[Debian-ha-maintainers] Bug#847295: Bug#847295: pcs: cluster auth does not generate tokens file & hangs on find

[Rafael David Tinoco]

On 8 Dec 2016, at 20:59, Valentin Vidic <Valentin.Vidic at CARNet.hr <mailto:Valentin.Vidic at CARNet.hr>> wrote:
> 
> On Thu, Dec 08, 2016 at 06:27:48PM -0200, Rafael David Tinoco wrote:
>> Definitely. Since PCS is a package to *configure* corosync. I do agree
>> with your statement (above and bellow).  
> 
> True, but there is another use-case where you install it on an existing
> cluster to get a web interface. In this case you don't want it to touch
> existing setup by default.

Wasn't aware of that, my bad here. Somehow I assumed that it could only
manage a cluster configured by it from the beginning, I should have 
tested this more carefully. 

That is why I was removing corosync.conf. 

> 
>> Unfortunately PCS fails silently for a lot of things, including
>> missing files (or even missing corosync binary, that is why I put
>> it as a "dependency" -> https://goo.gl/sZxest <https://goo.gl/sZxest> <https://goo.gl/sZxest <https://goo.gl/sZxest>>).
> 
> Not sure how you got this because installing pcs all of the following
> get installed:
> 
>  pcs -> pacemaker -> corosync

APT::Install-Recommends "false" in apt.conf does that. Pacemaker is declared 
as "Recommended" and it should be as "Dependency":

https://www.debian.org/doc/debian-policy/ch-relationships.html <https://www.debian.org/doc/debian-policy/ch-relationships.html> 
" The Depends field should be used if the depended-on package is required 
for the depending package to provide a significant amount of functionality."

I choose to put corosync only as a Dependency. Maybe its better to have 
Pacemaker instead.

> 
>> Just trying to make PCS functional "by default" since now it will be
>> used as the clustering configuration tool for MSSQL Linux HA and they 
>> need it configured by default (or capable of). 
> 
> I would argue that pcs is functional as all three daemons are running:
> 
>  * corosync
>  * pacemaker
>  * pcsd
> 
> What is not functional is the cluster as a whole, but this will always
> require some manual configuration as it spans more than one host and
> we cannot handle that in packaging.

Yep, I got you now. 

> 
> In your case you would need to run something like:
> 
>  pcs cluster destroy
>  pcs cluster auth node1 node2
>  pcs cluster setup --start --name cluster node1 node2
>  pcs resource create ...
> 

Definitely. For LP: #1640923 the key is:

"sudo pcs cluster destroy" before moving on. 

It clears /etc/corosync and /var/lib/corosync/ (I just realized my upstream 
commit was wrong, just fixed it into: https://github.com/ClusterLabs/pcs/pull/120 <https://github.com/ClusterLabs/pcs/pull/120>.

With that said, sorry to bother and thank you much for reviewing this for me.
I'll make sure patches are good for Xenial/Yakkety and propose the debdiffs.

No change is needed in Debian. Should I close this case ? 
(Would you use it for changing recommends/depends ?)

Rafael


> -- 
> Valentin

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/debian-ha-maintainers/attachments/20161208/348f8b58/attachment-0001.html>