[TUTORIAL] PBS on Harddisks only and large amount of Backups timeout in PVE.

[itNGO]

Hi there,
just a quick tip. We are using disk-based PBS-Servers where SSDs are not used.
If you are using mostly rotating hard disks with PBS and you get timeouts when listing Backups in PVE you can change Directory/File-Cache-Setting to keep the file and directory Metadata in RAM.

This is done just by adjusting "vm.vfs_cache_pressure" in sysconf or during runtime "sysctl vm.vfs_cache_pressure=10". You need to have enough ram to keep your directory/file inodes in RAM. (Example here about 5000 backups and 32GB RAM in Server with XFS is enough)

This setting will also boost Garbage Collection runtime a lot.

With "find /mnt/datastore/ -ls > /dev/null" in Shell/Terminal you can "preload" the structure or just wait until the Server has loaded it in Cache/RAM when selecting Backupstore in PVE....

This does not mean you should not use SSDs in your PBS, but not everyone has that value to spend. Often Backupservers are created from "old" Systems which are replaced by newer ones.... ;-)

 

[enoch85]

adjusting "vm.vfs_cache_pressure" in sysconf or during runtime

Sorry, but does "during runtime" mean that I set it with a script before every backup, or as a general setting on the PBS server?

I have lots of timeouts, and need to reboot every now and then. Quite annoying tbh.

 

[itNGO]

Sorry, but does "during runtime" mean that I set it with a script before every backup, or as a general setting on the PBS server?

I have lots of timeouts, and need to reboot every now and then. Quite annoying tbh.

This is for speeding up Garbage Collection and Backup Listing, not the Backup-Job itself.
It is a Config-File-Setting which you only do once, but the preloading has to be done after a reboot to get the cache filed up fast. Can also be a CRON-Job...

 

[enoch85]

Ok, so someting like this would work?

if grep 'vm.vfs_cache_pressure' /etc/sysctl.conf then echo "VFS Cache already set, skipping..." else sysctl -w vm.vfs_cache_pressure=50 { echo "###################################################################" echo "# Tune PBS Garbage Collection" echo "vm.vfs_cache_pressure=50" } >> /etc/sysctl.conf fi

 

[itNGO]

Ok, so someting like this would work?

if grep 'vm.vfs_cache_pressure' /etc/sysctl.conf then echo "VFS Cache already set, skipping..." else sysctl -w vm.vfs_cache_pressure=50 { echo "###################################################################" echo "# Tune PBS Garbage Collection" echo "vm.vfs_cache_pressure=50" } >> /etc/sysctl.conf fi

More or less yes, but set "sysctl vm.vfs_cache_pressure=10" to have a good chance that cache will not be cleared from other services....