share library version problems

On Tue, Aug 22, 2006 at 12:26:53PM -0400, Bryan White wrote:

I am having problems with my libpq programs crashing. This seems to be
a version incompatibility and I want to find out how to best proceed.

<snip>

My dev server was running the same setup as the web server. The
difference is that it acts as its own database server. I was
uncomfortable running an older version of the server on my test system
then on the live system. So yesterday I removed the OS supplied
PostgreSQL RPMs and installed 8.1.4 from RPMs on the PostgreSQL download
site.

Today I discovered that programs that I compile on my dev server will
segfault when run on the live web server.

Can you provide the output of ldd? The libraries are supposed to be
reasonably compatable.

In any case, you should try to run both servers against the same set of
libs and headers. You can have multiple copies of libpq around and
select it at compile time. The client library doesn't really have to
match the server version...

Hope this helps,
--
Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/

Bryan White <bryan@arcamax.com> writes:

I am having problems with my libpq programs crashing. This seems to be
a version incompatibility and I want to find out how to best proceed.

My main database is running Fedora Core 5 with the supplied PostgreSQL
8.1.4.

My web server is running Fedora Core 4 with the supplied PostgreSQL 8.0.8.

My dev server was running the same setup as the web server. The
difference is that it acts as its own database server. I was
uncomfortable running an older version of the server on my test system
then on the live system. So yesterday I removed the OS supplied
PostgreSQL RPMs and installed 8.1.4 from RPMs on the PostgreSQL download
site.

Today I discovered that programs that I compile on my dev server will
segfault when run on the live web server.

Can you get a core dump and provide a gdb backtrace from the segfault?
Right offhand I see no difference in the claimed API of 8.0 and 8.1
libpq except that 8.1 adds lo_create(), which I suppose you're not
using. So while this isn't good practice in general, I don't see
why it wouldn't work in this particular case.

One thing you should check is whether both libs were built with the same
options (compare pg_config --configure output from the 8.0 and 8.1
installations).

regards, tom lane

Tom Lane wrote:

Bryan White <bryan@arcamax.com> writes:

I am having problems with my libpq programs crashing. This seems to be
a version incompatibility and I want to find out how to best proceed.

My main database is running Fedora Core 5 with the supplied PostgreSQL
8.1.4.

My web server is running Fedora Core 4 with the supplied PostgreSQL 8.0.8.

My dev server was running the same setup as the web server. The
difference is that it acts as its own database server. I was
uncomfortable running an older version of the server on my test system
then on the live system. So yesterday I removed the OS supplied
PostgreSQL RPMs and installed 8.1.4 from RPMs on the PostgreSQL download
site.

Today I discovered that programs that I compile on my dev server will
segfault when run on the live web server.

Can you get a core dump and provide a gdb backtrace from the segfault?
Right offhand I see no difference in the claimed API of 8.0 and 8.1
libpq except that 8.1 adds lo_create(), which I suppose you're not
using. So while this isn't good practice in general, I don't see
why it wouldn't work in this particular case.

I have used both gdp and valgrind with full debug builds. The segfault
does not seem to occur in Postgres related code. It occurs before any
database connection is established. If it makes a difference, the code
is all written in C++.

The reason I suspect the Postgres lib is because there have been no
changes to this code. After upgrading the dev server yesterday, I
rolled out a small fix this morning and started seeing the segfault.
Reverting the change did not fix it. Compiling a clean subversion
checkout on both boxes confirmed that code compiled on the dev box will
not run on the web server but code compiled on the web server runs on
either box.

One thing you should check is whether both libs were built with the same
options (compare pg_config --configure output from the 8.0 and 8.1
installations).

I think that might be the problem. These are the differences in
pg_config --configure output:

dev server:
'--host=i686-redhat-linux-gnu'
'--build=i686-redhat-linux-gnu'
'--target=i686-redhat-linux'
'--with-includes=/usr/include'
'--with-libraries=/usr/lib'
'CFLAGS=-O2 -g -march=i686 -I/usr/include/et'
'CPPFLAGS= -I/usr/include/et'
'build_alias=i686-redhat-linux-gnu'
'host_alias=i686-redhat-linux-gnu'
'target_alias=i686-redhat-linux'

web server:
'--build=i386-redhat-linux'
'--host=i386-redhat-linux'
'--target=i386-redhat-linux-gnu'
'--with-tcl'
'--with-tclconfig=/usr/lib'
'--enable-thread-safety'
'CFLAGS=-O2 -g -pipe -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -m32
-march=i386 -mtune=pentium4 -fasynchronous-unwind-tables'
'build_alias=i386-redhat-linux'
'host_alias=i386-redhat-linux'
'target_alias=i386-redhat-linux-gnu'

I note that Postgres is packaged in the following RPMS:
postgresql-devel
postgresql-libs
postgresql-server
postgresql

Does it work to install the postgresql-server RPM from the 8.1 version
and the others from the Fedora 4 included 8.0 version?

--
Bryan White, ArcaMax Publishing Inc.

Martijn van Oosterhout wrote:

Can you provide the output of ldd? The libraries are supposed to be
reasonably compatable.

web server:
linux-gate.so.1 => (0x00ab0000)
libexpat.so.0 => /usr/lib/libexpat.so.0 (0x00411000)
libpq.so.4 => /usr/lib/libpq.so.4 (0x00324000)
libssl.so.5 => /lib/libssl.so.5 (0x005c2000)
libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x0069c000)
libm.so.6 => /lib/libm.so.6 (0x0049e000)
libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x00bd7000)
libc.so.6 => /lib/libc.so.6 (0x00d82000)
libpthread.so.0 => /lib/libpthread.so.0 (0x00ece000)
libcrypto.so.5 => /lib/libcrypto.so.5 (0x00c00000)
libkrb5.so.3 => /usr/lib/libkrb5.so.3 (0x00afc000)
libcrypt.so.1 => /lib/libcrypt.so.1 (0x00111000)
libresolv.so.2 => /lib/libresolv.so.2 (0x00f0c000)
libnsl.so.1 => /lib/libnsl.so.1 (0x00f88000)
libgssapi_krb5.so.2 => /usr/lib/libgssapi_krb5.so.2 (0x00b70000)
libcom_err.so.2 => /lib/libcom_err.so.2 (0x00a9f000)
libk5crypto.so.3 => /usr/lib/libk5crypto.so.3 (0x00ad6000)
libdl.so.2 => /lib/libdl.so.2 (0x0013f000)
libz.so.1 => /usr/lib/libz.so.1 (0x003fc000)
/lib/ld-linux.so.2 (0x00267000)
libkrb5support.so.0 => /usr/lib/libkrb5support.so.0 (0x00ac3000)

dev server:
linux-gate.so.1 => (0x00497000)
libexpat.so.0 => /usr/lib/libexpat.so.0 (0x0065b000)
libpq.so.4 => /usr/lib/libpq.so.4 (0x00524000)
libssl.so.5 => /lib/libssl.so.5 (0x00210000)
libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x00775000)
libm.so.6 => /lib/libm.so.6 (0x002b7000)
libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x006de000)
libc.so.6 => /lib/libc.so.6 (0x00bff000)
libcrypto.so.5 => /lib/libcrypto.so.5 (0x00101000)
libkrb5.so.3 => /usr/lib/libkrb5.so.3 (0x00d74000)
libcrypt.so.1 => /lib/libcrypt.so.1 (0x009b6000)
libresolv.so.2 => /lib/libresolv.so.2 (0x00395000)
libnsl.so.1 => /lib/libnsl.so.1 (0x00248000)
libgssapi_krb5.so.2 => /usr/lib/libgssapi_krb5.so.2 (0x0025d000)
libcom_err.so.2 => /lib/libcom_err.so.2 (0x00d44000)
libk5crypto.so.3 => /usr/lib/libk5crypto.so.3 (0x00d49000)
libdl.so.2 => /lib/libdl.so.2 (0x00275000)
libz.so.1 => /usr/lib/libz.so.1 (0x004e1000)
/lib/ld-linux.so.2 (0x00a8b000)
libkrb5support.so.0 => /usr/lib/libkrb5support.so.0 (0x00d6f000)

In any case, you should try to run both servers against the same set of
libs and headers. You can have multiple copies of libpq around and
select it at compile time. The client library doesn't really have to
match the server version...

As I asked in another thread: Does it work to install the
postgresql-server RPM from the 8.1 version and the others from the
Fedora 4 included 8.0 version?

--
Bryan White, ArcaMax Publishing Inc.

Bryan White <bryan@arcamax.com> writes:

Tom Lane wrote:

One thing you should check is whether both libs were built with the same
options (compare pg_config --configure output from the 8.0 and 8.1
installations).

I think that might be the problem. These are the differences in
pg_config --configure output:

dev server:
'--host=i686-redhat-linux-gnu'
'--build=i686-redhat-linux-gnu'
'--target=i686-redhat-linux'
'--with-includes=/usr/include'
'--with-libraries=/usr/lib'
'CFLAGS=-O2 -g -march=i686 -I/usr/include/et'
'CPPFLAGS= -I/usr/include/et'
'build_alias=i686-redhat-linux-gnu'
'host_alias=i686-redhat-linux-gnu'
'target_alias=i686-redhat-linux'

web server:
'--build=i386-redhat-linux'
'--host=i386-redhat-linux'
'--target=i386-redhat-linux-gnu'
'--with-tcl'
'--with-tclconfig=/usr/lib'
'--enable-thread-safety'
'CFLAGS=-O2 -g -pipe -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -m32
-march=i386 -mtune=pentium4 -fasynchronous-unwind-tables'
'build_alias=i386-redhat-linux'
'host_alias=i386-redhat-linux'
'target_alias=i386-redhat-linux-gnu'

I'd bet on --enable-thread-safety, or perhaps -fexceptions or
-fasynchronous-unwind-tables as being the problem.

Does it work to install the postgresql-server RPM from the 8.1 version
and the others from the Fedora 4 included 8.0 version?

I think the RPMs have interdependencies that would prevent that,
although you could try it.

My advice is to get your 8.1.* RPMs from Fedora 5 rather than PGDG.
Your "web server" options look like Red Hat's standard build
environment, so the Fedora RPMs should have those same build options or
close enough. And there's no difference worth noticing other than the
Red Hat-ish build options, unless Devrim or I screwed up ;-)

If the Fedora 5 RPMs won't install on your FC4 machine, grab the FC5
SRPM and do a quick rpmbuild --rebuild to make custom RPMs for your
environment.

regards, tom lane

Tom Lane wrote:

If the Fedora 5 RPMs won't install on your FC4 machine, grab the FC5
SRPM and do a quick rpmbuild --rebuild to make custom RPMs for your
environment.

After about 5 minutes of compiling I get this:
======
pg_regress: initdb failed
Examine ./log/initdb.log for the reason.

make: *** [check] Error 2
error: Bad exit status from /var/tmp/rpm-tmp.67109 (%build)

RPM build errors:
Bad exit status from /var/tmp/rpm-tmp.67109 (%build)
====== /usr/src/redhat/BUILD/postgresql-8.1.4/
src/test/regress/log/initdb.log
Running in noclean mode. Mistakes will not be cleaned up.
initdb: cannot be run as root
Please log in (using, e.g., "su") as the (unprivileged) user that will
own the server process.

--
Bryan White, ArcaMax Publishing Inc.

Bryan White <bryan@arcamax.com> writes:

Tom Lane wrote:

If the Fedora 5 RPMs won't install on your FC4 machine, grab the FC5
SRPM and do a quick rpmbuild --rebuild to make custom RPMs for your
environment.

After about 5 minutes of compiling I get this:

initdb: cannot be run as root

Don't do the rpmbuild as root. Alternatively, I believe there's a
%define you can set to skip the regression test ... but that's
probably not a good idea.

regards, tom lane

Tom Lane wrote:

Don't do the rpmbuild as root. Alternatively, I believe there's a
%define you can set to skip the regression test ... but that's
probably not a good idea.

I think I have it solved now. I am not to familiar with the process of
building from source RPMs. You said to not do it as root but that meant
I did not have write access to /usr/src/redhat. I tried to options to
build from a different location without much luck. In the end I moved
/usr/src/redhat to /usr/src/redhat.old and created a new one (including
sub-directories) and made myself the owner. It then builds fine. Seems
like there has to be an easier way.

Anyway, after installing the new RPMs on my FC4 dev server and
rebuilding my programs, the programs do now run on my web server (stock
FC4 PostgreSQL).

Thanks for your help

--
Bryan White, ArcaMax Publishing Inc.

Bryan White <bryan@arcamax.com> writes:

Tom Lane wrote:

Don't do the rpmbuild as root.

I think I have it solved now. I am not to familiar with the process of
building from source RPMs. You said to not do it as root but that meant
I did not have write access to /usr/src/redhat.

Well, nobody at Red Hat builds RPMs in /usr/src/redhat anymore ;-).
The setup I use involves creating a directory ~/rpmwork, making
these subdirectories in it:
BUILD/ RPMS/ SOURCES/ SPECS/ SRPMS/
and making a file ~/.rpmmacros containing just
%_topdir /home/tgl/rpmwork
(adjust to suit where your work dir actually is of course). To build,
copy the SRPM into the SRPMS subdirectory, cd there, and go
rpmbuild --rebuild srpmfilename
The only part you need root for is actually installing the built RPMs
(which end up under the RPMS directory of course).

regards, tom lane