VM Migration time very long

[_seb_]

Hello team,

I have 3 servers with PVE 7.2-3 in CEPH mode.

I notice that the migration time of a VM is very long when I cut the power supply of the PVE server hosting the VM.
The VM is a simple Debian 10 x64 (10G disk storage and 2G of Ram).

Here is the time when I cut the power supply of the PVE server before the VM fails over to another PVE host:

Ping OK
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping FAIL
Ping OK

Is this normal?

Thank you

 

[fabian]

yes, it takes 60s for a node to self-fence, and then some time for recovery on a still available node. the other nodes have to wait since they can't know whether the node is just not reachable from their end, or already gone.

 

[_seb_]

Thank Fabian !

If I understood correctly there is no solution to shorten the delay because this is the normal behavior ?

 

[fabian]

yes, it is required (lowering it would be possible in theory, but that would increase the number of false positive fencing events which has even bigger downsides)

 

[_seb_]

Ok thanks Fabian it's clear