Ceph, proxmox4, monitor load

[Melanxolik]

Hi, I have problem with performance ceph. One of many customers create load on ceph storage, at LVM storage I can run: iostat -x 1 -N and search LVM drive with need me problem customer, but How can I control problem client at ceph storage?
I need control iops and mbps/rw per customer.

I know about disk throttle, but I need monitor disks performance.

Could you help me with this?

 

[Q-wulf]

In ceph you have the following command:
"ceph -w" it tells you in real time whats happening on your cluster with regards to reads/writes and iops. However that does not tell you where its coming from.

I'd go via "datacenter > node > vm > summary" and have a look at Disk IO for the time period in question, to figure out who it is.

I am not aware how you'd QOS in Ceph on a pool basis. I'd use the "Disk throttle" function to rate-limit the customers vDisk via proxmox/rdb. Never comes up in our use-case, so i can not give you any gurantees on that.

ps.: if it turns out its a general issue of your ceph-cluster (unable to handle iops from multiple customers) you might wanna look into
SSD / M.2 NVME based Cache-Tiers: http://docs.ceph.com/docs/master/rados/operations/cache-tiering/

can tell you from experience they truly work wonders.