On Tue, Sep 27, 2016 at 12:08 PM, Sanjeewa Malalgoda <[email protected]>
wrote:

> Hi,
> Before we need to take any decision about moving to pub/sub approach we
> need to consider some of the facts.
>
> As of now token revoke responses carrying revoke token as transport header
> and at gateway cache clear handler will remove revoked token from cache.
> Then it will replicate to other nodes through clustering.
>
> If we are planning to have pub/sub solution then either oauth2
> implementation or cache clear handler at gateway should push this event to
> topic.
> Both cases we need to maintain message broker in gateway or key manager
> while all gateway workers subscribe to topic available there.
> Then we need to think about how high availability works for this
> broker(since we have multiple gateways and key managers in most of the
> deployments). IMHO it will add more complexity to deployment.
>
> My main point is even if we use pub/sub in order to achieve HA we need
> multiple broker instances. Then those need to synchup with each other and
> we need some sort of group communication for that(again clustering comes to
> picture).
>
> May be we can evaluate some solution like kafka for this. We are
> evaluating it for traffic manager update retrieving process(through kafka
> event publisher in CEP and extension to subscribe topics to fetch updates).
> It will not be a first class support but through extensions we may be able
> to do that. Even with that we need to maintain zoo-keeper cluster when we
> have multiple brokers(again this make deployment bit complicate).
>

What if we add this as an deployment option? We can ship it as a component
that is switched on with configs.


>
> If need we can wait 15 minutes for cache timeout rather adding this kind
> of feature if users do not like to use gateway clustering.
>
> And clustering is required to replicate validation information cache
> across gateway nodes. Otherwise when LB not routing requests without
> session awareness gateways may do same key validation call again and again.
> In this case usually cluster communication is cheaper than another key
> validation call. So if we remove clustering completely then we need to
> think about this as well.
>

I think the best here is switching on IP hashing at LB.

Cos think of a case where we have to support a large TPS. Having this would
be so easy. For large TPS (say .. XXXXX TPS range) they wouldn't even mind
having a their own broker in HA.

thanks,
Dimuthu



>
> Thanks,
> sanjeewa.
>
>
>
> On Tue, Sep 27, 2016 at 11:31 AM, Dimuthu Leelarathne <[email protected]>
> wrote:
>
>> Hi,
>>
>> If we publish OAuth key revocation to a topic (we can do so using by
>> writing an extension to WSO2IS), we can remove clustering in the APIM
>> gateway. Are there better ways for achieving the same? Can we prioratise
>> this for next APIM release?
>>
>> thanks,
>> Dimuthu
>>
>> --
>> Dimuthu Leelarathne
>> Director, Solutions Architecture
>>
>> WSO2, Inc. (http://wso2.com)
>> email: [email protected]
>> Mobile: +94773661935
>> Blog: http://muthulee.blogspot.com
>>
>> Lean . Enterprise . Middleware
>>
>
>
>
> --
>
> *Sanjeewa Malalgoda*
> WSO2 Inc.
> Mobile : +94713068779
>
> <http://sanjeewamalalgoda.blogspot.com/>blog :http://sanjeewamalalgoda.
> blogspot.com/ <http://sanjeewamalalgoda.blogspot.com/>
>
>
>


-- 
Dimuthu Leelarathne
Director, Solutions Architecture

WSO2, Inc. (http://wso2.com)
email: [email protected]
Mobile: +94773661935
Blog: http://muthulee.blogspot.com

Lean . Enterprise . Middleware
_______________________________________________
Architecture mailing list
[email protected]
https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture

Reply via email to