Forgot your password?
Views
TapeBackupProcedures
Backups at the HPCf
Abstract
This document describes the backup procedures for user and system data at the HPCf. It includes an overview of the backup process, a description of the software in use, installation and operation instructions. I also describe the physical storage of the backup tapes.
See Also
ManualTapeHandling - for information on driving the tape robot and drive manually (also usefull for restoring amanda backups).
Introduction
The HPCf currently houses approximately 400GB disk space, with plans
to triple this capacity in the near future. This document seeks to
explain how to back up this data, to ensure it's availability for
users. We use the Advanced Maryland Automatic Network Disk Archiver
(amanda, http://www.amanda.org/) to backup our entire
network. Amanda is a client server package, the server runs on
boreas.hpcf.upr.edu, clients run on each machine to be backed
up. Amanda drives an 8 tape DLT stacker, each tape has a capacity of
35GB uncompressed. On each client, amanda uses the native
dump program to perform backups, these are then compressed
and sent over the network to boreas, which spools several backups into
it's holding disk (/BACKUP/amanda1) before sending the
backups to tape.
Obtaining the software
Tape server software
Amanda is a client server package, the server software is installed onto the machine that hosts the tape drive. At the HPCf this is currently boreas.hpcf.upr.edu. The 8 tape DLT stacker used at the HPCf is driven by the chg-scsi program in amanda. This program will not compile on IRIX systems in the official release of amanda. Instead, we use the beta version of amanda on boreas. Beta 1 of amanda available from sourceforge
Client software
We use amanda version 2.4.1p1 on the client stations, as the beta 1 has some problems on sgi machines (it leaves xfsdump processes running on dartagnan). Amanda for IRIX is available on the sgi freeware CD. I installed the package fw_amanda.sw_amanda, which expects a user called amanda in the password file.
On the linux boxes, I've installed VA Research's amanda RPMs?, as the ones I found for redhat didn't include dump support. In addition, there seems to be a bug in the dump package included with redhat 6.0. it dumps core backing up the /home filesystem on astraeus. I've upgraded dump there to version dump-0.4b15-1. The VA Research rpms are available from
ftp://ftp.valinux.com/pub/software/VALinux/6.1.3/i386/RedHat/RPMS/
NOTE It seems like the va linux rpms no longer have dump support. I downloaded the amanda source rpm from valinux and built it on a machine that has dump installed. These rpms are available from:
http://hpcfinfo.hpcf.upr.edu/amanda/
RedHat? 7.1 can use the dump, gnuplot, and amanda rpms included with the system.
Install dump, amanda and amanda-client on the linux machine you wish to backup. You need to create an .amandahosts file in the amanda user home directory, containing "boreas.hpcf.upr.edu backup", this file must be readable by group disk, and the amanda home directory must allow access by group disk as well.
RedHat? 7 installed amanda user with home in /root, Redhat 7.1 puts amanda's home in /var/lib/amanda, please change the following accordingly.
# chgrp disk /root # chmod 710 /root # vi /root/.amandahosts # chgrp disk /root/.amandahosts # chmod 640 /root/.amandahosts
You should create /etc/dumpdates, and make it writable by group disk. Any partitions you wish to backup must be readable by group disk.
Make sure you have an amanda line in your /etc/services file:
amanda 10080/udp # amanda backup services
To allow the amanda daemon to run, add an entry to inetd (or xinetd), here's the one I use on astraeus:
#Amanda Client amanda dgram udp wait operator.disk /usr/sbin/tcpd /usr/lib/amanda/amandad
After adding the line to /etc/inetd.conf, reload the inetd file (/etc/rc.d/init.d/inet reload)
RedHat? 7 uses xinetd instead of inetd, and the rpms of amanda seem to be broken.
You will have to create /etc/xinetd.d/amandad by hand
# default: off # # description: Part of the Amanda server packageservice amandad { disable = no socket_type = dgram protocol = udp wait = yes user = operator group = disk server = /usr/lib/amanda/amandad port = 10080 }
RedHat? 7.1 amanda xinetd file is correct, just remember to chkconfig --level 345 amanda on
Configuration
Server configuration
I configured amanda on boreas according to the instructions in the source package (see: /usr/people/backup/amanda-2.4.1p1/docs/INSTALL ). We've created a user called backup that owns the amanda packages. The configured source code for amanda is in /usr/people/backup/amanda-2.4.1p1. The software was configured with the command:
# ./configure --with-user=backup --with-group=backup --with-amandahosts
Adding a disk
To add a disk to the backup rotation on a machine that already has the backup software installed you need to add the disk to the disklist file on boreas, and ensure the backup user has read access.
To add the disk to the disklist, login as backup on boreas
$ cd /usr/local/etc/amanda/HPCf $ vi disklist
Add a line like
machine /partition {dumptype}
to the disklist file, I usually use the mount path instead of the partition (easyer to remember), and use comp-user or comp-root as the dumptype.
Here's the actual line for dartagnan:/usr/people partition:
dartagnan /usr/people comp-user
To allow the backup user to access the drive, on linux simply make the device mode 0640, owner root, group disk (the backup user is already in the disk group).
$ df -h Filesystem Size Used Avail Use% Mounted on /dev/sda3 8.0G 2.5G 5.1G 33% / /dev/sda1 125M 5.3M 113M 5% /boot /dev/sdb1 8.4G 6.3G 1.6G 79% /home$ ls -l /dev/sda3 brw-r----- 1 root disk 8, 3 May 5 1998 /dev/sda3
On the sgi's, the devices get remade every time the machine reboots, so you need to chmod the raw device for the partition to 0640 root.sys, and add a line to /etc/ioperms to do this on reboot.
% df -h Filesystem Type Size use avail %use Mounted on /dev/root xfs 17G 5.4G 11G 33% / /dev/dsk/dks0d2s7 xfs 17G 9.7G 7.2G 58% /usr/people% cat /etc/ioperms /hw/rdisk/root 0640 root sys /dev/rdsk/dks0d2s7 0640 root sys
% ls -l /dev/rdsk/dks0d2s7 crw-r----- 1 root sys 0,236 Jan 17 15:11 /dev/rdsk/dks0d2s7
Machines running sgi propak 1.3 use devfs, and also need to make sure device permissions are saved and restored correctly.
Modify /etc/devfs.conf, look for the part that reads:
# # Uncomment this if you want permissions to be saved and restored for the main # devfs (usually mounted on /dev) REGISTER . COPY /dev-state/$devname $devpath CHANGE . COPY $devpath /dev-state/$devname CREATE .* COPY $devpath /dev-state/$devname
Then reboot, now make changes to the device permissions so that group disk can read the partitions:
# ls -l /dev/hda1 lr-xr-xr-x 1 root root 33 Jun 11 07:39 /dev/hda1 -> ide/host0/bus0/target0/lun0/part1 # chgrp disk /dev/ide/host0/bus0/target0/lun0/part1 # chmod g+r /dev/ide/host0/bus0/target0/lun0/part1 # ls -l /dev/ide/host0/bus0/target0/lun0/part1 brw-r----- 1 root disk 3, 1 Dec 31 1969 /dev/ide/host0/bus0/target0/lun0/part1
As the backup user on boreas, check to make sure the permisions are OK by executing
$ amcheck -c HPCf
this will alert you if any problems are found.
-- Main.HumbertoOrtiz - 07 Dec 2000
Updates to installation on redhat 7 and adding disks to devfs equipped machines.
-- Main.HumbertoOrtiz - 20 Jun 2001
Updated for RedHat? 7.1
-- Main.HumbertoOrtiz - 09 Nov 2001
