Thread: BUG #4296: Server crashes by restoring database

BUG #4296: Server crashes by restoring database

From
"Michal Szelag"
Date:
The following bug has been logged online:

Bug reference:      4296
Logged by:          Michal Szelag
Email address:      msz@szel.pl
PostgreSQL version: 8.3.3
Operating system:   2.6.23-1-amd64[debian etch]
Description:        Server crashes by restoring database
Details:

Hallo,

very sory for my bad english:

i have a server crash by retoring database.

I try to retore my database from a custom package with pg_restore and prom a
plain package with psql. In both i have the same problem, psl and pg_restore
outputs:
pg_restore: [archiver (db)] error returned by PQputCopyData: server closed
the connection unexpectedly
and i have messages in my log file:
server process (PID 11351) exited with exit code 255
terminating any other active server processes
all server processes terminated; reinitializing

my database contains composite types and array composite types.

Can you tell me, haw can i restore my database?

Best Regerds, M.Sz.

Re: BUG #4296: Server crashes by restoring database

From
Tom Lane
Date:
"Michal Szelag" <msz@szel.pl> writes:
> i have a server crash by retoring database.

> I try to retore my database from a custom package with pg_restore and prom a
> plain package with psql. In both i have the same problem, psl and pg_restore
> outputs:
> pg_restore: [archiver (db)] error returned by PQputCopyData: server closed
> the connection unexpectedly

You're evidently tripping over some kind of bug, but we can't help you
without a lot more detail than this.  Try to identify the exact command
that's crashing (hint: turn on log_statements and watch the postmaster
log).  See if you can put together a self-contained script that causes
the crash.  That will let us fix the problem, and it might let you find
a workaround to reload your data even in advance of a real fix.

            regards, tom lane

Re: BUG #4296: Server crashes by restoring database

From
Michał Szeląg
Date:
Dnia piątek, 11 lipca 2008, Tom Lane napisał:
> "Michal Szelag" <msz@szel.pl> writes:
> > i have a server crash by retoring database.
> >
> > I try to retore my database from a custom package with pg_restore and
> > prom a plain package with psql. In both i have the same problem, psl and
> > pg_restore outputs:
> > pg_restore: [archiver (db)] error returned by PQputCopyData: server
> > closed the connection unexpectedly
>
> You're evidently tripping over some kind of bug, but we can't help you
> without a lot more detail than this.  Try to identify the exact command
> that's crashing (hint: turn on log_statements and watch the postmaster
> log).  See if you can put together a self-contained script that causes
> the crash.  That will let us fix the problem, and it might let you find
> a workaround to reload your data even in advance of a real fix.
>
>             regards, tom lane

Hallo,
the crash occurs by this command in my dump:

COPY users_history (users_history_id, users_id, uh_date, uh_php_session_id,
uh_gpsc, uh_ip, uh_browser, uh_referer, uh_url) FROM stdin;

so, i excluded this table from my dump and the same error occurs by this
command:
ALTER TABLE ONLY invoices_lines ADD CONSTRAINT invoices_lines_pkey PRIMARY KEY
(nagid, o, linid);

The plain dump of my database has size of 4GB.

I have divided my dump with head, tail, grep, dd, etc. and i have restored my
database(a have spend a lot of time by this).

Best regerds, Michał Szeląg.
--
==================================
Michał Szeląg
tel.: +48 693 11 37 55


Re: BUG #4296: Server crashes by restoring database

From
Tom Lane
Date:
Michał Szeląg <msz@szel.pl> writes:
> the crash occurs by this command in my dump:

> COPY users_history (users_history_id, users_id, uh_date, uh_php_session_id,
> uh_gpsc, uh_ip, uh_browser, uh_referer, uh_url) FROM stdin;

> so, i excluded this table from my dump and the same error occurs by this
> command:
> ALTER TABLE ONLY invoices_lines ADD CONSTRAINT invoices_lines_pkey PRIMARY KEY
> (nagid, o, linid);

You still haven't given anywhere near enough information to identify the
bug.  I wonder though if any of these tables involve non-built-in
datatypes?  An out-of-date .so file for an add-on datatype could easily
lead to crashes in these operations.
        regards, tom lane


Re: BUG #4296: Server crashes by restoring database

From
Michał Szeląg
Date:
Dnia poniedziałek, 14 lipca 2008, Tom Lane napisał:
> Michał Szeląg <msz@szel.pl> writes:
> > the crash occurs by this command in my dump:
> >
> > COPY users_history (users_history_id, users_id, uh_date,
> > uh_php_session_id, uh_gpsc, uh_ip, uh_browser, uh_referer, uh_url) FROM
> > stdin;
> >
> > so, i excluded this table from my dump and the same error occurs by this
> > command:
> > ALTER TABLE ONLY invoices_lines ADD CONSTRAINT invoices_lines_pkey
> > PRIMARY KEY (nagid, o, linid);
>
> You still haven't given anywhere near enough information to identify the
> bug.  I wonder though if any of these tables involve non-built-in
> datatypes?  An out-of-date .so file for an add-on datatype could easily
> lead to crashes in these operations.
>
>             regards, tom lane

I have composite types in my dump, there are some arrays of them too. There
are no over datatypes.


earlier i had messages by restoring this database like this:

Consider increasing the configuration parameter "checkpoint_segments".
checkpoints are occurring too frequently (17 seconds apart).

So i have increased this parameter and shared memory, but the crash stays.

Best regards, M.Sz.

--
==================================
Michał Szeląg
tel.: +48 693 11 37 55


Re: BUG #4296: Server crashes by restoring database

From
Tom Lane
Date:
Michał Szeląg <msz@szel.pl> writes:
> Dnia poniedzia�ek, 14 lipca 2008, Tom Lane napisa�:
>> You still haven't given anywhere near enough information to identify the
>> bug.  I wonder though if any of these tables involve non-built-in
>> datatypes?  An out-of-date .so file for an add-on datatype could easily
>> lead to crashes in these operations.

> I have composite types in my dump, there are some arrays of them too. There
> are no over datatypes.

We're not going to be able to do much about this bug report unless you
can provide a test case.  There's just not enough information here to
even begin looking for a cause.
        regards, tom lane