[SOLVED] Prevent node fencing while updating corosync config

[budy]

Hi,

I am about to manually change the corosync config on my PVE cluster to introduce a 2ng Ring-Interface. I have read up on how to do that and although I am pretty sure, I got the config right, I was wondering if I could somehow prevent my nodes to be fenced, should I have messed up the new config nevertheless. Since HA is nothing you can simply switch off, I was thinking of removing all HA resources using the API. Would that work or is there another way of temporarily disabling fencing?

Thanks,
budy

 

[budy]

Just read on some other thread, that once all HA resources are deleted, it's necessary to restart the lrm daemon on each node like this…

systemctl restart pve-ha-lrm

We'll see, how it goes…

 

[t.lamprecht]

Hi,

it's actually saver to just stop the HA daemons - which puts the HA resources into the freeze state and release the watchdog (responsible for resets).

The order in which you stop them matters though:

1. Stop all LRM services systemctl stop pve-ha-lrm.service (or use the webinterface to stop them)
2. Wait untill the LRMs are stopped and all services (which aren't in ignore state) are frozen
3. Stop only then all CRM services systemctl stop pve-ha-crm.service (or use the webinterface to stop them)

Now the watchdog is released and you can do cluster maintenance without it interfering.

 

[t.lamprecht]

that once all HA resources are deleted, it's necessary to restart the lrm daemon on each node like this…

Yes, for the removal of resources this is required.

You could actually "remove" them by:
mv /etc/pve/ha/resources.cfg /etc/pve/ha/resources.cfg.tmp
then restart all LRMs

After maintenance was done you could again re-add them with: mv /etc/pve/ha/resources.cfg.tmp /etc/pve/ha/resources.cfg
(no restart required for that)

 

[budy]

Worked like a charm - thank you!