[ovirt-users] question regarding fencing proxies

[Martin Perina]

Hi,

engine executes power management (fencing) operation using different host
in the same cluster/DC as target host reside. It's an oVirt requirement
that each host in the same cluster and data center is able to connect to
power management deviced of all other hosts in the same cluster/DC.
​

Here's simplified flow for defaults (fence proxy may be selected in the
same cluster or DC)

1. Find all existing hosts in the same cluster​ which don't have connection
issues (status Up, Maintenance, NonOperational for a reason different than
network issues)
2. Select first host in Up (if no host is Up, then pick any other host, if
no host found continue with DC in step 5) and try to execute fence operation
3. If fence operation is successful, exit
4. If not, retry 2 times and if still not successful go back to step 2 but
exclude already problematic fence proxy
5. Find all existing hosts in the same data center​ which don't have
connection issues (status Up, Maintenance, NonOperational for a reason
different than network issues)
6. Select first host in Up (if no host is Up, then pick any other host, if
no host found then fail and exit) and try to execute fence operation
7. If fence operation is successful, exit
8. If not, retry 2 times and if still not successful go back to step 6 but
exclude already problematic fence proxy
9. If still no success, fail

As you see your setup with only 2 hosts is kinda fragile (and not only for
fencing operations), you need to assure that 1 hosts is always in perfect
condition. If possible I'd recommend you to add at least one other host if
possible to make your setup more resistant to failures.

Martin Perina


On Fri, Jan 13, 2017 at 6:10 PM, cmc <iucounu at gmail.com> wrote:

> Hi,
>
> Can someone tell me how the engine decides which power management
> proxy/proxies to use (using default cluster/dc config)? I am using
> drac 7 for a fence agent in my two host cluster, and have noticed that
> one of the hosts cannot contact the drac. My guess is that the engine
> is using one host to as a power management proxy and hosts cannot
> reach their own drac as they are on the same interface + vlan.
>
> Example scenario:
>
> Engine uses host 2 as power management proxy. It can contact host 1’s
> drac, but cannot contact its own drac. In the case of host 2 being
> unreachable/kdumping etc, would the engine switch to use host 1 as the
> proxy to contact host 2’s drac?
>
> Thanks,
>
> Cam
>
> PS: I'd like to use the APC as an additional fencing agent, each host
> has two PSUs connected to two different APCs. Is there a guide on how
> to specify two ports on two different PDUs to control power on a host?
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20170114/1ece391d/attachment.html>