rgmanager running per cli but not pve

[e100]

The fence_na.log seems very similar to the issue in this thread: http://forum.proxmox.com/archive/index.php/t-8637.html

 

[dietmar]

What is the output of

# fence_tool ls

I observer exactly that behavior when fencing is in action, and fencing agent returns errors

# grep fencing /var/log/syslog

 

[RobFantini]

here are results from 2 of the nodes:

fbc240 s009 ~ # fence_tool ls
fence domain
member count  4
victim count  0
victim now    0
master nodeid 1
wait state    none
members       1 3 4 5 

fbc240 s009 ~ # grep fencing /var/log/syslog 
fbc240 s009 ~ # 


fbc240 s009 ~ # fence_tool ls
fence domain
member count  4
victim count  0
victim now    0
master nodeid 1
wait state    none
members       1 3 4 5 

fbc240 s009 ~ # grep fencing /var/log/syslog 
fbc240 s009 ~ #

 

[RobFantini]

Hi Dietmar

In case there is not a software update for rgmanager coming soon, do you have a suggestion to fix this problem?

 

[dietmar]

In case there is not a software update for rgmanager coming soon, do you have a suggestion to fix this problem?

I will try to upload an update to pvetest next week.

 

[RobFantini]

I will try to upload an update to pvetest next week.

Thank you for the information.

 

[RobFantini]

Hi Dietmar
Does it look like the updated rgmanager will get to pvetest sometime in the next week or so? If not I'll try shutting down all the hosts and restart to see if rgmanager works in pve.

I know there may be other priorities .
thanks
Rob

 

[dietmar]

Sorry, there are still open issues with the new version.

 

[RobFantini]

this morning I tried turning off all nodes, then starting one at a time.

that did not fix the issue.

rgmanager is running per cli, but not per pve > datacenter > summary .

any suggestions to fix this so we can have high availability kvm's ?

 

[dietmar]

this morning I tried turning off all nodes, then starting one at a time.

And you set expexted votes after each node start (else you do not have quorum and rgmanager will not start)?

 

[RobFantini]

And you set expexted votes after each node start (else you do not have quorum and rgmanager will not start)?

No I have not set that. Will do some research on it. I mostly tried to follow wiki and forum suggestions and must have missed that.

 

[RobFantini]

And you set expexted votes after each node start (else you do not have quorum and rgmanager will not start)?

I'd like advice on what I may do , which is adding this to our cluster conf file

<cman expected_votes="2"/>

as we have a 3-node cluster .

 

[RobFantini]

I added

expected_votes="2"

to cluster.conf and activated.

rgmanager is running per cli and the services tab for each node.

but in data center summary it is not listed.

It would be nice to get HA cluster working. We have drbd but can not use heartbeat in version 2 . Heartbeat and drbd were our high available set up
prior to this for 4 years..

I'll start another thread regarding manual switch over.

 

[dietmar]

I added

expected_votes="2"

to cluster.conf and activated.

No, please remove that. You completely miss-understood me.

I you cold start a cluster, and you start one node after another, you clearly do not have quorum until there are enough nodes started.
So fencing and rgmanager does not start on those nodes.

So after you get quorum, you need to start those services on all nodes manually:

# /etc/init.d/cman/start
# /etc/init.d/rgmanager start

To get quorum you can also set the expected votes temporarily:

# pvecm expected X

hope that is more clear now.

 

[RobFantini]

Dietmar
that is very clear thank you. I'll remove the line now .

On the weekend will try turning off all nodes , then starting 1-st the main node and doing as you suggest to try to get rgmanager working properly.

 

[giuseppe.torrisi]

Hi,
i have the same problem, deadlock rgmanager when the nodes lost connection between them.
I have only found a patch for redhat system but not for debian.
when there will be a patch for debian ?
Best Regards
Giuseppe

 

[dietmar]

when there will be a patch for debian ?

What patch do you talk about?

 

[giuseppe.torrisi]

I refer to this dialogue

"06-16-2012
I will try to upload an update to pvetest next week. Best regards,
Dietmar"

the bug I'm talking about is:
https://bugzilla.proxmox.com/show_bug.cgi?id=105

Best Regards
Giuseppe

 

[dietmar]

I refer to this dialogue

"06-16-2012
I will try to upload an update to pvetest next week. Best

That is already uploaded to stable.

 

[e100]

That is already uploaded to stable.

Has bug 105 been tested to confirm the issue is fixed?
Bug 105 is not closed and the only comments on it are mine.

Maybe the issue RobFantini reported is fixed but not the issue giuseppe.torrisi is having?