postgres/doc/FAQ_Linux

711 lines
27 KiB
Plaintext

=======================================================
Frequently Asked Questions (FAQ) for PostgreSQL >= V6.1
Linux Specific
TO BE READ IN CONJUNCTION WITH THE NORMAL FAQ
=======================================================
last updated: Fri Jun 19 13:35:00 BST 1998
current maintainer: Andrew C.R. Martin (martin@biochem.ucl.ac.uk)
original author: Andrew C.R. Martin (martin@biochem.ucl.ac.uk)
Changes in this version (* = modified, + = new, - = removed):
+1.23) Why does the compile fail with messages about F_BOOLIN, F_BOOLOUT
and F_BYTEAIN being undeclared?
This file is divided approximately as follows:
1.*) Installing PostgreSQL
2.*) Compiling accessory programs
3.*) Runtime Problems
Questions answered:
1.1) What changes do I need to make to src/Makefile.global or
src/Makefile.custom and are there any other changes needed?
1.2) Why do I get problems with missing libreadline?
1.3) [REDHAT] Why do I get problems with missing libdl and dlfcn.h?
1.4) [SLACKWARE 3.1] Why do I get problems with missing libdl and dlfcn.h?
1.5) My compile of the backend dies complaining about the include file
dlfcn.h missing
1.6) GCC complains about an ignored option -fpic
1.7) I get warnings of the form
warning: cast from pointer to integer of different size
1.8) [SuSE-Linux 4.2-4.4] Where is curses and termcap?
1.9) Why do I get problems with ld.so?
1.10) Why do I get `yy_flush_buffer undefined' errors?
1.11) How do I compile PostgreSQL on an a.out system?
1.12) Why does make fail with:
yacc -d /disk2/PostgreSQL/src/backend/parser/gram.y
make: /usr/bin/make: cannot execute binary file
1.13) What are the references in X11_LIB to libsocket and libnsl in
src/Makefile.global?
1.14) [DEBIAN] Where is libtermcap?
1.15) [REDHAT] Can I get PostgreSQL as an RPM?
1.16) While trying to compile a development version under Linux, compilation
fails with a message like:
In file included from /usr/include/sys/sem.h:8,
from ipc.c:37:
/usr/include/asm/bitops.h:32: warning: no previous prototype for Set_bi
t'
....
make: *** [ipc.o] Error 1
1.17) When compiling postgres, gcc reports signal 11 and aborts.
1.18) Can I install 6.1.1 under MkLinux?
1.19) Why does make exit or crash?
1.20) How can I optimise for 486 or pentium processors
1.21) Why do I get strange results with printing times (for example
in the regression test 'timespan')?
1.22) Why don't I get any shared libraries for libpq when I compile
6.3.2?
1.23) Why does the compile fail with messages about F_BOOLIN, F_BOOLOUT
and F_BYTEAIN being undeclared?
2.1) The linker fails to find libX11 when compiling pgtclsh
3.1) I get an error reporting _fUnKy_POSTPORT_sTuFf_ undefined when
running scripts like createuser
3.2) I run postmaster and after that system says 'Bad system call(Core
dumped)'
3.3) When I try to start the Postmaster, why do I get an error of the form
Failed Assertion("!(file != 0):(null)", File:
"/usr/local/PostgreSQL/src/backend/storage/file/fd.c", Line: 257)
!(file != 0) (0)
initdb: could not create template database
initdb: cleaning up.
3.4) Why doesn't createuser work?
3.5) Why do I get an error like:
IpcMemoryCreate: memKey=155356396 , size=760632 ,
permission=384IpcMemoryCreate: shmget(..., create, ...)
failed: Invalid argument
3.6) Why does psql fail with:
psql: can't load library 'libpq.so.1'
3.7) Other strange behaviour
3.8) When PostgreSQL is running when the system is shutdown, Linux
always fsck's the disk when rebooted.
3.9) Why does Query 32 in the regression tests take so long?
3.10) Why do I get funny rounding results in some date/time arithmetic,
such as
select '4 hours'::timespan;
returning '3 hours 59 minutes 60 seconds'?
----------------------------------------------------------------------
Section 1: Compiling PostgreSQL
----------------------------------------------------------------------
1.1) What changes do I need to make to src/Makefile.global or
src/Makefile.custom and are there any other changes needed?
Changes to the makefiles are most easily made by running the
customize shell script in the src directory which will write a
Makefile.custom for you.
The only other change you may have to make is to replace
Flex if you have version 2.5.3 which has a bug which generally
manifests itself as createuser failing (See Question 3.4)
If you modify the makefiles by hand, you *must* set the
following variable:
PORTNAME= linux
You will also need to change the following to match your own
installation:
POSTGRESDIR
If you switch on the USE_TCL option, you will need to set these:
TCL_INCDIR=
TCL_LIBDIR=
TCL_LIB=
TK_INCDIR=
TK_LIBDIR=
TK_LIB=
X11_INCDIR=
X11_LIBDIR=
X11_LIB=
On my Slackware3.0 system, these are:
TCL_INCDIR= /usr/include/tcl
TCL_LIBDIR= /usr/lib
TCL_LIB= -ltcl
TK_INCDIR= /usr/include/tcl
TK_LIBDIR= /usr/lib
TK_LIB= -ltk
X11_INCDIR= /usr/include/X11
X11_LIBDIR= /usr/X386/lib
X11_LIB= -lX11
You may also make any other changes you need as documented in
the INSTALL file and in Makefile.global
1.2) Why do I get problems with missing libreadline?
Linux systems generally don't come with the GNU readline library
installed. Either ensure you do not activate the readline options
in src/Makefile.global or src/Makefile.custom or install the GNU
readline library.
Note that Debian Linux (like FreeBSD) does come with readline
installed.
1.3) [REDHAT] Why do I get problems with missing libdl and dlfcn.h?
This manifests itself as being unable to link with functions
such as dlopen(), dlclose(), etc. during the last phase of the
compilation.
The libdl library is used for dynamic linking of user-supplied
functions at run-time. For some reason this library was missed out
from the Redhat distribution. It seems that the latest Redhat 4.0
(Colgate) fixes this.
RedHat now have a new ld.so RPM package on their FTP site.
Simply grab:
ftp://ftp.redhat.com/pub/redhat/devel/i386/RedHat/RPMS/ld.so-1.
7.14-4.i386.rpm
Install the RPM file in the usual way and off you go!
NOTE! You will have to re-run configure and do a make clean
after installing the library and before recompiling.
There has been one report of a corrupted system resulting from
programs accessing these libraries while updating them (not
altogether surprising). Consequently it is a good idea to reboot
the system before installing the new libraries and to
have as little running as possible during this upgrade. Going
into single-user mode is probably a good idea!
If you want to do it the hard way, you can obtain the library and the
header file from:
ftp://tsx-11.mit.edu/pub/linux/packages/GCC/ld.so-1.7.14.tar.gz
Alternatively, you may find precompiled binaries in
distributions/debian/buzz/binary-i386/base/ld.so-1.7.14-4.deb
on the same site, or follow the instructions given for question 1.2 for
correcting the same error with early releases of Slackware 3.1.
Don't use this method unless you know what you are doing!
1.4) [SLACKWARE 3.1] Why do I get problems with missing libdl and dlfcn.h?
This manifests itself as being unable to link with functions
such as dlopen(), dlclose(), etc. during the last phase of the
compilation.
See the answer to question 1.3. Slackware up to version 3.0 was
supplied with this library and include file and they seem to be
back in again in the latest versions of 3.1, but the early 3.1
releases (before 9th September 1996) had them missing and many
CD-ROM versions will have been pressed from the first 3.1 releases.
There has been one report of a corrupted system resulting from
programs accessing these libraries while updating them (not
altogether surprising). Consequently it is a good idea to reboot
the system before installing the new libraries and to
have as little running as possible during this upgrade. Going
into single-user mode is probably a good idea!
The easiest fix is to obtain the file ldso.tgz from the a4 disk of
a more recent Slackware distribution and unpack this file
from the root (/) directory, then do
sh install/doinst.sh
to complete the installation. Follow this with
ldconfig
NOTE! You will have to re-run configure and do a make clean
after installing the library and before recompiling.
If you want to install manually, you must first install the file
dlfcn.h in /usr/include.
Second, install the file libdl.so.1.7.14 (or whatever the latest
release is) in /lib, then do:
cd /lib
ln -sf libdl.so.1.7.14 libdl.so.1
ln -sf libdl.so.1 libdl.so
On some systems (depending on your GCC configuration) it may be
necessary to do:
cd /usr/lib
ln -sf /lib/libdl.so .
Finally
ldconfig
NOTE! You will have to re-run configure and do a make clean
after installing the library and before recompiling.
1.5) My compile of the backend dies complaining about the include file
dlfcn.h missing
See the answer to question 1.3/1.4. Don't forget that if you are using
an a.out system you must first have installed the dld package
(which is not supplied with most a.out systems) to have dlfcn.h
at all. See Question 1.11.
1.6) GCC complains about an ignored option -fpic
Earlier versions of GCC accepted either -fpic or -fPIC.
It appears that more recent versions (V2.7.2?) require -fPIC.
If you are using an ELF version of Linux, this can safely be
ignored as -fPIC is the default.
You can correct this by editing src/Makefile.global and
changing CFLAGS_SL
1.7) I get warnings of the form
warning: cast from pointer to integer of different size
These were seen in earlier versions of Postgres95 and could
safely be ignored. PostgreSQL V6.0 should compile with no warnings
except those related to system header files (which can also
be safely ignored).
1.8) [SuSE-Linux 4.2-4.4] Where is curses and termcap?
SuSE-Linux 4.2 has ncurses but not curses. 4.4 appears to have both.
SuSE-Linux also has the termcap library is in /usr/lib/termcap
instead of in /usr/lib.
PostgreSQL (up to V6.0)
-----------------------
Set the value of CURSES_LIB in src/Makefile.custom to -lncurses
(or do this through the customize script).
Add the following line to src/Makefile.custom:
LDADD_BE+= -L/usr/lib/termcap
You may need to edit src/bin/psql/Makefile and change:
ifeq ($(PORTNAME), linux)
LD_ADD+=
to:
ifeq ($(PORTNAME), linux)
LD_ADD+= -ltermcap
PostgreSQL (V6.1)
-----------------
The configure script doesn't know to look in /usr/lib/termcap for
the termcap library, so you should specify this as one of the
library directories when asked for additional directories to
search.
If this doesn't work (I don't have SuSE to verify that it does)
then after running configure, you need to edit
src/Makefile.global and add -ltermcap to the LDFLAGS line
(after -lreadline). (Alternatively, you can modify
src/Makefile.custom before running configure.)
Some versions of SuSE provide only ncurses, so you may need
to force use of ncurses rather than curses by changing
-lcurses to -lncurses. (Reported true for SuSE 5.1)
1.9) Why do I get problems with ld.so?
If you get problems with ld.so, another library required under
ELF for dynamic loading, then you have messed up your installation
or (more likely) upgrade of Linux.
See the answers to Question 1.3/1.4. You may need to install
ld.so.x.y.z in /lib and run ldconfig.
The most recent stable release of the ld package is 1.7.14
At the time of writing, 1.8.x versions of ld are experimental.
1.10) Why do I get `yy_flush_buffer undefined' errors?
This isn't really Linux specific, but is common on older Linux
installations. You must have a recent version of flex (2.5.2 or later)
to compile PostgreSQL. Note that flex 2.5.3 has a bug: see
Question 3.4.
1.11) How do I compile PostgreSQL on an a.out system?
First, you must install the dld library. This may be obtained
from Sunsite as:
Linux/libs/dld.3.2.7.tar.gz
(ftp://sunsite.unc.edu/pub/Linux/libs/dld.3.2.7.tar.gz)
Second, add the following line to src/Makefile.custom:
LINUX_ELF=
(or use the customize script)
1.12) Why does make fail with:
yacc -d /disk2/PostgreSQL/src/backend/parser/gram.y
make: /usr/bin/make: cannot execute binary file
This was a problem in earlier versions of Postgres95. The default
for PostgreSQL is to use bison -y rather than yacc.
yacc is generally implemented as a script which invokes bison -y
For some reason (certain versions of make? certain versions of
bash?) make is unable to execute this script file.
To correct this, simply edit src/mk/port/postgres.mk.linux
and, at the end of the file, change:
# YACC = bison -y
to
YACC = bison -y
1.13) What are the references in X11_LIB to libsocket and libnsl in
src/Makefile.global?
This was a problem in 1.08 (they are Sun Solaris specific).
It is fixed in 1.09 and 6.0
1.14) [DEBIAN] Where is libtermcap?
Debian Linux comes without the termcap library and uses ncurses
(which uses terminfo instead). There is no need to change the
CURSES_LIB variable in src/bin/psql/Makefile since Debian provides
a link from libncurses to libcurses (unlike SuSE-Linux --- see
Question 1.8).
You may need to edit src/bin/psql/Makefile and comment out the
change:
ifeq ($(PORTNAME), linux)
LD_ADD+= -ltermcap
to:
ifeq ($(PORTNAME), linux)
LD_ADD+=
1.15) [REDHAT] Can I get PostgreSQL as an RPM?
Yes! Michal Mosiewicz
(http://www.pdi.lodz.pl/~mimo) has kindly put together an RPM
for PostgreSQL V6.0 on Intel architectures which he has uploaded to
ftp://ftp.redhat.org/pub/Incoming/Postgres-6.0-1.i386.rpm
This is a pre-compiled version, the source RPM is on its
was as I write (3rd Feb 1997).
1.16) While trying to compile a development version under Linux, compilation
fails with a message like:
In file included from /usr/include/sys/sem.h:8,
from ipc.c:37:
/usr/include/asm/bitops.h:32: warning: no previous prototype for Set_bi
t'
....
make: *** [ipc.o] Error 1
The problem is that Linux provides no prototypes for these
inline functions. The solution is to go into the
.../src/backend/storage/ipc directory and edit the Makefile.
Change the line
CFLAGS+=$(INCLUDE_OPT)
to
CFLAGS+=$(INCLUDE_OPT) -Wno-error
Do the same in the ../src/backend/storage/lmgr directory.
1.17) When compiling postgres, gcc reports signal 11 and aborts.
More specifically:
gcc: Internal compiler error: program cc1 got fatal
signal 11
This may be a hardware/memory problem. PortgreSQL is a big
program, and large gcc compilations (such as building
PostgreSQL or bebuilding the kernel) stress memory like
few other programs, resulting in errors that do not occur
in normal operation. Lesser operating systems are also
unlikely to stress the hardware to this degree so you
may never see any problems under DOS/Windows.
More information on this problem at:
http://www.bitwizard.nl/sig11
From this Sig11 FAQ, there appears to be a specific problem
with Redhat 5.0 gcc running on Cyrix processors. See the
URL above for more details!
1.18) Can I install 6.1.1 under MkLinux?
Tatsuo Ishii has done this under
MkLinux DR2.1 update2 after a small patch available from:
ftp://ftp.sra.co.jp/pub/cmd/postgres/6.1.1/mklinux.patch.gz
1.19) Why does make exit or crash?
There have been a couple of reports of gmake either just
exiting early or seg faulting. The latter problem was reported
with gmake 3.74 - upgrading to 3.76.1 solved the problem.
However, 3.74 is known to work fine in other people's setups.
In short, try upgrading gmake to the latest version you can
find before reporting this as a problem
1.20) How can I optimise for 486 or pentium processors
The default compiler flags perform no optimisation for 486
or Pentium processors. To add such optimisation, edit
Makefile.custom and add:
CFLAGS+= -m486
or (for the new compilers that most people are not yet running)
CFLAGS+= -mpentium
or
CFLAGS+= -mpentiumpro
1.21) Why do I get strange results with printing times (for example
in the regression test 'timespan')?
The times are appearing as: '4 hours 59 mins 60.00 secs'
instead of '5 hours'
This is a problem with the glibc2 library which comes with
Redhat 5.0. Update your glibc to the latest RedHat version
for v5.0/hurricane. Anything prior to glibc-2.0.7 is likely
to have the problem.
1.22) Why don't I get any shared libraries for libpq when I compile
6.3.2?
There was some last minute breakage of the Linux configuration for
v6.3.2. Look in ftp://postgresql.org/pub/patches/ for a few fix-ups,
including a linux_elf patch.
1.23) Why does the compile fail with messages about F_BOOLIN, F_BOOLOUT
and F_BYTEAIN being undeclared?
The actual messages are something like:
-I/usr/include/readline -O2 -Wall -Wmissing-prototypes -I..
-Wno-error -c bootstrap.c -o bootstrap.o
bootstrap.c:160: `F_BOOLIN' undeclared here (not in a function)
bootstrap.c:160: initializer element for `Procid[0].inproc' is not
constant
bootstrap.c:160: `F_BOOLOUT' undeclared here (not in a function)
bootstrap.c:160: initializer element for `Procid[0].outproc' is not
constant
bootstrap.c:161: `F_BYTEAIN' undeclared here (not in a function)
bootstrap.c:161: initializer element for `Procid[1].inproc' is not
constant
This is tricky unless you know why it happens, as these constants
don't seem to be defined anywhere.
The solution is to make sure that cpp is included in your path
before you start the make.
On Redhat 5.1, cpp is in /usr/lib/gcc-lib/i386-redhat-linux/2.7.2.3
----------------------------------------------------------------------
Section 2: Compiling accessory programs
----------------------------------------------------------------------
2.1) The linker fails to find libX11 when compiling pgtclsh
Add the following to src/Makefile.custom
X11_LIBDIR = /usr/X11R6/lib
----------------------------------------------------------------------
Section 3: Runtime Problems
----------------------------------------------------------------------
3.1) I get an error reporting _fUnKy_POSTPORT_sTuFf_ undefined when
running scripts like createuser
This is a bug in V1.06-V1.07 of Postgres and is fixed in V1.08
and above.
3.2) I run postmaster and after that system says 'Bad system call(Core
dumped)'
This indicates that you have not compiled shared memory support
into your kernel. You need to recompile the Linux kernel to add this
feature.
3.3) When I try to start the Postmaster, why do I get an error of the form
Failed Assertion("!(file != 0):(null)", File:
"/usr/local/PostgreSQL/src/backend/storage/file/fd.c", Line: 257)
!(file != 0) (0)
initdb: could not create template database
initdb: cleaning up.
Your permissions on the file /dev/null are wrong.
ls -l /dev/null should give you something like:
crw-rw-rw- 1 root wheel 2, 2 Oct 8 18:41 /dev/null
Correct the permissions using:
chmod a+rw /dev/null
3.4) Why doesn't createuser work?
There is a problem with Version 2.5.3 of GNU flex and createuser.
Your options are to downgrade flex to V2.5.2, upgrade to V2.5.4
or apply a patch to V2.5.3 which is supplied in doc/README.flex
You may obtain V2.5.4 from
ftp://prep.ai.mit.edu/pub/gnu/flex-2.5.4.tar.gz
3.5) Why do I get an error like:
IpcMemoryCreate: memKey=155356396 , size=760632 ,
permission=384IpcMemoryCreate: shmget(..., create, ...)
failed: Invalid argument
You haven't build IPC support into your Linux kernel. You
will have to rebuild the kernel and switch on this option.
3.6) Why does psql fail with:
psql: can't load library 'libpq.so.1'
Psql has been compiled to link dynamically with the libpq
library.
To solve this, you should log in as root and edit the file
/etc/ld.so.conf
Add a single line at the end which gives the name of the
PostgreSQL library directory (the lib subdirectory of your
PostgreSQL installation) and run
/sbin/ldconfig -v
Alternatively, (and if you don't have root access), you may
use the LD_LIBRARY_PATH environment variable.
The LD_LIBRARY_PATH variable contains a colon separated list of
paths to be searched for shared libraries. This list is
searched before the libraries specified by ldconfig.
Therefore under Bash, you would do something like:
export LD_LIBRARY_PATH='PathToPGSQL'/lib
or, using tcsh
setenv LD_LIBRARY_PATH 'PathToPGSQL'/lib
replacing 'PathToPGSQL' with the appropriate path to your top level
PostgreSQL directory
Note that the ldd command can be used on a dynamically linked
executable to list the paths to all the shared libraries upon
which the executable depends.
3.7) Other strange behaviour
I'm not sure what the symptoms might be except for nothing
working properly, but it has been pointed out that one needs
to be careful that the dynamic loader loads the correct version
of the libpq library. If you have old versions lying around
in your library path (for example in /usr/lib) these may get
loaded instead of the new version you intended to load. Make
sure you get them out of the way and look at Q3.6 for
details of loading libraries.
3.8) When PostgreSQL is running when the system is shutdown, Linux
always fsck's the disk when rebooted.
There have been some reports of this happening and it seems
to be a result of running PostgreSQL from /etc/inittab as
suggested in the INSTALL document.
You are therefore recommended to start the postmaster from an
rc script. Under a Slackware-like release, you would modify
/etc/rc.d/rc.local to start the postmaster. Under a RedHat-like
release you should create a SysV style script in
/etc/rc.d/rc3.d based on the /etc/rc.d/init.d skeleton file.
There's a sample file in contrib/linux/postgres.init
Here's another sample file supplied by John Robinson
which you should modify as needed:
#!/bin/sh
#
# postgreSQL.init This shell script takes care of starting and stopping
# the PostgreSQL postmaster.
#
# Source function library.
. /etc/rc.d/init.d/functions
# Source networking configuration.
. /etc/sysconfig/network
# Check that networking is up.
[ ${NETWORKING} = "no" ] && exit 0
# See how we were called.
case "$1" in
start)
# Start daemons.
echo -n "Starting postgres Postmaster daemon:"
if [ -z "`pidofproc postmaster`" ]
then
su postgres -c "/usr/local/pgsql/bin/postmaster -D /home/postgr
eSQL/data -p 5432 &"
echo -n " postmaster"
else
echo -n " (already running)"
fi
echo
touch /var/lock/subsys/postgres
;;
stop)
# Stop daemons.
echo -n "Shutting down postgres Postmaster daemon: "
killall -TERM postmaster 2>/dev/null
killall -TERM postgres 2>/dev/null
echo
rm -f /var/lock/subsys/postgres
;;
*)
echo "Usage: postgres {start|stop}"
exit 1
esac
exit 0
3.9) Why does Query 32 in the regression tests take so long?
This is due to a bug in regression scripts which only happens
on linux boxes. There are two workarounds as far as I know
(information from Tatsuo Ishii ):
1. change following in regress.sh:
time postgres -texecutor -tplanner -Q bench < bench.sql
to:
postgres -texecutor -tplanner -Q bench < bench.sql
2. after running the test, remove a line at the very end of
bench.out something like:
85.86user 114.47system 4:49.20elapsed 69%CPU (0avgtext+0avgdata
0maxresident)k
then type:
sh ./perquery < bench.out > & bench.out.perquery
3.10) Why do I get funny rounding results in some date/time arithmetic,
such as
select '4 hours'::timespan;
returning '3 hours 59 minutes 60 seconds'?
You are running the new glibc2 libraries and have a version earlier tha
n
2.0.7. It is a math rounding problem in the library. Upgrade your libra
ry.
----------------------------------------------------------------------------
Dr. Andrew C.R. Martin University College London
EMAIL: (Work) martin@biochem.ucl.ac.uk (Home) andrew@stagleys.demon.co.uk
URL: http://www.biochem.ucl.ac.uk/~martin
Tel: (Work) +44(0)171 419 3890 (Home) +44(0)1372 275775