qmrestore and high load

[tincboy]

Yesterday I was restoring a 100 GB tar file backup ( which was created by vzdump for a KVM guest ) but the load of server increased to 20 or 30 and other live VMs on that host was inaccessible in the period of restoring.
I've used this command to find out the process number:
ps -aux | grep qmrestore
and then this command to reduce the IO pressure on the server:
ionice -c3 -pxxxxx

but it didn't fix anything.

Any suggestion about reducing the IO pressure of qmrestore?

 

[tincboy]

Also I've notified that the IO scheduler of my HDD drives are deadline
cat /sys/block/[sh]d[a-z]*/queue/scheduler
noop anticipatory [deadline] cfq
noop anticipatory [deadline] cfq

Does it mean that the ionice will not work?

 

[dietmar]

Does it mean that the ionice will not work?

Yes. Linux supports io scheduling priorities only with the CFQ io scheduler.

 

[tincboy]

If it require CFQ then why vzdump uses ionice -c2 -n7 while deadline is the default io scheduler in proxmox?

 

[dietmar]

while deadline is the default io scheduler in proxmox?

cfq is the default in proxmox.

 

[tincboy]

Isn't it new ( does proxmox 1.5 use cfq as default scheduler ? ) ? I didn't change the scheduler but the cat /sys/block/[sh]d[a-z]*/queue/scheduler returns deadline as current scheduler.

 

[tincboy]

I've 6 proxmox servers and all of them are using deadline as default scheduler and I didn't made any change in scheduler configurations.

 

[dietmar]

I've 6 proxmox servers and all of them are using deadline as default scheduler and I didn't made any change in scheduler configurations.

What is the output of

# pveversion -v

 

[dietmar]

I've 6 proxmox servers and all of them are using deadline as default scheduler and I didn't made any change in scheduler configurations.

My guess is that you use an older 2.6.32 kernel - please update.

 

[tincboy]

yes my kernel is 2.6.32-2-pve ,
Does 2.6.35 support CFG as default scheduler?

 

[dietmar]

yes my kernel is 2.6.32-2-pve ,
Does 2.6.35 support CFG as default scheduler?

yes

 

[cks]

Hello

my host is an old one upgraded from 1.6 (itself upgraded from 1.5) :

# pveversion -v
pve-manager: 1.7-10 (pve-manager/1.7/5323)
running kernel: 2.6.18-4-pve
proxmox-ve-2.6.18: 1.7-10
pve-kernel-2.6.18-4-pve: 2.6.18-10
qemu-server: 1.1-25
pve-firmware: 1.0-9
libpve-storage-perl: 1.0-16
vncterm: 0.9-2
vzctl: 3.0.24-1pve4
vzdump: 1.2-9
vzprocps: 2.0.11-1dso2
vzquota: 3.0.11-1dso1
pve-qemu-kvm-2.6.18: 0.9.1-10

I guess ionice has no effect for me since :

# cat /sys/block/[sh]d[a-z]*/queue/scheduler
noop anticipatory deadline [cfq]

if i'm right, will ionice work with 2.6.18 kernel if I force another scheduler ? if yes can you explain how to do that ?

thanks

 

[dietmar]

I guess ionice has no effect for me since

Why (you already use 'cfq' scheduler).

 

[cks]

Why (you already use 'cfq' scheduler).

Because OP stated that "deadline" mean "ionice to not work"... but if I misunderstood and if you confirm that ionice is active for me, then all's fine

 

[mmenaz]

The output of:
# cat /sys/block/[sh]d[a-z]*/queue/scheduler
noop anticipatory deadline [cfq]

as far as I understand, lists the schedulers your kernel is capable of, and, between square brackets, the one is currently active, so [cfg] is the current onr

 

[cks]

The output of:
# cat /sys/block/[sh]d[a-z]*/queue/scheduler
noop anticipatory deadline [cfq]

as far as I understand, lists the schedulers your kernel is capable of, and, between square brackets, the one is currently active, so [cfg] is the current onr

Ok, makes sense.... sorry to have bother you with this, I should have read doc about /sys/block/[sh]d[a-z]*/queue/scheduler content by myself before asking