[SOLVED] How to change mount options for drives that are not in fstab?

[C.G.B. Spender]

Hi I'm running Proxmox 6.0-11 in my homelab, I've been running into issues with high io wait and low fsyncs, after some googling I managed to sort it on / by adding barrier=0 to it's mount options, however I have 2 more ext4 formatted ssd drives connected (1 sata 1 usb) which do not appear in /etc/fstab and 1 of them still exhibit low fsyncs. The sata one is used as vm storage, the usb is a for daily backup. I understand the implications of the barrier=0 option, the server is backed by ups and in the end it's just a "playground" no data I would miss on it.

/ is a supermicro sata dom 32gb drive connected to internal sata3 dom port
/mnt/pve/vm-backup is sandisk x110 ssd in usb3 external enclosure
/mnt/pve/vm-storage is crucial mx300 ssd connected to internal sata3 port

/etc/fstab

root@pve:~# cat /etc/fstab
# <file system> <mount point> <type> <options> <dump> <pass>
/dev/pve/root / ext4 barrier=0,errors=remount-ro 0 1
UUID=E6D0-BE2A /boot/efi vfat defaults 0 1
/dev/pve/swap none swap sw 0 0
proc /proc proc defaults 0 0

relevant lines from mount output

root@pve:~# mount | grep ext4
/dev/mapper/pve-root on / type ext4 (rw,relatime,nobarrier,errors=remount-ro)
/dev/sdd1 on /mnt/pve/vm-backup type ext4 (rw,relatime)
/dev/sdb1 on /mnt/pve/vm-storage type ext4 (rw,relatime)

the pveperf tests were done on idle server with all vm's and containers shut down.

pveperf /

root@pve:~# pveperf /
CPU BOGOMIPS:      24000.00
REGEX/SECOND:      3750826
HD SIZE:           7.07 GB (/dev/mapper/pve-root)
BUFFERED READS:    456.85 MB/sec
AVERAGE SEEK TIME: 0.17 ms
FSYNCS/SECOND:     2672.89
DNS EXT:           40.82 ms
DNS INT:           2.79 ms (lan)

pveperf /mnt/pve/vm-backup

root@pve:~# pveperf /mnt/pve/vm-backup
CPU BOGOMIPS:      24000.00
REGEX/SECOND:      3742463
HD SIZE:           233.73 GB (/dev/sdd1)
BUFFERED READS:    363.87 MB/sec
AVERAGE SEEK TIME: 0.31 ms
FSYNCS/SECOND:     2402.12
DNS EXT:           33.57 ms
DNS INT:           3.81 ms (lan)

pveperf /mnt/pve/vm-storage

root@pve:~# pveperf /mnt/pve/vm-storage
CPU BOGOMIPS:      24000.00
REGEX/SECOND:      3744982
HD SIZE:           686.67 GB (/dev/sdb1)
BUFFERED READS:    431.09 MB/sec
AVERAGE SEEK TIME: 0.10 ms
FSYNCS/SECOND:     437.49
DNS EXT:           39.03 ms
DNS INT:           3.94 ms (lan)

Before adding the barrier=0 to fstab for / fsyncs on / were around 500, similar to what my vm-storage mount shows. However what confuses me is how come both the vm-storage and vm-backup mounts are mounted with same options, but the usb drive shows 6x more fsyncs/sec than the internal ssd.

I tried adding the drive to /etc/fstab manually both by uuid or by /dev/sd* and rebooted the server, but they were still mounted with the same mount options.

So now my questions are:
If I want to change mount options for the drives that are not in /etc/fstab where would I do so?
Are the low fsyncs on the internal ssd drive due to the model of the drive or due to the barrier=0 option not being there?

Thanks for any input!

 

[wolfgang]

Hi,

/ by adding barrier=0

This is dangerous and can corrupt your FS in the worst case.
In case that you ssd you should use mount options like discard and atime.

If you provision a storage with the PVE GUI the mount is done by systemd.
/etc/systemd/system/mnt-pve-<storagename>

 

[C.G.B. Spender]

Thanks,
I added barrier=0 to relevant file in /etc/systemd/system/ and my fsyncs on vm-storage skyrocketed 10times (while running all the vms and containers):

root@pve:~# pveperf /mnt/pve/vm-storage/
CPU BOGOMIPS:      24000.00
REGEX/SECOND:      3116234
HD SIZE:           686.67 GB (/dev/sdb1)
BUFFERED READS:    405.06 MB/sec
AVERAGE SEEK TIME: 0.11 ms
FSYNCS/SECOND:     5325.62
DNS EXT:           37.06 ms
DNS INT:           2.60 ms (lan)

I'm a layman when it comes to my homelab / linux etc. All I know back on older proxmox on ext3 with same hardware I had good fsyncs then the defaults for mounts in later kernel changed or whatnot and barrier became default and my fsyncs tanked. I googled around, but as with most information related to linux I came up with obscure overly technical explanation that left me more confused than before. I might misunderstand the problematic, but to me it seemed all it can cause is file corruption on power loss, which is a nonissue for me.

Bottom line is IT and my homelab are my hobby not my job, and in my playground I prefer performance over enterprise-grade data protection. Worst case scenario I restore the daily backup and lose 1 day worth of graylog and pihole statistics.