This document exists to explain the layout of the RPMs for PostgreSQL, to describe various RPM specifics, and to document special features found in the RPMset. This document is written to be applicable to version @PGMAJORVERSION@ of PostgreSQL, which is the current version of the RPMs as of this writing. More to the point, versions prior to @PGMAJORVERSION@ are not documented here. This document is intended for use only with the RPMs supplied in Red Hat Enterprise Linux, CentOS and Fedora. Note that there are also "PGDG" RPMs available directly from the upstream PostgreSQL project. Those are slightly different. For a fresh installation, you will need to initialize the cluster first (as a root user): PGSETUP_ROOT_COMMAND(@NAME_BINARYBASE@-setup --initdb) and it will prepare a new database cluster for you. Then you will need to start PostgreSQL. Now, as root, run: PGSETUP_ROOT_COMMAND([PGSETUP_SERVICE_START(@NAME_SERVICE@)]) This command will start a postmaster that will listen on localhost and Unix socket 5432 only. Edit @PGDATADIR@/postgresql.conf and pg_hba.conf if you want to allow remote access -- see the section on Grand Unified Configuration. You will probably also want to do PGSETUP_ROOT_COMMAND([PGSETUP_SERVICE_ENABLE(@NAME_SERVICE@)]) so that the postmaster is automatically started during future reboots. The file @POSTGRES_HOMEDIR@/.bash_profile is packaged to help with the setting of environment variables. You may edit this file, and it won't be overwritten during an upgrade. However, enhancements and bugfixes may be added to this file, so be sure to check .bash_profile.rpmnew after upgrading. The user 'postgres' is created during installation of the server subpackage. This user by default is UID and GID 26. The user has the default shell set to bash, and the home directory set to @POSTGRES_HOMEDIR@. This user also has no default password, so the only way to become this user is to su to it from root. If you want to be able to su to it from a non-root account or log in directly as 'postgres' you will need to set a password using passwd. Test section 2. For a minor-version upgrade (such as 9.3.1 to 9.3.4; last number changes), just install the new RPMs; there's usually nothing more to it than that. Upgrading across a major release of PostgreSQL (for example, from 9.2.x to 9.3.x) requires more effort. If you are upgrading across more than one major release of PostgreSQL (for example, from 8.3.x to 9.0.x), you will need to follow the "traditional" dump and reload process to bring your data into the new version. That is: *before* upgrading, run pg_dumpall to extract all your data into a SQL file. Shut down the old postmaster, upgrade to the new version RPMs, perform initdb, and run the dump file through psql to restore your data. In some major releases, the RPMs also support faster upgrade from concrete subset of previous releases. You can run the: PGSETUP_COMMAND(@NAME_BINARYBASE@-setup --upgrade-ids) to see what previous versions you are able to upgrade from. This is much faster than a dump and reload. To do a faster upgrade: shut down the old postmaster running against old data optionally make a backup of data directory (recommended!) install the new version's RPMs (install all the ones you had before, plus @NAME_PACKAGE@-upgrade) as root, run "@NAME_BINARYBASE@-setup --upgrade [[--upgrade-from ID]]" update the configuration files @PGDATADIR@/*.conf with any customizations you had before (your old configuration files are in old data directory or in @PGDATADIR@-old/ if you've done in-place upgrade) as root, run "PGSETUP_SERVICE_START(@NAME_SERVICE@)" the @NAME_PACKAGE@-upgrade package can be removed after the update is complete, as can old data directory NOTE: The in-place upgrade process is new and relatively poorly tested, so if your data is critical it's a really good idea to make a tarball backup of old data directory before running the upgrade. This will let you get back to where you were in case of disaster. PostgreSQL is split up into multiple packages so that users can 'pick and choose' what pieces are needed, and what dependencies are required. PackageDescription @NAME_PACKAGE@: Key client programs and basic documentation @NAME_PACKAGE@-libs: Client shared libraries @NAME_PACKAGE@-server: Server executables and data files @NAME_PACKAGE@-test: The regression tests and associated files @NAME_PACKAGE@-upgrade: Support files for upgrading from previous major version @NAME_PACKAGE@-docs: Full documentation in HTML and PDF, the tutorial files @NAME_PACKAGE@-contrib: Add-on loadable modules and programs @NAME_PACKAGE@-plperl: PL/Perl procedural language @NAME_PACKAGE@-plpython: PL/Python procedural language (for Python 2) @NAME_PACKAGE@-plpython3: PL/Python procedural language (for Python 3) @NAME_PACKAGE@-pltcl: PL/Tcl procedural language You have to install @NAME_PACKAGE@ and @NAME_PACKAGE@-libs to do anything. @NAME_PACKAGE@-server is needed unless you only plan to use the clients to work with a remote PostgreSQL server. The others are optional. Note that there are no @NAME_PACKAGE@-perl, @NAME_PACKAGE@-jdbc, @NAME_PACKAGE@-odbc, @NAME_PACKAGE@-python, @NAME_PACKAGE@-tcl, or @NAME_PACKAGE@-tk subpackages any longer. Those programs have been split off into separate source distributions. They are still available, but in some cases not under those RPM names. To be in compliance with the Linux FHS, the PostgreSQL RPMs install files in a manner not consistent with most of the PostgreSQL documentation. According to the standard PostgreSQL documentation, PostgreSQL is installed under the directory /usr/local/pgsql, with executables, source, and data existing in various subdirectories. Different distributions have different ideas of some of these file locations. In particular, the documentation directory can be /usr/doc, /usr/doc/packages, /usr/share/doc, /usr/share/doc/packages, or some other similar path. However, this installation (which usually matches the Red Hat / CentOS / Fedora RPM's) install the files like: DescriptionDirectoryExecutables@bindir@Libraries@libdir@Documentation@pgdocdir@/htmlPDF documentation@pgdocdir@Contrib documentation@pgcontribdocdir@Sourcenot installedData@PGDATADIR@Backup area@sharedstatedir@/pgsql/backupsTemplates@datadir@/pgsqlProcedural Languages@libdir@/pgsqlDevelopment Headers@includedir@/pgsqlOther shared data@datadir@/pgsqlRegression tests@libdir@/pgsql/test/regress (in the -test package) While it may seem gratuitous to place these files in different locations, the FHS requires it -- distributions should not ever touch /usr/local. It may also seem like more work to keep track of where everything is -- but, that's the beauty of RPM -- you don't have to keep track of the files, RPM does it for you. These RPMs are designed to be LSB-compliant -- if you find this not to be the case, please let us know by way of the pgsql-pkg-yum@postgresql.org mailing list. m4_ifelse([@WANT_SYSVINIT@], [1], m4_dnl [ The @NAME_PACKAGE@-server RPM contains an "initscript" that is used to start the postmaster. The current version of this script has logic to be able to start multiple postmasters, with different data areas, listening on different ports, etc. To use this functionality requires root access. As an example, let us create a secondary postmaster called, creatively enough, 'secondary'. Here are the steps: create a hard link in /etc/rc.d/init.d (or equivalent location) to '@NAME_SERVICE@' named 'secondary'. PGSETUP_ROOT_COMMAND(ln @NAME_SERVICE@ secondary) Pick a name not already used in /etc/rc.d/init.d! create a file in /etc/sysconfig/pgsql named secondary. This file is a shell script -- typically you would define PGDATA, PGPORT, and PGOPTS here. Since $PGDATA/postgresql.conf may override many of these settings, except PGDATA, you might be surprised on startup. create the target PGDATA Initdb the targe PGDATA as documented in the main documentation. Automatic initdb may or may not work for you, so a manual one is preferred. This must be done as user 'postgres' Edit postgresql.conf to change the port, address, tcpip settings, etc. Start the postmaster with 'service secondary start'. Note that there may be problems with the standard symlink -- consider this support experimental at this point in time. ], m4_dnl [ The postgresql-server package contains a systemd "unit" files @NAME_SERVICE@.service and @NAME_SERVICE@@.service. The first file is used solely to start the default PostgreSQL server. The second one is designed to allow instantiating additional PostgreSQL servers on same machine. As an example, let us create a secondary PostgreSQL service called, creatively enough, 'postgresql@secondary'. Here are the steps: Run the following command to create the necessary configuration and to initialize the new database cluster $ @NAME_BINARYBASE@-setup --initdb \ --unit postgresql@secondary \ --new-systemd-unit \ --datadir /path/to/data/directory \ --port NNNN Replace the "/path/to/data/directory" path and NNNN port with appropriate settings that don't conflict with any other PostgreSQL setup. Make sure that the parent directory of specified path has appropriate ownership and permissions. Note the SELinux issues mentioned below. Edit postgresql.conf in the target 'datadir' directory to change settings as needed. Start the new service with this command: PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_START(@NAME_SERVICE@@secondary)) You will probably also want to run the command PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_ENABLE(@NAME_SERVICE@@secondary)) so that the new service is automatically started in future reboots. When doing a major-version upgrade of a secondary service, add the service name to the @NAME_BINARYBASE@-setup command, for example: PGSETUP_ROOT_COMMAND(@NAME_BINARYBASE@-setup --upgrade --unit @NAME_SERVICE@@secondary) This will let @NAME_BINARYBASE@-setup find the correct data directory from the proper configuration file. If you are running SELinux in enforcing mode (which is highly recommended, particularly for network-exposed services like PostgreSQL) you will need to adjust SELinux policy to allow the secondary server to use non-default PGPORT or PGDATA settings. To allow use of a non-default port, say 5433, do this as root: PGSETUP_ROOT_COMMAND(semanage port -a -t postgresql_port_t -p tcp 5433) To allow use of a non-default data directory, say /special/pgdata, do: PGSETUP_ROOT_COMMAND(semanage fcontext -a -t postgresql_db_t "/special/pgdata(/.*)?") If you already created the directory, follow that with: PGSETUP_ROOT_COMMAND(restorecon -R /special/pgdata) These settings are persistent across reboots. For more information see "man semanage". ]) If you install the @NAME_PACKAGE@-test RPM then you can run the PostgreSQL regression tests. These tests stress your database installation and produce results that give you assurances that the installation is complete, and that your database machine is up to the task. To run the regression tests under the RPM installation, make sure that the PostgreSQL server has been started (if not, su to root and do PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_START(@NAME_SERVICE@)) su to postgres, cd to @libdir@/pgsql/test/regress and execute "make check". This command will start the regression tests and will both show the results to the screen and store the results in the file regress.out. If any tests fail, see the file regression.diffs in that directory for details, and read the "Regression Tests" section of the PostgreSQL documentation to find out whether the differences are actually significant. If you need help interpreting the results, contact the pgsql-general list at postgresql.org. After testing, run "make clean" to remove the files generated by the test script. Then you can remove the @NAME_PACKAGE@-test RPM, if you wish. Fedora / Red Hat / CentOS use the systemd package to manage server startup. A systemd unit file for PostgreSQL is provided in the server package, as @systemdunitsdir@/@NAME_SERVICE@.service. To start the postmaster manually, as root run PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_START(@NAME_SERVICE@)) To shut the postmaster down, PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_STOP(@NAME_SERVICE@)) These two commands only change the postmaster's current status. If you want the postmaster to be started automatically during future system startups, run PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_ENABLE(@NAME_SERVICE@)) To undo that again, PGSETUP_ROOT_COMMAND(PGSETUP_SERVICE_DISABLE(@NAME_SERVICE@)) m4_ifelse([@WANT_SYSVINIT@], [1], m4_dnl [See "man chkconfig" and "man service" for other possible subcommands.],m4_dnl [See "man systemctl" for other possible subcommands.])) The PostgreSQL server has many tunable parameters -- the file @PGDATADIR@/postgresql.conf is the master configuration file for the whole system. The RPM ships with a mostly-default file -- you will need to tune the parameters for your installation. In particular, you might want to allow nonlocal TCP/IP socket connections -- in order to allow these, you will need to edit the postgresql.conf file. The line in question contains the string 'listen_addresses' -- you need to both uncomment the line and set the value to '*' to get the postmaster to accept nonlocal connections. You'll also need to adjust pg_hba.conf appropriately. By default, the postmaster's stderr log is directed into files placed in a pg_log subdirectory of the data directory (ie, @PGDATADIR@/pg_log). The out-of-the-box configuration rotates among seven files, one for each day of the week. You can adjust this by changing postgresql.conf settings. If your distribution is not supported by the binary RPMs from PostgreSQL.org, you will need to rebuild from the source RPM. If you have not previously rebuilt any RPMs, set up the required environment: make a work directory, say ~/rpmwork, then cd into it and do PGSETUP_ROOT_COMMAND(mkdir BUILD BUILDROOT RPMS SOURCES SPECS SRPMS) Then make a file ~/.rpmmacros containing %_topdir full_path_to_work_directory_here Download the postgresql .src.rpm for the release you want and place it in the SRPMS subdirectory, then cd there and execute PGSETUP_COMMAND(rpmbuild --rebuild postgresql-nnn.src.rpm) The results will appear under the RPMS subdirectory. You will have to have a full development environment to rebuild the RPM set. If rpmbuild complains of lack of certain packages, install them and try again. In some cases, you can disable features to avoid needing some development packages, as detailed next. This release of the RPMset includes the ability to conditionally build sets of packages. The parameters, their defaults, and the meanings are: VariableDefaultCommentbeta0build with cassert and do not strip the binariesrunselftest1do "make check" during the buildtest1build the postgresql-test packageupgrade1build the postgresql-upgrade packageplpython1build the PL/Python procedural language packageplpython31build the PL/Python3 procedural language packagepltcl1build the PL/Tcl procedural language packageplperl1build the PL/Perl procedural language packagessl1build with OpenSSL supportkerberos1build with Kerberos 5 supportldap1build with LDAP supportnls1build with national language supportpam1build with PAM supportsdt1build with SystemTap supportxml1build with XML supportpgfts1build with --enable-thread-safetyselinux1build contrib/selinuxuuid1build contrib/uuid-ossp To use these defines, invoke a rebuild like this: $ rpmbuild --rebuild \ --define 'plpython 0' \ --define 'pltcl 0' \ --define 'test 0' \ --define 'runselftest 0' \ --define 'kerberos 0' \ postgresql-9.2.0-1.src.rpm This command would disable the plpython, pltcl, and test subpackages, disable the regression test run during build, and disable kerberos support. You might need to disable runselftest if there is an installed version of PostgreSQL that is a different major version from what you are trying to build. The self test tends to pick up the installed libpq.so shared library in place of the one being built :-(, so if that isn't compatible the test will fail. Also, you can't use runselftest when doing the build as root. More of these conditionals will be added in the future. The contents of the contrib tree are packaged into the -contrib subpackage and are processed with make and make install. There is documentation in @pgcontribdocdir@ for these modules. Most of the modules are in @libdir@/pgsql for loadable modules, and binaries are in @bindir@. In the future these files may be split out, depending upon function and dependencies. You can get more information at http://www.postgresql.org and http://yum.postgresql.org Please help make this packaging better -- let us know if you find problems, or better ways of doing things. You can reach us by e-mail at pgsql-pkg-yum@postgresql.org or fail a bug against postgresql component on bugzilla.redhat.com.