[PVE-User] Proxmox Fencing

[Eneko Lacunza]

Hi Alex,

>> El 5/7/21 a las 13:18, Alex K escribió:
>>> Hi all,
>>>
>>> I'm new to proxmox and trying to setup a 2 + 1 node active/active HA
>>> cluster on top glusterfs using latest community
>> pve-manager/6.4-4/337d6701
>>> (running kernel: 5.4.106-1-pve). The third node is  used for gluster
>>> arbitration and perhaps I have to configure in it a quorum disk also to
>>> keep quorum in case of a node failure (not clear yet at my mind, still
>>> reading the docs).
>> If you have 3 nodes, you want all them in Proxmox cluster for proper
>> quorum majority. No need for quorum disk that way. (note that I don't
>> know how gluster works).
>>
> Gluser has a similar concept for quorum so as to keep writes on the
> storage. Hence I am placing a third node in the setup. Due to cost
> limitations, the third node has minimal specs and is not meant to host VMs.
> It is a mini-PC thats why I did not add it as a proxmox host. I am
> wondering if it is possible to add it as a proxmox host and put a
> constraint to avoid VMs migrating into it. In this way I will achieve the
> required quorum levels without adding a full spec host.

Yes, you can create node-groups in HA groups, and add the desired nodes 
to the group. Then when adding a VM/CT to HA, configure the group there too.
>> I am stuck at the moment at the fencing part of the setup. Reading through
>>> the docs it seems that I have only the option to setup hardware watchdog
>>> fencing. I would expect to be able to use external media such as IPMI,
>>> iDrac, HP iLO or UPS based power management (APC) though I can't find any
>>> info how these are configured at current version of Proxmox.
>> Currently by default Proxmox uses a software watchdog. I'm not sure if
>> hardware watchdog support was introduced, others may help with this.
>>
> According to the docs it seems there is hardware watchdog option:
> https://pve.proxmox.com/pve-docs/chapter-ha-manager.html
> Q+++
> hardware watchdog - if not available we fall back to the linux kernel
> software watchdog (softdog)
> +++Q
Never used that, sorry.
>>> In case of a network partition and not a node hardware issue, how is the
>>> watchdog going to behave? Is a healthy but disconnected node going to be
>>> power cycled? I will soon proceed with testing as soon as I manage to
>> setup
>>> fencing though I wanted to better understand this part of fencing.
>> The node that drops out of quorum will be rebooted. If there where
>> CM/CTs configured for HA in that node, Proxmox will attempt to restart
>> them in another node.
>>
> So soft-fencing is done from ha-manager? How are the other nodes notified
> that the rebooted host is indeed rebooted so as to start the HA VMs?
There is a time delay that allows the fended node time to reboot before 
other nodes take over the HA VMs. It's like 1-2 minutes. The fenced node 
(the one out of the quorum) will reboot in max 60s.

Cheers

Eneko Lacunza
Zuzendari teknikoa | Director técnico
Binovo IT Human Project

Tel. +34 943 569 206 | https://www.binovo.es
Astigarragako Bidea, 2 - 2º izda. Oficina 10-11, 20180 Oiartzun

https://www.youtube.com/user/CANALBINOVO
https://www.linkedin.com/company/37269706/