * from disks where diskid in ( select
> distinct diskid from tracks where artist = 'Led Zeppelin');", argv=0x0,
> typev=0x0, nargs=0) at postgres.c:597
> #21 0xa90c4 in PostgresMain (argc=1122304, argv=0xeffffc38) at postgres.c:1373
> #22 0x68788 in main (argc=11, argv=0xeffffd04) at main.c:97
>
>
>


--
Bruce Momjian
maillist@candle.pha.pa.us

>
> Hi All,
>
> I'm getting a SIGSEGV with the following subselect:-
>
> select * from disks where diskid in
>  ( select distinct diskid from tracks where artist = 'Led Zeppelin');
>
> The source tree is a cvsup from 22nd Feb (AM GMT)
> Platform is Linux/SPARC
>
> I'm afraid my debugging skills are limited to generating a backtrace
> so here goes....

This is a partial fix.  In make_unique(), if Query->uniqueAttr is '*',
Unique->uniqueAttr gets null.  However, with this fix, I get:

    test=> select usesysid from pg_user where usesysid in (select distinct
    usesysid from pg_user);
    ERROR:  ExecReScan: node type 24 not supported

I have not installed this patch yet until Vadim comments and perhaps has
an idea on the ExecReScan message:


---------------------------------------------------------------------------

*** ./backend/nodes/copyfuncs.c.orig    Sun Feb 22 18:58:59 1998
--- ./backend/nodes/copyfuncs.c    Sun Feb 22 19:03:37 1998
***************
*** 549,555 ****
       *    copy remainder of node
       * ----------------
       */
!     newnode->uniqueAttr = pstrdup(from->uniqueAttr);
      newnode->uniqueAttrNum = from->uniqueAttrNum;
      Node_Copy(from, newnode, uniquestate);

--- 549,558 ----
       *    copy remainder of node
       * ----------------
       */
!     if (newnode->uniqueAttr)
!         newnode->uniqueAttr = pstrdup(from->uniqueAttr);
!     else
!         newnode->uniqueAttr = NULL;
      newnode->uniqueAttrNum = from->uniqueAttrNum;
      Node_Copy(from, newnode, uniquestate);



--
Bruce Momjian
maillist@candle.pha.pa.us

Bruce Momjian wrote:
>
> This is a partial fix.  In make_unique(), if Query->uniqueAttr is '*',
> Unique->uniqueAttr gets null.  However, with this fix, I get:
>
>         test=> select usesysid from pg_user where usesysid in (select distinct
>         usesysid from pg_user);
>         ERROR:  ExecReScan: node type 24 not supported
>
> I have not installed this patch yet until Vadim comments and perhaps has
> an idea on the ExecReScan message:

Patch is Ok. DISTINCT is yet another node without re-scan support.
Hope to add it in 1-2 days.

Vadim

>
> Bruce Momjian wrote:
> >
> > This is a partial fix.  In make_unique(), if Query->uniqueAttr is '*',
> > Unique->uniqueAttr gets null.  However, with this fix, I get:
> >
> >         test=> select usesysid from pg_user where usesysid in (select distinct
> >         usesysid from pg_user);
> >         ERROR:  ExecReScan: node type 24 not supported
> >
> > I have not installed this patch yet until Vadim comments and perhaps has
> > an idea on the ExecReScan message:
>
> Patch is Ok. DISTINCT is yet another node without re-scan support.
> Hope to add it in 1-2 days.

I have applied my patch.

--
Bruce Momjian
maillist@candle.pha.pa.us