| Prev | Chapter 6. Maintenance | Next |
|---|
The purpose of backup is to enable recovery. Backup and recovery are always risky; here are some steps that minimize the chance recovery is necessary:
Store everything on a fault-tolerant disk array (RAID 1 or 5 or better).
Use battery backup.
Use more reliable hardware, such as SCSI instead of IDE.
These steps improve the chances of successful recovery:
Store backups on a third disk on another controller
Store backups on a different computer on a different network in a different physical location. (Compared to off-line backup such as tapes and CDRs, on-line backup is faster and more likely to succeed, but requires maintenance of another machine.)
Plan and configure for recovery from the beginning.
Test your recovery strategy from time to time.
Make it easy to maintain and test your recovery strategy, so that you are more likely to do it.
OpenACS installations comprise files and database contents. If you follow the reference install and put all files, including configuration files, in /var/lib/aolserver/service0/, and back up the database nightly to a file in /var/lib/aolserver/service0/database-backup, then you can apply standard file-based backup strategies to /var/lib/aolserver/service0
This section describes how to make a one-time backup and restore of the files and database. This is useful for rolling back to known-good versions of a service, such as at initial installation and just before an upgrade. First, you back up the database to a file within the file tree. Then, you back up the file tree. All of the information needed to rebuild the site, including the AOLserver config files, is then in tree for regular file system backup.
Back up the database to a file.
Oracle.
Download the backup script. Save the file export-oracle.txt as /tmp/export-oracle.txt
Login as root. The following commands will install the export script:
joeuser:~$ su -
Password: ***********
root:~# cp /tmp/export-oracle.txt /usr/sbin/export-oracle
root:~# chmod 700 /usr/sbin/export-oracleSetup the export directory; this is the directory where backups will be stored. We recommend the directory /ora8/m02/oracle-exports.
root:~# mkdir /ora8/m02/oracle-exports
root:~# chown oracle.dba /ora8/m02/oracle-exports
root:~# chmod 770 /ora8/m02/oracle-exportsNow edit /usr/sbin/export-oracle and change the SERVICE_NAME and DATABASE_PASSWORD fields to their correct values. If you want to use a directory other than /ora8/m02/oracle-exports, you also need to change the exportdir setting.
Test the export procedure by running the command:
root:~# /usr/sbin/export-oracle
mv: /ora8/m02/oracle-exports/oraexport-service_name.dmp.gz: No such file or directory
Export: Release 8.1.6.1.0 - Production on Sun Jun 11 18:07:45 2000
(c) Copyright 1999 Oracle Corporation. All rights reserved.
Connected to: Oracle8i Enterprise Edition Release 8.1.6.1.0 - Production
With the Partitioning option
JServer Release 8.1.6.0.0 - Production
Export done in US7ASCII character set and US7ASCII NCHAR character set
. exporting pre-schema procedural objects and actions
. exporting foreign function library names for user SERVICE_NAME
. exporting object type definitions for user SERVICE_NAME
About to export SERVICE_NAME's objects ...
. exporting database links
. exporting sequence numbers
. exporting cluster definitions
. about to export SERVICE_NAME's tables via Conventional Path ...
. exporting synonyms
. exporting views
. exporting stored procedures
. exporting operators
. exporting referential integrity constraints
. exporting triggers
. exporting indextypes
. exporting bitmap, functional and extensible indexes
. exporting posttables actions
. exporting snapshots
. exporting snapshot logs
. exporting job queues
. exporting refresh groups and children
. exporting dimensions
. exporting post-schema procedural objects and actions
. exporting statistics
Export terminated successfully without warnings.PostGreSQL. Create a backup file and verify that it was created and has a reasonable size (several megabytes).
[root@localhost root]# su - service0
[service0@localhost service0]$ pg_dump -f /var/lib/aolserver/service0/database-backup/before_upgrade_to_4.6.dmp service0
[service0@localhost service0]$ ls -al /var/lib/aolserver/service0/database-backup/before_upgrade_to_4.6.dmp
-rw-rw-r-x 1 service0 service0 4005995 Feb 21 18:28 /var/lib/aolserver/service0/database-backup/before_upgrade_to_4.6.dmp
[service0@localhost service0]$ exit
[root@localhost root]#
su - service0
pg_dump -f /var/lib/aolserver/service0/database-backup/before_upgrade_to_4.6.dmp openacs-dev
ls -al /var/lib/aolserver/service0/database-backup/before_upgrade_to_4.6.dmp
exitBack up the file system. Back up all of the files in the service, including the database backup file but excluding the auto-generated supervise directory, which is unneccesary and has complicated permissions.
In the tar command,
c create a new tar archive
p preserves permissions.
s preserves file sort order
j compresses the output with bz2.
The --exclude clauses skips some daemontools files that are owned by root and thus cannot be backed up by the service owner. These files are autogenerated and we don't break anything by omitting them.
The --file clause specifies the name of the output file to be generated; we manually add the correct extensions.
The last clause, /var/lib/aolserver/service0/, specifies the starting point for backup. Tar defaults to recursive backup.
[root@yourserver root]# su - service0 [service0@yourserver service0]$ tar -cpsj --exclude /var/lib/aolserver/service0/etc/daemontools/supervise --file /tmp/service0-backup.tar.bz2 /var/lib/aolserver/service0/ tar: Removing leading `/' from member names [service0@yourserver service0]$
Suffer a catastrophic failure on your production system. (We'll simulate this step)
[root@yourserver root]# svc -d /service/service0
[root@yourserver root]# mv /var/lib/aolserver/service0/ /var/lib/aolserver/service0.lost
[root@yourserver root]# rm /service/service0
rm: remove symbolic link `/service/service0'? y
[root@yourserver root]# ps -auxw | grep service0
root 1496 0.0 0.0 1312 252 ? S 16:58 0:00 supervise service0
[root@yourserver root]# kill 1496
[root@yourserver root]# ps -auxw | grep service0
[root@yourserver root]# su - postgres
[postgres@yourserver pgsql]$ dropdb service0
DROP DATABASE
[postgres@yourserver pgsql]$ dropuser service0
DROP USER
[postgres@yourserver pgsql]$ exit
logout
[root@yourserver root]#Restore the operating system and required software. You can do this with standard backup processes or by keeping copies of the install material (OS CDs, OpenACS tarball and supporting software) and repeating the install guide. Recreate the service user (service0).
Restore the OpenACS files and database backup file.
[root@yourserver root]# su - service0 [service0@yourserver service0]$ cd /var/lib/aolserver [service0@yourserver aolserver]$ tar xjf /tmp/service0-backup.tar.bz2 [service0@yourserver aolserver]$ chmod -R 775 service0 [service0@yourserver aolserver]$ chown -R service0.web service0
Restore the database
Oracle.
Set up a clean Oracle database user and tablespace (more information).
Invoke the import command
imp service0/service0 FILE=/var/lib/aolserver/service0/database-backup/nighty_backup.dmpPostgres.
Because of a bug in Postgres backup-recovery, database objects are not guaranteed to be created in the right order. To compensate, we pre-creating some critical items first, which leads to some harmless errors.
[root@yourserver root]# su - postgres
[postgres@yourserver pgsql]$ createuser service0
Shall the new user be allowed to create databases? (y/n) y
Shall the new user be allowed to create more new users? (y/n) y
CREATE USER
[service0@yourserver web]$ createdb service0
CREATE DATABASE
[service0@yourserver web]$ psql -f /var/lib/aolserver/service0/packages/acs-kernel/sql/postgresql/postgresql.sql service0
(many lines omitted)
[service0@yourserver web]$ psql service0 < /var/lib/aolserver/service0/database-backup/database-backup.dmp
(many lines omitted)
[service0@yourserver web]$ exit
[postgres@yourserver pgsql]$ exit
logout
Activate the service
[root@yourserver root]# ln -s /var/lib/aolserver/service0/etc/daemontools /service/service0 [root@yourserver root]# sleep 10 [root@yourserver root]# svgroup web /service/service0 [root@yourserver root]#
Backup can encompass all files in /var/lib/aolserver/service0. Use one cron job to back up the database to a file in /var/lib/aolserver/service0/database-backup, and a second cron job to back up the entire file tree.
Backing up the database consists of creating a file which is a picture of the database at a particular moment. Postgres can be backed up while running.
Depending on your overall backup strategy, you can create a series of database backup files (recommended for a development server where you are using the section called “Using CVS for backup-recovery”), or you can create a single nightly backup file which is then collected into a bigger backup file that includes the other parts of the service. The latter technique is more generally recommended. To make a new file every night, edit the crontab file for service0:
[service0@yourserver service0]$ export EDITOR=emacs;crontab -e
Add this line to the file. The numbers and stars at the beginning are cron columns that specify when the program should be run - in this case, whenever the minute is 0 and the hour is 1, i.e., 1:00 am every day.
0 1 * * * /usr/local/pgsql/bin/pg_dump -f /var/lib/aolserver/service0/database-backup/service0_`date +\%Y-\%m-\%d`.dmp service0
If you plan to back up the whole /var/lib/aolserver/service0 directory, then it would be redundant to keep a history of database backups. In that case, set up the cron job to overwrite the previous backup each time:
0 1 * * * /usr/local/pgsql/bin/pg_dump -f /var/lib/aolserver/service0/database-backup/service0_nightly.dmp service0
On a test service, make sure that your backup-recovery process work. After backing up the database and file system, delete the service as detailed below and then recover it.
Earlier strategies, included here because this section hasn't been fully updated yet.
Edit the backup script to save the backup file in /var/lib/aolserver/service0/database-backup. While you're working with Oracle, you should configure it to do automatic exports. An export is a separate backup copy of the database. This copy includes all of the database's state at the time that the export was initiated. If your database is corrupted, you can restore from one of these backups. You should do this step as root.
Automating backups is accomplished using the UNIX crontab facility.
While still root, run the following command. You can replace the EDITOR="emacs -nw" portion with whatever editor your prefer, such as EDITOR=vi.
root:~# export EDITOR="emacs -nw"
root:~# crontab -eNow add the following line on a line by itself
0 23 * * * /usr/sbin/export-oracleSave the file, exit the editor. Verify that the addition succeeded by checking the output of the following command.
root:~# crontab -l | grep export-oracle
0 23 * * * /usr/sbin/export-oracle
root:~# exit
; LogoutIf you see the line, go ahead and log out.
This is an alternate method to the crontabls - backup. Dowload this script to /tmp. At the top of the script are several variables that you'll need to customize:
bak - location where you want local backups to be saved
servername - name of your server (and database instance)
ftp_user - username on your ftp account
ftp_password - password on your ftp account
ftp_dir - path on the remote server where your backups will be uploaded
ftp_server - your ftp server
Next, we'll save this file to our server's tcl directory so that it will be loaded on startup. It will automatically be run every night at midnight. Note that this script only backs up the database - not the OpenACS scripts and file content.
joeuser:~$ cp /tmp/acs-pgbackup-init.txt ~/var/lib/aolserver/birdnotes/tcl/acs-pgbackup-init.tcl
joeuser:~$ restart-aolserver birdnotesThat's it! The script will email you with each successful backup (or if it fails, it will send you an email with the reason)
If you are already using CVS, you probably don't need to do anything to back up your files. Just make sure that your current work is checked into the system. You can then roll back based on date - note the current system time, down to the minute. For maximum safety, you can apply a tag to your current files. You will still need to back up your database.
Note that, if you did the CVS options in this document, the /var/lib/aolserver/service0/etc directory is not included in cvs and you may want to add it.
[root@localhost root]# su - service0
[service0@localhost service0]$ cd /var/lib/aolserver/service0
[service0@localhost service0]$ cvs commit -m "last-minute commits before upgrade to 4.6"
cvs commit: Examining .
cvs commit: Examining bin
(many lines omitted)
[service0@localhost service0]$ cvs tag before_upgrade_to_4_6
cvs server: Tagging bin
T bin/acs-4-0-publish.sh
T bin/ad-context-server.pl
(many lines omitted)
[service0@localhost service0]$ exit
[root@localhost root]#
su - service0
cd /var/lib/aolserver/service0
cvs commit -m "last-minute commits before upgrade to 4.6"
cvs tag before_upgrade_to_4_6
exitTo restore files from a cvs tag such as the one used above:
[root@localhost root]# su - service0
[service0@localhost service0]$ cd /var/lib/aolserver/service0
[service0@localhost service0]$ cvs up -r current
[service0@localhost service0]$ exit
su - service0
cd /var/lib/aolserver/service0
cvs up -r current