Обсуждение: Segmentation fault: pg_upgrade 9.1 to 9.3: pg_dump: row number 0 is out of range 0..-1

On Sat, Sep 14, 2013 at 09:40:01PM -0400, Robert Nix wrote:
> Running a pg_upgrade task is causing Segmentation fault:
>
> command: "/usr/lib/postgresql/9.3/bin/pg_dump" --host "/var/lib/postgresql"
> --port 50432 --username "postgres" --schema-only --quote-all-identifiers
> --binary-upgrade --format=custom  --file="pg_upgrade_dump_6064585.custom" "u"
> >> "pg_upgrade_dump_6064585.log" 2>&1
> pg_dump: row number 0 is out of range 0..-1
> Segmentation fault (core dumped)

This error is coming from libpq, specifically
check_tuple_field_number(), which is called by PQgetvalue(),
PQgetlength(), and PQgetisnull().  As I have never seen this error
before, there might be something wrong with the system catalogs on that
cluster.  It would be good to run pg_dump --binary-upgrade on that
database and try to identify the object being dumped that is causing the
crash.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + Everyone has their own god. +

I ran into something tonight that seems relevant here, or certainly related:

I recently updated my app(s) libpq version from 9.1 to 9.3 and immediately I starting seeing:

row number 0 is out of range 0..-1

spewed to stdout.

I traced it down to this code:

    if (PQresultStatus(result) == PGRES_TUPLES_OK){
        if (!PQgetisnull(result, 0, 0)){
            ......
        }
    }

if the result contains 0 tuples, the message is spit to stdout.  This is new behavior, I went back and rebuilt using
libversions (according to PQlibVersion) 90102, 90203, and 90300.  Neither version 90102 & 90203 spewed to stdout. 

I don't know if this is causing the OP's crash, but it seems to me that there is something going on in libpq that needs
tobe looked at.  If nothing else than to get rid of the spewing to stdout. 

alan


On Oct 10, 2013, at 9:47 AM, Bruce Momjian <bruce@momjian.us> wrote:

> On Sat, Sep 14, 2013 at 09:40:01PM -0400, Robert Nix wrote:
>> Running a pg_upgrade task is causing Segmentation fault:
>>
>> command: "/usr/lib/postgresql/9.3/bin/pg_dump" --host "/var/lib/postgresql"
>> --port 50432 --username "postgres" --schema-only --quote-all-identifiers
>> --binary-upgrade --format=custom  --file="pg_upgrade_dump_6064585.custom" "u"
>>>> "pg_upgrade_dump_6064585.log" 2>&1
>> pg_dump: row number 0 is out of range 0..-1
>> Segmentation fault (core dumped)
>
> This error is coming from libpq, specifically
> check_tuple_field_number(), which is called by PQgetvalue(),
> PQgetlength(), and PQgetisnull().  As I have never seen this error
> before, there might be something wrong with the system catalogs on that
> cluster.  It would be good to run pg_dump --binary-upgrade on that
> database and try to identify the object being dumped that is causing the
> crash.
>
> --
>  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
>  EnterpriseDB                             http://enterprisedb.com
>
>  + Everyone has their own god. +
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general

Alan Nilsson <anilsson@apple.com> writes:
> I ran into something tonight that seems relevant here, or certainly related:
> I recently updated my app(s) libpq version from 9.1 to 9.3 and immediately I starting seeing:

> row number 0 is out of range 0..-1

> spewed to stdout.

> I traced it down to this code:

>     if (PQresultStatus(result) == PGRES_TUPLES_OK){
>         if (!PQgetisnull(result, 0, 0)){
>             ......
>         }
>     }

I can find no code like that in the PG sources --- everyplace that does
"PQgetisnull(result, 0, 0)" prefaces it with a check that there's at least
one row in the result.  I think you're blaming the messenger.

            regards, tom lane

oops sorry to be unclear, the code below is my code in my app.

What I am saying is that something changed in 90300 that causes libpq to spew to stdout where it had not in libpq 90102
&90203. 

I guess i am blaming the messenger because there should be no messenger.  Regardless of how badly I mangle the use of
libpq,it should not be sending anything to stdout. 

alan

On Oct 26, 2013, at 6:53 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:

> Alan Nilsson <anilsson@apple.com> writes:
>> I ran into something tonight that seems relevant here, or certainly related:
>> I recently updated my app(s) libpq version from 9.1 to 9.3 and immediately I starting seeing:
>
>> row number 0 is out of range 0..-1
>
>> spewed to stdout.
>
>> I traced it down to this code:
>
>>     if (PQresultStatus(result) == PGRES_TUPLES_OK){
>>         if (!PQgetisnull(result, 0, 0)){
>>             ......
>>         }
>>     }
>
> I can find no code like that in the PG sources --- everyplace that does
> "PQgetisnull(result, 0, 0)" prefaces it with a check that there's at least
> one row in the result.  I think you're blaming the messenger.
>
>             regards, tom lane
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general

Alan Nilsson <anilsson@apple.com> writes:
> What I am saying is that something changed in 90300 that causes libpq to spew to stdout where it had not in libpq
90102& 90203. 

Well, that's interesting, but the issue is not in libpq, and you've still
provided no information that would help anyone diagnose where it is
(like, say, the query that's producing different results).

> I guess i am blaming the messenger because there should be no messenger.  Regardless of how badly I mangle the use of
libpq,it should not be sending anything to stdout. 

Would you prefer it dumped core?  It doesn't have very many choices for
reporting that you've passed invalid arguments.  In any case, the short
answer is that if you'd like this code fragment to do X when the query
has returned zero rows, you should explicitly code a test for that.
Otherwise you're at the mercy of somebody else's idea of what to do
with the error condition.

Also, if your gripe is not so much that it resorted to printing a message
as exactly how it printed it, see libpq's options for inserting your own
message processing hooks.

            regards, tom lane