[slurm-users] Database backup best practices
Ole Holm Nielsen
Ole.H.Nielsen at fysik.dtu.dk
Thu Dec 10 16:05:09 UTC 2020
Backing up the Slurm database is important for disaster recovery or
migration of the database, but this is one of those Slurm tasks which
isn't well documented. I would like to find out some best practices, and
I'm sure there's a lot of experience on this in the Slurm community.
I've been working a bit on the database backup, but I'm not blessed with a
deep knowledge of databases. Anyway, here's my proposal which I hope
people would comment on:
Our setup is a slurmdbd server running on CentOS 7.9 with a MariaDB 5.5
database from the distro.
Propositions:
1. Backup using the mysqldump command. Only the database named
slurm_acct_db should be backed up. Don't back up the other system
databases with, e.g., '--all-databases', see ref. [1].
2. Since the Slurm database uses InnoDB tables, the mysqldump command
should include '--single-transaction' to ensure a consistent data dump and
not locking the database, see refs. [1] and [2].
3. A number of daily backups should be made and rotated. The logrotate
command is an ideal tool for this.
My solution in ref. [3] is the following logrotate script
/etc/logrotate.d/slurm_acct_db_backup which will make 8 backup versions in
/var/log/mariadb/:
/var/log/mariadb/slurm_acct_db_backup.bz2 {
daily
dateext
dateyesterday
rotate 8
nocompress
create 640 root adm
postrotate
/usr/bin/mysqldump --user=root --password=<DATABASE-PASSWORD>
--single-transaction -B slurm_acct_db | bzip2 > /root/slurm_acct_db_backup.bz2
endscript
}
You must first create an empty backup file:
touch /var/log/mariadb/slurm_acct_db_backup.bz2
Some further details are given in ref. [3], and this setup seems to be
working well.
Question: Are there better ways, or can you suggest improvements to my
approach?
Thanks,
Ole
References:
1. https://bugs.schedmd.com/show_bug.cgi?id=10295
2. https://serversforhackers.com/c/mysqldump-with-modern-mysql
3.
https://wiki.fysik.dtu.dk/niflheim/Slurm_database#backup-and-restore-of-database
--
Ole Holm Nielsen
PhD, Senior HPC Officer
Department of Physics, Technical University of Denmark
More information about the slurm-users
mailing list