cluster performance degradation

[kellogs]

Do you see backup task in your task window?
You can try to stop it from there

 

[ness1602]

try with qm unlock 102 or 103 on that host(pve2?)

 

[toto-ets]

in the activities window it does not exist, this problem occurred when ceph was under construction and the vm was not reachable, I would like to try to restart the node but I'm afraid, do I finish all the bkp first, or can I do something else to change the status?

 

[toto-ets]

try with qm unlock 102 or 103 on that host(pve2?)

tell me the exact command

 

[waltar]

root@pve2:# qm unlock 102 ; qm unlock 103

 

[toto-ets]

ok, it worked, now I can start the migration, thanks 1000 friend

 

[ness1602]

Just do the backups again on them,just for peace of mind.

 

[toto-ets]

yes sure, I'm doing bkp of all the vm

 

[toto-ets]

Can you suggest me the best ZFS configuration? I'm doing DAIDZ with 5 16TB HDDs, what type of compression should I use to get more performance? By default it's on, but it seems to me that lz4 is better or am I wrong? Can you suggest me?

 

[ness1602]

Use Raid10 if you want best performance, not any Raidz.

 

[toto-ets]

if I do raid 10 I have to put 6 HDDs then, but I lose 3 of the total right?

 

[ness1602]

Yeah that is true.

 

[toto-ets]

How much do I increase performance like this, from raidz to raid 10? And above all, what compression should I use? Should I leave it on by default?

 

[ness1602]

i can't give you exact number,but lets say 50% . And compression by default.

 

[toto-ets]

if it's around 50% it's worth it then losing 3 HDDs on the total, but at least I'll have more performance

 

[waltar]

I would say 10% performance win but 50% capacity win with compression which defaults to lz4.
Assume often used 10 disk combi in zfs 5 vdev 2-mirror come to 50% but with compression get up 50% from there and you are at 75% (-5% should never used)- so that's near which in raid6 gets 80% space for.

 

[ness1602]

Hardly 10% when raid5 reads are okay,but writes are the IOPS of one disk. Whereas raid10 IOPS *3

 

[waltar]

Uups, I mean 10% win with compression enabled ... not perf win from raidz to mirror

 

[toto-ets]

My intention is to have more performance at the expense of space if necessary, so which path should I take?

 

[waltar]

zfs mirror. Other pool extensions could be done later if needed too (special, log, cache).