Proxmox fencing

[koniambo]

Hi everyone,

Got some interrogation about fencing in proxmox.

I wanna add fencing to my cluster I saw that there is 2 way which one is the best : Hardware (ILO 6) or software ?

what do we have to use ?

- fencing-pve
- It seems that watchdog is unusable
- whatcdog-mux

How to setup fencing in proxmox ? In the docs, /etc/default/redhat-cluster-pve is the way but it doesnt exist anymore

As you see may see im very new to proxmox, thx for any help

 

[aaron]

Configuring hardware watchdogs that the server provides can lead to unexpected reboots, depending on how good that is implemented. We did have some support tickets where the HW watchdogs caused unexpected reboots.

In our experience, the default software watchdog of the Linux kernel works well. Keep in mind, this is used in situations where the host is running HA guests but lost the cluster connection (Corosync) to the (hopefully) still running rest of the cluster. Letting it expire after a grace period is done to make sure that the HA guests are definitely powered off, before the remaining cluster will start these guests.

How to setup fencing in proxmox ? In the docs, /etc/default/redhat-cluster-pve is the way but it doesnt exist anymore

Please check at the beginning of documents if there is a note regarding this information being out of date!

To enable fencing on nodes that have HA guests running on them, you don't need to do anything at all as by default the software watchdog of the Linux kernel is used.

I hope that clears up some of the confusion

 

[koniambo]

Thx for your fast reply !

To enable fencing on nodes that have HA guests running on them, you don't need to do anything at all as by default the software watchdog of the Linux kernel is used.

Is softdog related to watchdog-mux service ?
If i uderstand well with watchdog-mux, a node that leave the qorum will be shooting down due to fencing ? or corosync did it by himself ?

 

[aaron]

the watchdog-mux is there to make it possible that more than one service can interact with the underlying watchdog device of the Linux kernel. See https://git.proxmox.com/?p=pve-ha-m...a85455173299ae4aeb02e56352bed7609;hb=HEAD#l42 for a more detailed reasoning.
A host that cannot establish a connection to the quorate part of a cluster will renew the watchdog for 1 minute (grace period), and then will let the watchdog expire. That renewal is handled by the pve-ha-lrm service which will enable, renew and stop it when stopped gracefully.

 

[koniambo]

Ok I got it now

 

[spirit]

Configuring hardware watchdogs that the server provides can lead to unexpected reboots, depending on how good that is implemented. We did have some support tickets where the HW watchdogs caused unexpected reboots.

I can confirm that I had tried with Dell idrac some year ago , and I had a lot of bug with unexpected reboot. Never had problem with softdog.
(The only benefit was to have log of the watchdog in the idrac log)