Installing Oracle9i on RedHat Linux

Introduction

This version of Oracle is easier to install than many of the previous ones. However there are still some traps and issues to get around. if you just need a database with transaction support, get Postgresql (www.postgresql.org) - it is very easy to install.

Before You Get Oracle

This release recommends 512 Mb of RAM, 1 GB of swap and 3 GB of disk ! You can get away with 320 MB, 750 Mb and 2.5 Gb respectively - provided you do not attempt to create a database at install time (wait until later !)

Do not attempt to install at all if you have less than 200 Mb of RAM as the installer will crash at link time ! (If you are an Oracle install pro then you can link manually to complete the process, but its a bit of a pest).

This release requires a 2.4.4 Kernel and Glibc 2.2.2. However you can install with 2.4.2 (Redhat 7.1) or 2.4.3 (Mandrake 8.0) Kernels.

There is an undocumented dependency on Binutils. The version that comes with Redhat 7.1 causes Oracle to be unlinkable. The solution is to use the version that comes with Redhat 7.0 (binutils-2.10.0.18) or Mandrake 8.0 (binutils-2.10.1.0.2).

The tested combinations were :

1. Redhat 7.0 with binutils-2.10.0.18
2. Mandrake 8.0 with binutils-2.10.1.0.2.

Getting Oracle

Visit www.oracle.com and click on "Oracle technology Network". The current URL for downloads is here. Select "9i Database" and then pick the "Linux" entry. You must then agree to the license before the actual download page appears.

There are three big files :

1. Linux9i_Disk1.cpio.gz (400Mb)
2. Linux9i_Disk2.cpio.gz (620 Mb)
3. Linux9i_Disk3.cpio.gz (80 Mb)

They are actually cpio archives ( not gzipped cpio archives ! ). You need fast internet access...or a friend with fast internet access.

Once downloaded then expand the archives somewhere where you have 2 GB of space :

  #  cd /var/bigspace
  #  cpio -idmv Linux9i_Disk1.cpio.gz
  #  cpio -idmv Linux9i_Disk2.cpio.gz
  #  cpio -idmv Linux9i_Disk3.cpio.gz

These expanded archives can be burned to three CDs to provide a convenient and professional looking installation.

Preparation

Decide where you want Oracle to be installed and create a directory for it ( I use /usr/oracle/[version] but its up to you). In addition create a group and user for Oracle (login as root for this) :

  #  groupadd dba
  #  useradd oracle
  #  mkdir /usr/oracle
  #  mkdir /usr/oracle/9.0
  #  chown -R oracle:dba /usr/oracle

Remember you need 3 GB or so !

In addition I usually create the file /etc/oratab used to record databases :

  #  touch /etc/oratab
  #  chown oracle:dba /etc/oratab

There is some folklore about setting a plethora of environment variables and a thing called "Optimal Flexible Architecture". This business is best summed up by saying : The procedure is neither optimal, flexible nor an architecture... but a way of laying out files that some Oracle folk like.

Ensure that you have installed X on your machine, as there is no longer a character mode installer.

Oracle, like most databases, needs to use IPC to create shared memory. Typically the default amount configured on most Linux distributions is minimal. To sort this out : ( as root here )

  # sysctl -w kernel.shmmax=100000000
  # echo "kernel.shmmax = 100000000" >> /etc/sysctl.conf

This enables a shared memory segment to be 100 Mb ( probably enough to start you off...)

Installation

Login as oracle and proceed to install :

  $  cd /var/bigspace
  $  cd Disk1
  $  export ORACLE_HOME=/usr/oracle/9.0
  $  ./runInstaller

The Java Gui installer will start and guide you through the process.

This part is now fairly easy - just pick "9i Database", then either of "Enterprise" or "Standard Edition" depending whether you want all the fancy bits or not ( bitmap indexes, partitions, transportable tablespaces etc require Enterprise Edition).

The next step asks about what database you want, I usually say "Software Only" and do the database later.

You will be asked where JDK is... if you don't have a JDK it does not matter ( unless you wish to do Java development of course).

Then click on install and wait for a while.

At some point you will be asked about the "other locations" for the remaining software archives.. its fairly straightforward what to be there ( if you have cut CDs, ensure your terminal window is not still cd'ed into the cdrom mount directory... )

Post Installation

Now comes creating a database. There is a reasonable Gui tool for this :

  $ export PATH=$PATH:$ORACLE_HOME/bin
  $ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME/lib
  $ dbca &

Additionally there is a network configuration tool with a similar UI :

  $ netca &

However I will detail a command line creation and configuration of a database so that some understanding of what is happening can be gained:

Set up configuration for a database called db1 :

  $ cat "db1:/usr/oracle/9.0:Y >> /etc/oratab"
  $ cd $ORACLE_HOME/dbs
  $ cat initdw.ora |sed s/"#db_name = MY_DB_NAME"/"db_name = db1"/|sed s/#control_files/control_files/ > initdb1.ora

Start and create database :

  $ export PATH=$PATH:$ORACLE_HOME/bin
  $ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME/lib
  $ export ORACLE_SID=db1
  $ sqlplus /nolog <<!
  connect / as sysdba
  startup nomount
  create database db1;
  !

This creates a default database with files in $ORACLE_HOME/dbs

Now add the database meta data to actually make it useful :

$ sqlplus /nolog <<!
  connect / as sysdba
  @?/rdbms/admin/catalog
  @?/rdbms/admin/catproc
  !

Now create a user and give it wide ranging permissions :

  $ sqlplus /nolog <<!
  connect / as sysdba
  create user myuser identified by password;
  grant create session,create any table to myuser;
  grant unlimited tablespace to myuser;
  !

Now the usual sort of database stuff can be done :

  $ sqlplus myuser/password <<!
  create table mytable(id integer, val varchar2(100));
  insert into mytables values(1,'xxxxx');
  select * from mytable;
  !

You can shut down the database with :

  $ sqlplus /nolog <<!
  connect / as sysdba
  shutdown 
  !

And start it again with :

  $ sqlplus /nolog <<!
  connect / as sysdba
  startup 
  !

Simple command line versions of these are provided :

  $ dbshut
  $ dbstart

The thing that is unusual about Oracle is that there is a one to one correspondence between database server (ORACLE_SID) and database (db_name in init$ORACLE_SID.ora) - i.e : one server manages one and only one database. Compare this to Mysql or Postgresql where one server manages many databases. If you want another database, then you must go through the above steps again with ORACLE_SID set differently and a new initialization file with a new db_name in it.

Networking

It is nice if remote clients can see your database, so here is the story :

  $ echo " LISTENER = (ADDRESS_LIST= (ADDRESS=(PROTOCOL=tcp)(HOST=localhost)(PORT=1521)) (ADDRESS=(PROTOCOL=ipc)(KEY=PNPKEY)))" > $ORACLE_HOME/network/admin/listener.ora
  $ lnsrctl start
  $ echo "DB1 =  (DESCRIPTION = (ADDRESS_LIST = (ADDRESS = (PROTOCOL = TCP)(HOST = < your host >)(PORT = 1521))) (CONNECT_DATA = (SERVICE_NAME = db1)))" > $ORACLE_HOME/network/admin/tnsnames.ora

These two commands tell the network listener where to listen, and clients were to send connection requests to. This can be tested by :

  $ tnsping db1 1

This should return (ok) [number]

Housekeeping

It is easy to make Oracle start automatically. Create a script called oracle in /etc/init.d that is like : ( you need to be root here )

  #!/bin/sh
  #
  # oracle        This shell script takes care of starting and stopping
  #               the oracle services.
  #
  # chkconfig: 345 90 10
  # description: Oracle server
  #   
  # probe: true
  
  # Source function library.
  . /etc/rc.d/init.d/functions

  # Source networking configuration.
  if [ ! -f /etc/sysconfig/network ]; then
      exit 0
  fi

  # Check for echo -n vs echo /c
  if echo '/c' | grep -s c > /dev/null 2 > &1 ; then
      ECHO_N="echo -n"
      ECHO_C=""
  else
      ECHO_N="echo"
      ECHO_C='/c'
  fi



  # See how we were called.
  case "$1" in
    start)
        # Start server.
        $ECHO_N "Starting Oracle: "$ECHO_C
        su -l oracle  -c sqlplus /nolog >/dev/null <<!
                          connect / as sysdba
                          startup
  !
          su -l oracle -c "lsnrctl start > /dev/null"
        ;;
    stop)
        # Stop server.
        $ECHO_N "Stopping Oracle: "$ECHO_C
        su -l oracle  -c sqlplus /nolog > /dev/null <<!
                          connect / as sysdba
                          shutdown abort
  !
          su -l oracle -c "lsnrctl stop > /dev/null"
        ;;
    *)
        echo $"Usage: $0 {start|stop}"
        exit 1
        ;;
  esac

  exit 0

Then tell the system to run it : ( as root again )

  #  chkconfig --add oracle

It is nice to set up the Oracle .bash_profile to get ORACLE_SID and PATH working by default :

  # .bash_profile

  # Get the aliases and functions
  if [ -f ~/.bashrc ]; then
    . ~/.bashrc
  fi
  
  export BASH_ENV=$HOME/.bashrc

  export ORAENV_ASK=NO
  export ORACLE_SID=db1
  if [ -f /usr/local/bin/oraenv ];then
    . /usr/local/bin/oraenv
  fi
  ORAENV_ASK=YES
  export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME/lib

Finally a more sophisticated script to create a database in /var/oradata ( for example ) : ( as oracle )

  #!/bin/bash
  #  ---------------------------------------------------------------------------
  #  file         : createdb1.sql
  #  ---------------------------------------------------------------------------
  
  DB=db1
  DBPATH=/var/oradata
  ORACLE_SID=$DB
  
  
  mkinit()
  {
  export ORACLE_SID=$DB
  export ORAENV_ASK=NO
  . /usr/local/bin/oraenv
  cp init$DB.ora $ORACLE_HOME/dbs
  }
  
  mkdb()
  {
  sqlplus /nolog <<!
  CONNECT / AS SYSDBA
  STARTUP NOMOUNT
  CREATE DATABASE $DB 
  DATAFILE '$DBPATH/$DB/sys.dat' SIZE 200M
  AUTOEXTEND ON
  NEXT 10M
  LOGFILE GROUP 1 '$DBPATH/$DB/log01.dat' SIZE 50M,
          GROUP 2 '$DBPATH/$DB/log02.dat' SIZE 50M
  ;
  !
  }
  
  scdb()
  {
  sqlplus /nolog  <<!
  CONNECT / AS SYSDBA
  @?/rdbms/admin/catalog
  @?/rdbms/admin/catproc
  !
  }
  
  mktb()
  {
  sqlplus /nolog  <<!
  CONNECT / AS SYSDBA
  CREATE TABLESPACE rollback
  DATAFILE '$DBPATH/$DB/rollback.dat' SIZE 100M
  AUTOEXTEND ON
  NEXT 100M
  DEFAULT STORAGE ( INITIAL 10M NEXT 10M PCTINCREASE 0)
  ;
  CREATE TABLESPACE temp
  DATAFILE '$DBPATH/$DB/temp.dat' SIZE 100M
  AUTOEXTEND ON
  NEXT 100M
  DEFAULT STORAGE ( INITIAL 10M NEXT 10M PCTINCREASE 0)
  TEMPORARY
  ;
  CREATE TABLESPACE data
  DATAFILE '$DBPATH/$DB/data.dat' SIZE 200M
  AUTOEXTEND ON
  NEXT 100M
  EXTENT MANAGEMENT LOCAL 
  SEGMENT SPACE MANAGEMENT AUTO
  ;
  !
  }
  
  
  mkoj()
  {
  sqlplus /nolog  <<!
  CONNECT / AS SYSDBA
  CREATE ROLLBACK SEGMENt t0 ;
  ALTER ROLLBACK SEGMENT t0 ONLINE;
  SET TRANSACTION USE ROLLBACK SEGMENT t0;
  CREATE ROLLBACK SEGMENT r0 TABLESPACE rollback;
  CREATE ROLLBACK SEGMENT r1 TABLESPACE rollback;
  ALTER ROLLBACK SEGMENT t0 OFFLINE;
  DROP ROLLBACK SEGMENT t0;
  !
  echo "rollback_segments = (r0,r1)" >> $ORACLE_HOME/dbs/init$DB.ora
  }
  
  resp=n
  echo -n "Create initial files ?"
  read resp
  case $resp in
    y) echo "Creating initialization files"
       mkinit
       ;;
  esac
  
  
  resp=n
  echo -n "Create database ?"
  read resp
  case $resp in
    y) echo "Creating $DB"
       mkdb
       ;;
  esac
  
  
  resp=n
  echo -n "Run database scripts ?"
  read resp
  case $resp in
    y) echo "Running scripts for $DB"
       scdb
       ;;
  esac
  
  
  resp=n
  echo -n "Create tablespaces ?"
  read resp
  case $resp in
    y) echo "Creating tablespaces for $DB"
       mktb
       ;;
  esac
  
  
  resp=n
  echo -n "Create objects ?"
  read resp
  case $resp in
    y) echo "Creating some objects for $DB"
     mkoj
     ;;
esac