Обсуждение: Error: could not find a feasible split point for "obs_v_file_id_index"

Hey All,

While doing some daily processing, I ran across the following error:

ERROR:  could not find a feasible split point for "obs_v_file_id_index"
STATEMENT: UPDATE obs_v SET mag = obs_v.imag + zp.zero_v + cg.color_v *
(obs_v.imag - i.imag), use = true FROM color_groups AS cg, zero_pair AS zp,
obs_i AS i, files AS f, groups AS g WHERE obs_v.star_id = i.star_id AND
obs_v.file_id = f.file_id AND cg.group_id = g.group_id AND g.night_id =
f.night_id AND g.group_id = 149 AND zp.pair_id = f.pair_id

version: 7.4.1.

table schema:

---------+---------+------------------------------------------------
 x       | real    | not null
 y       | real    | not null
 imag    | real    | not null
 smag    | real    | not null
 loc     | spoint  | not null
 obs_id  | integer | not null default nextval('"obs_id_seq"'::text)
 file_id | integer | not null
 use     | boolean | default false
 solve   | boolean | default false
 star_id | integer |
 mag     | real    |
Inexes:    "obs_v_file_id_index" btree (file_id)
    "obs_v_loc_index" gist (loc)
    "obs_v_obs_id_index" btree (obs_id)
    "obs_v_star_id_index" btree (star_id)
    "obs_v_use_index" btree (use)
Foreign-key constraints:
    "obs_v_files_constraint" FOREIGN KEY (file_id) REFERENCES files(file_id) ON
DELETE CASCADE
    "obs_v_star_id_constraint" FOREIGN KEY (star_id) REFERENCES
catalog(star_id) ON DELETE SET NULL
Triggers:
    obs_v_trig BEFORE INSERT OR DELETE OR UPDATE ON obs_v FOR EACH ROW EXECUTE
PROCEDURE observations_trigger

Help?

Cheers,
Rob

--
 06:52:56 up 15 days, 17:15,  3 users,  load average: 2.05, 2.12, 2.22
Linux 2.6.5-02 #8 SMP Mon Jul 12 21:34:44 MDT 2004

Robert Creager <Robert_Creager@LogicalChaos.org> writes:
> While doing some daily processing, I ran across the following error:

> ERROR:  could not find a feasible split point for "obs_v_file_id_index"

>  file_id | integer | not null
> Inexes:    "obs_v_file_id_index" btree (file_id)

On an index on an *integer* column??  That's a can't-happen error in the
first place, and it's particularly silly when the index entries are
small and fixed-width.  I surmise that you have a corrupted index, but
it would be interesting to examine the entrails.  Is it repeatable?
Is the database small enough that you could ship me a tarball copy
(or if not, would you be willing to let me into your system to look at
it)?

            regards, tom lane

When grilled further on (Mon, 30 Aug 2004 11:31:12 -0400),
Tom Lane <tgl@sss.pgh.pa.us> confessed:

> Robert Creager <Robert_Creager@LogicalChaos.org> writes:
> > While doing some daily processing, I ran across the following error:
>
> > ERROR:  could not find a feasible split point for "obs_v_file_id_index"
>
> >  file_id | integer | not null
> > Inexes:    "obs_v_file_id_index" btree (file_id)
>
> On an index on an *integer* column??  That's a can't-happen error in the
> first place, and it's particularly silly when the index entries are
> small and fixed-width.  I surmise that you have a corrupted index, but
> it would be interesting to examine the entrails.  Is it repeatable?
> Is the database small enough that you could ship me a tarball copy
> (or if not, would you be willing to let me into your system to look at
> it)?
>
>             regards, tom lane

The error is repeatable (with a script that runs in a few minutes).  A recent
dump 528Mb compressed (10Gb on disk).  As for access, presumably you want in as
postgres?

Any chance you have a gpg key somewhere for encryption?  I did a quick check and
didn't find any...

Cheers,
Rob

--
 20:00:25 up 16 days,  6:22,  3 users,  load average: 3.79, 3.97, 3.99
Linux 2.6.5-02 #8 SMP Mon Jul 12 21:34:44 MDT 2004