[Linux-HA] Windows port of haclient

[Andrew Beekhof]

On Feb 8, 2008, at 10:14 AM, Andreas Mock wrote:

>> -----Ursprüngliche Nachricht-----
>> Von: linux-ha-bounces at lists.linux-ha.org
>> [mailto:linux-ha-bounces at lists.linux-ha.org] Im Auftrag von
>> Andrew Beekhof
>> Gesendet: Donnerstag, 7. Februar 2008 20:29
>> An: General Linux-HA mailing list
>> Betreff: Re: [Linux-HA] Windows port of haclient
>
> Hi all,
>
> only my opinion as I'm also interested in another GUI solution.
>
>
>>> Well, I was thinking more on reachability and device independent
>>> access to the cluster. One drawback I see now is that
>> admins have to
>>> ssh -X or directly vnc a linux machine that has the client
>> installed
>>> in order to operate with the cluster.
>>
>> one can also connect to the CIB from non-cluster nodes
>
> This has to be the real way, IMHO. All should keep remote  
> administration
> of cluster servers in mind. I don't like to have X stuff on a
> server. I would be also not so pleased to have something like an
> Apache on the cluster only to access a gui.

I should probably make the CLI tools support remote connections at  
some point.

>
>
> So, probably I would like to see a console gui.  :-)
> (Yes, I like curses-Yast)
>
>
>>> Being web-based it would become ubiquitous and could be
>> accessed by
>>> lots of devices. Being the communications between the web
>> client and
>>> the server part XML, it would become interoperable and
>> allow further
>>> developments in the client side. Imagine that the web client part
>>> could be interchangeable and replace it with a windows native
>>> client, or a web service that queries the cluster status before
>>> trying to perform a transaction ...
>
> I haven't seen yet a discussion about what functionality is needed
> in that gui. I would like to give some requirements from my point
> of view:
>
> a) It has to work. (That one I explicitly put in here for Andrew :-))

or at least not corrupt valid configurations :)

> b) Support for creating cib-snippets. The created IDs should have
> a meaningful naming schema so that you can understand the cib when
> you extract it by hand or you have to debug something. UUID are
> very hard to use as identifieres. You have to be aware that you
> don't need that often. After cluster is configured it's done.

on a semi related note, the dev tree now has something called cibpipe  
- a tool for performing/simulating multiple configuration changes  
"offline"

   eg. cibadmin -Q \
	| ./cibpipe -D -X '<group id="bad-group"/>' \
	| ./cibpipe -D -X '<rsc_colocation id="bad-group-colocation"/>'  \
	| ./cibpipe -M -X '<cib admin_epoch="admin_epoch++"/>'  \
	| cibadmin -R

>
>
> c) Good assistance for creating constraints so that you get the
> behaviour you want. (This is the most important topic!!). This
> part would create constraint snippets and would set cluster
> and resource attributes (e.g. stickyness) set right way.
> People are interested in a behaviour not in the machinery behind
> that. At the moment you have to have a deep understanding to set
> all parameters in a way to achieve what you like. GUI should hide
> here.

Basically we're talking wizards here.
Manual editing is fine when you've been using the cluster for a few  
months, but earlier on you dont know what you need.
Even a bad wizard helps by asking the right questions - a good one  
will fill in the right templates and upload the changes :-)

> d) Give overview over the status of the cluster. IMHO this is done
> already very good by crm_mon. Someone has to think about a good
> gui presentation if you have many nodes and many resources (kind
> of sorting by nodes, RA, last changes, faults)

Using crm_mon to create a HTML summary is one option.
Another might be to use xslt.

> Michael and others seemed to work with SNMP here,

Improving the SNMP support is something I'm quite serious about.
Its on my (not exactly short) to-do list for pacemaker 1.0.

> which is a path
> if you have to integrate monitoring in enterprise scale monitoring
> solutions. IMHO as soon as you HAVE to have a cluster because of HA
> you have to use external monitoring tools. Maybe just to measure
> your HA. (Boss: You spent so much time and money to make this HA.
> How HA are you now for the last months. Admin: Very high. Boss: What
> exactly? Admin: Aeh, I don't know. Of course much more that before.
> Boss: I need figures!!!)
>
> e) You should be able to manage the cluster more easily by gui.
> Starting, stopping resources; making node standby; moving
> resources around;
> But this can be done easily by some wrapper scripts around the
> current command line tools.
> But IMHO this is also not done very often.
>
> What do you think?
>
> Best regards
> Andreas Mock
>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems