On Wed, Apr 6, 2016 at 5:26 AM, David Strejc <[email protected]> wrote:

> Our whole team have agreed on following solution:
>
> * We will install nearly same service as ha-proxy router in Open Shift is.
>
> * We will have ha proxy in docker container inside Open Shift on each node
>   this will serve as connection point from our applications - we will
> create service out
>   of this HA Proxy dockers
>
> * In case of failure Kubernetes will know about issue and we can use
> readiness for
>   this purpose inside so we will have HA Proxy in front of our
> infrastructure of MariaDBs
>

​I guess it sounds like something that could work, but I suspect there
might be other ways to communicate endpoint state to the service definition
that don't require this extra hop.  Adding Ben Bennett from the team that
owns routing.

​



>
> Do we have to use the same system for GlusterFS?
>
> Use case:
>
> We got glusterfs on each openshift node and they serve as storage for
> data. We got
> Endpoints inside our template and it points to GlusterFS according to Open
> Shift
> documentation for GlusterFS.
>
> If one node fails - Kubernetes will reschedule all containers on different
> node, but
> what about GlusterFS - will there still be requests to failed node from
> our apps?
>
> Due to Kubernetes documentation it chooses Endpoints at random from all
> endpoints
> specified. Does this apply also to GlusterFS plugin? If so wee need to
> setup HAProxy
> also for GlusterFS storage.
>
> Does this make a little bit sense or are we completely wrong with our
> design?
>
> I can send design scheme if it will help.
>
> Many thanks for your help and time.
>
> David Strejc
> t: +420734270131
> e: [email protected]
>
> On Tue, Apr 5, 2016 at 6:28 PM, David Strejc <[email protected]>
> wrote:
>
>> And what about GlusterFS - does Open Shift driver support checking if
>> GlusterFS endpoint is up and running?
>>
>>
>>
>> David Strejc
>> t: +420734270131
>> e: [email protected]
>>
>> On Tue, Apr 5, 2016 at 5:54 PM, Ben Parees <[email protected]> wrote:
>>
>>>
>>>
>>> On Tue, Apr 5, 2016 at 11:26 AM, Tomáš Kukrál <[email protected]> wrote:
>>>
>>>> Hi,
>>>> is it really possible to attach Endpoint?
>>>>
>>>> I think readiness probes are limited to containers and thus is can not
>>>> solve David's problem because mysql servers are running outside of
>>>> OpenShift.
>>>>
>>>>
>>> ​sorry, didn't realize it's an external service, my advice would not
>>> apply if that is the case.
>>> ​
>>>
>>>
>>>
>>>> tom
>>>>
>>>>
>>>>
>>>> On 04-05 09:04, Ben Parees wrote:
>>>> > On Tue, Apr 5, 2016 at 3:59 AM, David Strejc <[email protected]>
>>>> wrote:
>>>> >
>>>> > > Dear all,
>>>> > >
>>>> > > my use case is as follows:
>>>> > >
>>>> > > I got MariaDB galera cluster on each node I got Open Shift
>>>> installed on
>>>> > > (they are bare metals).
>>>> > >
>>>> > > I've created endpoints such as this:
>>>> > >
>>>> > > - apiVersion: v1
>>>> > >   kind: Endpoints
>>>> > >   metadata:
>>>> > >     name: mysql
>>>> > >   subsets:
>>>> > >    - addresses:
>>>> > >      - ip: 10.3.0.1
>>>> > >      ports:
>>>> > >      - port: 3306
>>>> > >    - addresses:
>>>> > >      - ip: 10.3.1.1
>>>> > >      ports:
>>>> > >      - port: 3306
>>>> > >    - addresses:
>>>> > >      - ip: 10.3.2.1
>>>> > >      ports:
>>>> > >      - port: 3306
>>>> > >
>>>> > > And I got service which points to this endpoint.
>>>> > >
>>>> > > My question is - Kubernetes doesn't have any kind of "HA proxy
>>>> mode" so it
>>>> > > can tell that any of nodes behind Endpoint is failing? So lets say
>>>> I got
>>>> > > 10.3.0.1 mysql service stopped and kubernetes round robins
>>>> connection to
>>>> > > this host so every third request on my webapp is failing?
>>>> > >
>>>> > > This is what I've experienced in my setup.
>>>> > >
>>>> > > Am I doing something wrong? Or is this a standard behaviour so I
>>>> have to
>>>> > > deploy HA Proxy in between the endpoints and my MySQL database
>>>> cluster?
>>>> > >
>>>> >
>>>> > ​this what readiness checks are for.  If a container fails its
>>>> readiness
>>>> > check, the endpoint will be removed from the list so it does not serve
>>>> > requests:
>>>> >
>>>> >
>>>> https://docs.openshift.org/latest/dev_guide/application_health.html#container-health-checks-using-probes
>>>> >
>>>> > ​
>>>> >
>>>> >
>>>> >
>>>> > >
>>>> > > Many thanks for advice.
>>>> > >
>>>> > > David Strejc
>>>> > > t: +420734270131
>>>> > > e: [email protected]
>>>> > >
>>>> > > _______________________________________________
>>>> > > users mailing list
>>>> > > [email protected]
>>>> > > http://lists.openshift.redhat.com/openshiftmm/listinfo/users
>>>> > >
>>>> > >
>>>> >
>>>> >
>>>> > --
>>>> > Ben Parees | OpenShift
>>>>
>>>> > _______________________________________________
>>>> > users mailing list
>>>> > [email protected]
>>>> > http://lists.openshift.redhat.com/openshiftmm/listinfo/users
>>>>
>>>>
>>>> _______________________________________________
>>>> users mailing list
>>>> [email protected]
>>>> http://lists.openshift.redhat.com/openshiftmm/listinfo/users
>>>>
>>>>
>>>
>>>
>>> --
>>> Ben Parees | OpenShift
>>>
>>>
>>> _______________________________________________
>>> users mailing list
>>> [email protected]
>>> http://lists.openshift.redhat.com/openshiftmm/listinfo/users
>>>
>>>
>>
>


-- 
Ben Parees | OpenShift
_______________________________________________
users mailing list
[email protected]
http://lists.openshift.redhat.com/openshiftmm/listinfo/users

Reply via email to