You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@ignite.apache.org by Alexander Fedotov <al...@gmail.com> on 2017/04/14 12:44:25 UTC
ClusterGroup fallback
Hi Igniters!
I want to discuss if it possible to implement ClusterGroup fallback, i.e.
create
a chained cluster group ClusterGroup1->ClusterGroup2 that will refer to
nodes
satisfying conditions for ClusterGroup1 while there are such nodes and
if no such nodes become available then the nodes satisfying ClusterGroup2
conditions
will come into play.
As an example, consider a case when service per node should be deployed on
server nodes
located on a specific host. If all the server nodes on the host go down,
then all the remaining server nodes should be used for service deployment.
Any thoughts?
Kind regards,
Alex.
Re: ClusterGroup fallback
Posted by Valentin Kulichenko <va...@gmail.com>.
Also disagree, silent fallback is no good.
If we really need this, we can introduce some way to specify "preferred
locations" for service deployment, based on weights for example. But that's
a whole new feature and needs to be discussed in more detail.
-Val
On Fri, Apr 14, 2017 at 3:59 PM, Vladimir Ozerov <vo...@gridgain.com>
wrote:
> -1 from my side.
>
> If there are no nodes for a service this is an exceptional situation, which
> should be handled explicitly. Silent fallback will make things less
> obvious.
>
> On Fri, Apr 14, 2017 at 3:44 PM, Alexander Fedotov <
> alexander.fedotoff@gmail.com> wrote:
>
> > Hi Igniters!
> >
> > I want to discuss if it possible to implement ClusterGroup fallback, i.e.
> > create
> > a chained cluster group ClusterGroup1->ClusterGroup2 that will refer to
> > nodes
> > satisfying conditions for ClusterGroup1 while there are such nodes and
> > if no such nodes become available then the nodes satisfying ClusterGroup2
> > conditions
> > will come into play.
> >
> > As an example, consider a case when service per node should be deployed
> on
> > server nodes
> > located on a specific host. If all the server nodes on the host go down,
> > then all the remaining server nodes should be used for service
> deployment.
> >
> > Any thoughts?
> >
> > Kind regards,
> > Alex.
> >
>
Re: ClusterGroup fallback
Posted by Vladimir Ozerov <vo...@gridgain.com>.
-1 from my side.
If there are no nodes for a service this is an exceptional situation, which
should be handled explicitly. Silent fallback will make things less obvious.
On Fri, Apr 14, 2017 at 3:44 PM, Alexander Fedotov <
alexander.fedotoff@gmail.com> wrote:
> Hi Igniters!
>
> I want to discuss if it possible to implement ClusterGroup fallback, i.e.
> create
> a chained cluster group ClusterGroup1->ClusterGroup2 that will refer to
> nodes
> satisfying conditions for ClusterGroup1 while there are such nodes and
> if no such nodes become available then the nodes satisfying ClusterGroup2
> conditions
> will come into play.
>
> As an example, consider a case when service per node should be deployed on
> server nodes
> located on a specific host. If all the server nodes on the host go down,
> then all the remaining server nodes should be used for service deployment.
>
> Any thoughts?
>
> Kind regards,
> Alex.
>