[Hawkular-dev] Agent - EAP Domain mode support

[Thomas Heute]

On Mon, May 2, 2016 at 7:32 PM, John Mazzitelli <mazz at redhat.com> wrote:

> > The question is likely for Mazz, but I wanted to check how we will
> support the WF domain mode.
> >
> > As far as I understand, we are now able to run the agent in the host
> > controller but wanted to know a bit more on how we expect the setup to be
> > (one agent on the host controller for the whole domain ?)
>
> <TL;DR>
> I can run an agent in a host controller, but my main question is: What
> special functionality are these agents supposed to provide by having them
> run in the DC or HC (as opposed to running in all the standalone servers?)
> </TL;DR>
>



>
> Right now, I have the agent at least able to run in a host controller.
> Each machine where you run your app servers has one host controller on
> them. Additionally, somewhere in your network, there is a single host
> controller that is configured to run as the domain controller (i.e. a
> "domain controller" is just a specially configured host controller - there
> is only one DC per domain).
>
> We can run the agent today in a host controller (note: I have no agent
> configuration that tells the agent how to monitor that HC itself - there
> are not the same subsystems running in a HC than what you see running in a
> standalone WildFly server).
>
> Do we want to run the agent only in the domain controller? Or do we want
> to run one agent in each host controller? I suspect we want to do the
> latter, not the former, simply because it seems to follow the typical way
> agents are run (one per machine). However, I'm still not clear what it is
> people want these agents running in a DC or HC to do. What special
> functionality are these agents supposed to provide by having them run in
> the DC or HC (as opposed to running in all the standalone servers?)
>

We need to be able to "visualize" server groups and domains --> that
topology needs to fill in the inventory
We'll want to be able to execute operations on a domain or server group,
like deploy an application (without re-implementing the logic for failed
deployment on 1 or more servers for instance)

If the domain/host controller collects all metrics for all servers or if
each server reports itself is more of an implementation detail. Also I
understand that have all data going through the domain controller would
create a bottleneck, one per host seems more suitable on first thoughts.

Thomas



>
> A different approach to consider is that we can have the host controllers
> inject agents into the WildFly servers they spawn (so we can then monitor
> each of the standalone servers that are started by the host controller
> since each standalone server can get an agents installed via the HC). I am
> not sure if HCs can actually push out module subsystem extensions like they
> can deployments - we'd have to ask wildfly-dev. But it would be cool to add
> the agent subsystem to the domain's profiles so when a HC runs a WildFly
> server it can not only push out the deployments but also the agent
> subsystem extension.
>
>
> > Also I wanted to ask if older versions of the app server could be
> supported
> > or only WF 10 ? (in particular in the context of EAP7/EAP6 support)
>
> I have not tried to run this in a WF9 instance - but alot of this is
> fairly new in the WF codebase. There was a time when custom subsystem
> extensions could NOT run in these host controllers - wildfly devs had to
> add some stuff to get that to work. I suspect (though I do not know for
> sure) that it won't work on EAP6.
> _______________________________________________
> hawkular-dev mailing list
> hawkular-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/hawkular-dev
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/hawkular-dev/attachments/20160503/f56ce9fd/attachment.html