[PVE-User] Very high I/O-delay while migrating LXC

[Falco Kleinschmidt]

Dear PVE-List!

In the last days I migrated (Restart Mode) several LXC containers (from
small 10GB to big 400GB) from node to node, to get one node free for
testing microcode updates.

The destination node has a local-LVM on a RAID 10 consisting of 4 SSDs
which perform well.

The cluster (not HA) is running PVE 5.1-43 using "pve-no-subscription.list"

While migrating, the I/O-delay rises to very high values over 80% on the
destination node. Other containers and VMs on the destination node are
affected heavily from this and are nearly unreachable or even not
responding at all while migrating the last two times.

After the last migration finished, I realized that one VM with normally
very low workload was using 100% of one of two cores and was still not
responding, although the migration had finished. I tried to "Reset" the
VM without success. Then I tried to "Stop" it with success. This
produced a unbootable bootdisk and I had to restore the backup of the VM.

Is this high I/O-delay a known issue? Do I have a faulty configuration?

Does somebody has an idea how to reduce the I/O-delay?

Perhaps reducing the allowed network traffic? The traffic rises up to
100-120 MB/s - which is of course good, regarding the time used for
migrations.

Thank you for any advice!

Kind regards!
Falco

-- 

Datenfalke - Dipl. Inf. Falco Kleinschmidt
Adresse: Dinnendahlstr. 8 - 45136 Essen
Steuer-Nr: DE248267798
Telefon: +49-(0)201-6124650
Fax: +49-(0)201-6124651
Email: fk at datenfalke.de
WWW: http://www.datenfalke.de