Server Downing From Transfer

[infinityM]

Hey Guys,

I'm hoping for some advise.
I've got a cluster of 5 servers. Each has 1 10GB SFP+ port and 2 1Gb network ports connected.

The ip address ranges for each differ, i.e.
10G = 10.0.0.2-7
1G = 10.0.1.2-7
1G = 10.0.2.2-7

And i've setup corosync to use the last 1G port range while I use the 10.0.1.x range to do backups etc.
The only issue i've now noticed that if I try to transfer a server from 10.0.0.7 to any other server, that server then becomes unavailable and restarts itself...
Why would this be?
What should I correct?

 

[ph0x]

The migration probably happens over the wrong subnet, then.
Did you set the migration network under Datacenter->Options?

 

[infinityM]

The migration probably happens over the wrong subnet, then.
Did you set the migration network under Datacenter->Options?

I did not even know about that option thank you!.

So now the question becomes. Which IP rangedo I need to avoid sending the traffic over?
Meaning, If I need to avoid using the corosync range, or if I need to avoid the lice range or what would you recommend?

 

[ph0x]

With the three options you provided I would go for the 10G network, maybe within a separate VLAN.

 

[infinityM]

but the 10G is the main network. I would then rather move the 10G network to an internal range and the 1G network to the live IP facing ranges .

But the 1 question still remains. What's causing the reboot?
Was it because the traffic was on the live IP range? Or was something else causing it? because corosync does have it's own IP range...

 

[hepo]

Maybe watchdog functionality, if the network is saturate and corosync cannot operate correctly, the watchdog will kick in (automatic reboot when the watchdog counter runs out). Stop the network interface on which corosync operates, few seconds later the server will reboot.
The best practice is to separate corosync on it's own network.

 

[infinityM]

Maybe watchdog functionality, if the network is saturate and corosync cannot operate correctly, the watchdog will kick in (automatic reboot when the watchdog counter runs out). Stop the network interface on which corosync operates, few seconds later the server will reboot.
The best practice is to separate corosync on it's own network.

The Corosync is on the same switch, but it's own network range? I assume that's not an issue?

 

[hepo]

https://pve.proxmox.com/wiki/Separate_Cluster_Network

 

[ph0x]

The Corosync is on the same switch, but it's own network range? I assume that's not an issue?

So an own VLAN, I guess. This should be unaffected by high traffic in other subnets, then. Unless the switch reaches its limit but I tend do doubt that.