Why does writing to disk in kvm fail?

Question

KVM

KVM guest on ubuntu 14.04 is on the latest proxmox with virtio options cache tried none and writeback
there is a dull disk on the postgres replica
in iotop such a situation
Total DISK READ : 634.92 K/s | Total DISK WRITE : 2.96 M/s
Actual DISK READ: 634.92 K/s | Actual DISK WRITE: 1670.63 K/s
TID PRIO USER DISK READ DISK WRITE SWAPIN IO> COMMAND
1095 be/4 postgres 0.00 B/s 1670.63 K/s 0.00% 89.72% postgres: wal receiver process streaming 20D5/2EB71140 Disk
usage 89.72%
maximum 3 megabytes per second
ssd disks
have exactly the same replica on the same server without kvma where the disk load is near zero.
thin allocation of space has nothing to do with it, for the test I scored the disk with zeros to the eyeballs.
What could be the reason for such strange behavior?

Answer 1

In the guest itself, you need to install the virtio drivers.
And on the hypervisor, look at the HDD settings in the BIOS, or configure the RAID controller (if any), write caching. As an option, choose a file system adapted for SSD on the hypervisor, but this is troublesome. You may need to increase the RAM on the guest and hypervisor. Disable all unnecessary services in the guest, including the graphical interface. On the guest, increase the SWAP size.

Answer 2

And on what device the disks are hooked?
If on virtio-blk, then you can try to upload it to virtio-scsi. Maybe trim for the ssdshnik does not work and litters the speed? virtio-scsi supports it, unlike virtio-blk.
Is a separate thread dedicated to io, and what type is used: threads or native? You can also play around with the number of hardware io queues.
Well, at least try vhost-scsi. With him, according to tests, it seems, some space iops can be untwisted.