Forum OpenACS Q&A: Oracle hotbackup scripts?
I've written a nice little script to do nightly Oracle exports using named pipes, gzip, and split (as suggested in Expert One-on-One Oracle by Thomas Kyte), which I can upload here if anyone's interested, but I don't have a similar solution set up yet for doing periodic hot and cold physical backups of Oracle (using archivelog mode for hot backups, etc.)
There is a good example from a book of Oracle Press called Oracle 8i Backup and Recovery. The scripts from the book need some syntax check but they are quite useful.
I don't think I am entitled to put the scripts somewhere since I don't know about all the copyright stuff involved
telinit 3 restart-aolserver intranet /etc/rc.d/init.d/oracle8i stop nice tar cfz /tmp/ora8.tar.gz /ora8 /etc/rc.d/init.d/oracle8i start telinit 4Aolserver runs at runlevel 4, so the telinit 3 is turning Aolserver off. I have a separate script that puts the backed up file somewhere safe.
Just run "ora-export.sh -h
" to get help on the command
line options.
#!/bin/sh # # ora-export.sh # $Header: /home/cvsroot/unix/usr/local/adm/bin/ora-export.sh,v 1.3 2002/06/17 17:24:02 andy Exp $ # # by atp@piskorski.com, 2002/05/29 # # Script to do an Oracle export using a named pipe, gzip, and split, # as described in _Expert One-on-One Oracle_ by Thomas Kyte, # pg. 327-328, "Export to an OS Pipe". This lets us avoid exp's # possible 2 gig file size limit, use less work space on disk by not # first writing out an un-compressed export file, etc. # We assume that /etc/profile sets and exports ORACLE_SID, # ORACLE_HOME, etc.: . /etc/profile # Various defult settings, most of which may be overridden on the # command line: # This PASSWORD_FILE must be a Bourne shell script which these # variables like this: # ORA_USERNAME="username" # ORA_PASSWORD="password" # TODO: Putting the password in a file like this keeps us from having # to hard-code it into multiple script files, but the password STILL # DOES show up in ps -ef output - not good. PASSWORD_FILE="/usr/local/adm/bin/ora-system-password.sh" PUT_IN_DIR="/web/oracle-backups" MAX_SIZE=600 FULL_P=1 LOG_DATE_FORMAT='+%a %Y-%m-%d %H:%M:%S %Z' # TODO: For use on multiple machines, may want to move the above # default settings into a separate configuration file, and source that # file here. Perhaps just use the same $PASSWORD_FILE. CMDNAME=`basename $0` USAGE="Do nightly Oracle export, using exp, gzip, and split. Usage: ${CMDNAME} [-p FILE] [-d DIRECTORY] [-m SIZE] [-f {0|1}] [-o USER] [-r LOG_FILE] Where: -p : Username/password file to source to set the ORA_USERNAME and ORA_PASSWORD variables. Default: $PASSWORD_FILE -d : Directory where we will put the Oracle export file. Default: $PUT_IN_DIR -m : Max size of each dump file chunk, in megabytes. Default: $MAX_SIZE -f : Full instance export, 1 for yes, 0 for no. Default: $FULL_P -o : Oracle schema owner user name to export. Only needed if -f is no. -r : Redirect stdout and stdin to file. Set to empty string to turn off default redirection to the log file. " password_file=$PASSWORD_FILE put_in_dir=$PUT_IN_DIR max_size=$MAX_SIZE full_p=$FULL_P owner="" redirect_p=0 direct_p=1 compress_p=1 consistent_p=1 while getopts hp:d:m:f:o:r: OPT do case $OPT in h) echo "$USAGE" exit ;; p) password_file=$OPTARG ;; d) put_in_dir=$OPTARG ;; m) max_size=$OPTARG ;; f) full_p=$OPTARG ;; o) owner=$OPTARG ;; r) STDOUT=$OPTARG STDERR=$OPTARG redirect_p=1 ;; ?) echo 1>&2 echo "ERROR: One or more bad command line options." 1>&2 echo 1>&2 echo "$USAGE" 1>&2 exit 1 ;; esac done shift `expr $OPTIND - 1` ORA_USERNAME="" ORA_PASSWORD="" . $password_file weekday=`date "+%a"` # TODO: We only have one Oracle instance on this machine. If we add # another, we'll need to change things to optionally take the # ORACLE_SID as a command line option, or something like that: # --atp@piskorski.com, 2002/05/29 11:03 EDT if [ "$full_p" -eq 1 ] then full_opt="full=y" owner_opt="" dump_file="${ORACLE_SID}-${ORA_USERNAME}-full-${weekday}.dmp.gz" log_file="${ORACLE_SID}-${ORA_USERNAME}-full-${weekday}.log" else full_opt="full=n" if [ "$owner" = "" ] then owner="$ORA_USERNAME" fi owner_opt="owner=$owner" dump_file="${ORACLE_SID}-${owner}-${weekday}.dmp.gz" log_file="${ORACLE_SID}-${owner}-${weekday}.log" fi # Redirect stdout and stderr to where? Settings STDOUT and STDERR to # empty string will cause no redirection to occur at all: if [ "$redirect_p" -ne 1 ] then STDOUT=$log_file STDERR=$log_file fi # Truncate the old previous log file: cat -s '' > $log_file if [ "$ORA_USERNAME" = "" ] then echo "Error: ORA_USERNAME not set." 1>&2 1>>$STDOUT 2>>$STDERR exit 1 fi if [ "$ORA_PASSWORD" = "" ] then echo "Error: ORA_PASSWORD not set." 1>&2 1>>$STDOUT 2>>$STDERR exit 1 fi if [ "$direct_p" -eq 1 ] then direct_opt="direct=y" else direct_opt="direct=n" fi if [ "$compress_p" -eq 1 ] then compress_opt="compress=y" else compress_opt="compress=n" fi if [ "$consistent_p" -eq 1 ] then consistent_opt="consistent=y" else consistent_opt="consistent=n" fi # The process id of this running script: pid=$$ ora_exp_maxsize="${max_size}m" pipe="/tmp/${dump_file}-${pid}" cd $put_in_dir if [ -p $pipe ] then # We've included the PID in the pipe name, so this is pretty # unlikely: echo "WARNING: Named pipe $pipe already existed!" 1>&2 1>>$STDOUT 2>>$STDERR echo `ls -l $pipe` 1>&2 1>>$STDOUT 2>>$STDERR fi rm -f $pipe mknod $pipe p if [ $? -ne 0 ] then echo "ERROR: Failed to create named pipe: $pipe" 1>&2 1>>$STDOUT 2>>$STDERR exit 1 fi echo "$CMDNAME export STARTING: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR # Note: If this week's export is smaller than last week's, then simply # overwriting the files might leave one extra split file from LAST # week, which would mess things up when we re-combine the split files # in order to do an import. So go and delete the old dump files # first, rather than simply overwriting them: rm -f ${dump_file}.* gzip < $pipe | split -b $ora_exp_maxsize - ${dump_file}. & exp $ORA_USERNAME/$ORA_PASSWORD file=$pipe $full_opt $owner_opt $consistent_opt $direct_opt $compress_opt 1>>$STDOUT 2>>$STDERR echo "$CMDNAME export DONE: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR # To import the dump files we just created, something like this will work: # # cat `ls -1 ${dump_file}.* | sort` | gunzip > $pipe & # imp $ORA_USERNAME/$ORA_PASSWORD file=$pipe show=y rm -f $pipe
It's a real monster, but every bit of bloat was added for a reason.
I went to Curtis's "Oracle backup and recovery for sysadmins" tutorial at the last USENIX LISA. Highly recommended if you have the means.
You should move the cd $put_in_dir
line up to be
before the if [ "$redirect_p" -ne 1 ]
line, not
where it is now. If the cd isn't before the cat -s '' >
$log_file
truncation of the old log file, the log file never
gets truncated at all and it will just keep growing week after week.
#!/bin/sh # # ora-rman-hb.sh # $Header: /home/cvsroot/unix/usr/local/adm/bin/ora-rman-hb.sh,v 1.4 2002/06/27 22:12:43 andy Exp $ # # by atp@piskorski.com, 2002/06/19 # originally based on a similar script by Xuequn "Robert" Xu # <xux@arsdigita.com>, c. 2000/10/03. # Oracle physical hot backup using RMAN. Run as user oracle from # crontab something like so, to run every night at 0:45 am: # # 45 0 * * * /usr/local/adm/bin/ora-rman-hb.sh # We assume that /etc/profile sets and exports ORACLE_SID, # ORACLE_HOME, etc.: . /etc/profile BACKUP_ROOT="/web/oracle-backups" backup_root=$BACKUP_ROOT backup_dir="$backup_root/rman" backup_dir_old_1="$backup_root/rman-old-1" backup_dir_old_7="$backup_root/rman-old-7" ARCHIVED_LOGS_DIR="$ORACLE_BASE/admin/ora8/arch" PASSWORD_FILE="/usr/local/adm/bin/ora-system-password.sh" MAX_SIZE_KBYTES=614400 LOG_DATE_FORMAT='+%a %Y-%m-%d %H:%M:%S %Z' ORA_USERNAME="" ORA_PASSWORD="" . $PASSWORD_FILE CMDNAME=`basename $0` log_file=`basename $CMDNAME`.log USAGE="Do nightly Oracle 8.1.7 physical hot backup using RMAN. Usage: ${CMDNAME} [-h] [-t] Where: -h : Show this help and exit. -t : Don't do use the 'backup database' command, use 'backup tablespace' instead. -r : Redirect stdout and stdin to file. Set to empty string to turn off default redirection to the log file. Default: $backup_dir/$log_file -d : Directory root where we will place our directories of backup files. This must be an ABSOLUTE pathname, and the oracle unix user must be able to write to it. Default: $BACKUP_ROOT So by default, the backup files will be placed in: $backup_dir/ $backup_dir_old_1/ $backup_dir_old_7/ " tablespace_p=0 redirect_p=0 while getopts htr:d: OPT do case $OPT in h) echo echo "$USAGE" echo exit ;; t) tablespace_p=1 ;; r) STDOUT=$OPTARG STDERR=$OPTARG redirect_p=1 ;; d) backup_root=$OPTARG backup_dir="$backup_root/rman" backup_dir_old_1="$backup_root/rman-old-1" backup_dir_old_7="$backup_root/rman-old-7" ;; ?) echo 1>&2 echo "ERROR: One or more bad command line options." 1>&2 echo 1>&2 echo "$USAGE" 1>&2 exit 1 ;; esac done shift `expr $OPTIND - 1` # Redirect stdout and stderr to where? Settings STDOUT and STDERR to # empty string will cause no redirection to occur at all: if [ "$redirect_p" -ne 1 ] then STDOUT=$backup_dir/$log_file STDERR=$backup_dir/$log_file fi create_dir() { # Takes the path name of directory to create. Allows recursive # creation. If the file already exists, returns an error if it is a # normal file, does nothing if the file is a directory. return_code=0 if [ -f $1 ] then return_code=1 echo "Error: A file '$1' already exists. Cannot create directory '$1'." elif [ ! -d $1 ] then mkdir -p $1 return_code=$? fi return $return_code } # Delete the 2 day old backup, and move the yesterday's backup from # the current to the 1 day old directory. On Monday, instead of # deleting the old Saturday 12:45 am backup, move it into the week-old # directory: create_dir $backup_dir create_dir $backup_dir_old_1 create_dir $backup_dir_old_7 # 0 is Sunday: weekday_num=`date "+%w"` if [ $weekday_num -eq 1 ] then /bin/rm -f $backup_dir_old_7/* /bin/mv -f $backup_dir_old_1/* $backup_dir_old_7/ fi /bin/rm -f $backup_dir_old_1/* /bin/mv -f $backup_dir/* $backup_dir_old_1/ if [ "$tablespace_p" -eq 1 ] then # Using 'backup database' also includes the temp tablespace, which # we don't need, and which on our machien is currently ludicrously # large. So for now, we want to use 'backup tablespace' instead. # For some info on fixing the temp tablespace problem, see: # https://openacs.org/bboard/q-and-a-fetch-msg.tcl?msg_id=000542 # Therefore, generate the list of tablespaces we went to backup, # excluding the temp tablespace: # --atp@piskorski.com, 2002/06/20 09:36 EDT tablespace_list=`$ORACLE_HOME/bin/sqlplus -S $ORA_USERNAME/$ORA_PASSWORD <<EOS 1>>$STDOUT 2>>$STDERR set feed off linesize 500 pages 0 select tablespace_name from dba_tablespaces where tablespace_name not like 'TEMP%'; exit EOS` tablespace_list=`echo $tablespace_list | /bin/sed -e 's/ /, /g'` backup_what="tablespace $tablespace_list include current controlfile ;" else # Use 'backup datbase' rman command, NOT 'backup tablespace': backup_what="database ;" fi echo "$CMDNAME: We will backup: $backup_what" 1>>$STDOUT 2>>$STDERR # Run rman, executing the rman script lines up to the "EOS" sign: # # We set the max size on the channel of a single backup piece file to # MAX_SIZE_KBYTES. We pick a default value of 600 MB so that we're # certain we could fit each file onto a 650 MB CD-R disk if for some # reason we want to. (If 1 kb = 1023 bytes, 600 MB = 614400 kb. If 1 # kb = 1000 bytes, 600 MB = 629145.6 kb.) # # Note that while the Oracle docs are vague on the distinction, # filesperset is the maximum number of Oracle data files to read into # the backup set, NOT the max number of "backup piece" files to write # out! Presumably, as many backup pieces will be created as # necessary, and the max size of each backup piece is defined by the # 'set limit channel' command. # # We could also allocate additional channels in order to spread the # backup across multiple disks. Each single backup set goes to only # one channel, an each channel can have 1 or more backup sets. echo "$CMDNAME: rman backup STARTING: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR $ORACLE_HOME/bin/rman target / nocatalog <<EOS 1>>$STDOUT 2>>$STDERR run { allocate channel m01 type disk format '$backup_dir/%d-t%t-s%s-p%p.bak'; allocate channel m02 type disk format '$backup_dir/%d-t%t-s%s-p%p.bak'; allocate channel m03 type disk format '$backup_dir/%d-t%t-s%s-p%p.bak'; set limit channel m01 kbytes $MAX_SIZE_KBYTES; set limit channel m02 kbytes $MAX_SIZE_KBYTES; set limit channel m03 kbytes $MAX_SIZE_KBYTES; backup filesperset = 6 skip inaccessible skip offline $backup_what # Archives the current redo log plus any other non-archived redo # logs. The Oracle docs say you should do this immediately after # backing up the database: # http://otn.oracle.com/docs/products/oracle8i/doc_library/817_doc/server.817/a76990/troubler.htm#446214 # --atp@piskorski.com, 2002/06/19 21:11 EDT sql 'alter system archive log current'; # Creates a "trace file" script which when run will re-create the # control file, but all RMAN info in the control file will be lost. # We've already made a physical backup of th control file above, so # also doing this here is belt-and-suspenders: # --atp@piskorski.com, 2002/06/19 21:11 EDT sql 'alter database backup controlfile to trace'; } exit EOS echo "$CMDNAME: rman backup DONE: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR # Note that the 'backup database' command does NOT backup the archived # redo logs. Since our nightly file-system backs them up directly # from where Oracle puts them, this is ok, we don't need to do a # 'backup archivelog' command in RMAN. # Note that since we have put them under CVS control, here we are NOT # backing up the init.ora file, listener.ora, and other config files # located in: # $ORACLE_BASE/admin/ora8/pfile/ # $ORACLE_HOME/network/admin/ # TODO: The 'backup database' command backs up the control file, but # it seems unclear WHICH version of the control file it backs up - the # control file as it existed BEFORE RMAN started, or the version AFTER # the RMAN backup of the database files completes, when it will # include recovery catalog information for the database backup just # completed. If RMAN backs up the before version, perhaps we would # also want to do 'backup current controlfile' or 'include current # controlfile' after the 'backup database'. I recall seeing at least # one example in the Oracle docs that implies that this might be a # good thing to do, but none of the rest the examples do so, so it's # probably not necessary. # --atp@piskorski.com, 2002/06/19 19:56 EDT # It would be nice to have RMAN write to a named pipe and have gzip # read from the named pipe at the same time in the background, like we # do in the ora-export.sh script, but I'm not even really certain # whether that's feasible. So to keep things simple, just gzip # separately after RMAN is done: --atp@piskorski.com, 2002/06/25 09:19 EDT echo "$CMDNAME: gzip STARTING: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR /bin/gzip $backup_dir/*.bak 1>>$STDOUT 2>>$STDERR echo "$CMDNAME: gzip DONE: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR # Remove archived redo logs more than 12 days old: # # Since on Monday we move the previous Saturday backup into the week # old directory (2 days old), a minimum of ** 10 ** days of archived # redo logs (7 days + 2 + 1 extra to account for different times # during the day) should insure that we always have enough archived # logs on disk to take us all the way back to the oldest full backup # which we have on disk. /bin/find $ARCHIVED_LOGS_DIR ( -ctime +12 ) -exec /bin/rm {} ; 1>>$STDOUT 2>>$STDERR echo "$CMDNAME: everything DONE: `date "$LOG_DATE_FORMAT"`" 1>>$STDOUT 2>>$STDERR