Ok , so you are saying that Kubernetes make it easy to deploy OpenWhisk. Then we should really provide an helm chart, I think. Because it is the de-facto packaging for Kubernetes nowadays.
And contribute it here: https://github.com/kubernetes/charts What you think of this idea (I can volunteer to work on that)? I am still worried of the need of using Kubernetes but no one said so far it could be a good idea to skip it. -- Michele Sciabarra [email protected] On Wed, Mar 28, 2018, at 5:15 AM, Ben Browning wrote: > Michele, > > I have that dream as well! In fact, I've set up 3 production > deployments of OpenWhisk just last week across three different public > cloud providers. I'm deploying another production instance to a > private cloud later this week. In my case, I'm deploying OpenWhisk on > top of OpenShift (Kubernetes + some extras). Now, I'll admit that I > didn't have to install OpenShift on each of these clouds; someone else > did that for me. But, once OpenShift is there, I deploy OpenWhisk > identically across all 3. Deployment of a new production cluster takes > just a couple of minutes. I can use the same OpenShift template > (extension to regular Kubernetes yaml files - similar in spirit to > Helm Charts) to spin up a massive HA production cluster or a simple > temporary development cluster just by tweaking some parameter values > in a file. > > That's the beauty of Kubernetes, OpenShift, and the like. You can > deploy, manage, and monitor applications identically across any cloud > provider or even on bare metal servers. I understand if Kubernetes or > OpenShift aren't your thing, but these are the enablers of the dream > you described. > > Also, as an aside, CouchDB and Kafka clusters work great in > Kubernetes. I've been running HA clusters of each on OpenShift for a > while now. If you want more details on that, I'm happy to share since > it seems you've had some experiences that didn't turn out as well. > > We definitely have some work to do in the OpenWhisk project to > consolidate and document the Kubernetes, OpenShift, and Mesos > deployment options. All the pieces necessary to deploy everything are > a bit fragmented and it can be confusing to newcomers just trying to > get a production deployment going. But, I'm a big believer in using > some kind of container orchestration as the deployment target of the > future. > > Ben > > > On Tue, Mar 27, 2018 at 1:17 PM, Michele Sciabarra > <[email protected]> wrote: > > I have a dream: installing easily OpenWhisk , production grade, everywhere. > > Most notably, on AWS, but also on Azure, Digital Ocean and even on bare > > metal. > > > > I am aware of the Kubernetes deployment. Everyone is using that, I was told > > on Slack. > > > > Well, the fact is that in general installing Kubernetes is hard. And it > > adds a substantial burden on the system to run. > > > > There are installers for Kubernetes on AWS (kops) and more but they add a > > good level of complexity to the installation. > > > > Furthermore, there are Kafka and Couchdb and afaik they do not play so well > > with Kubernetes (I worked months deploying Kafka on Kubernetes so I know). > > Couchdb looks like to be on the same league, because of some restrictions > > on the deployment due to its design based on Erlang. > > > > In short, I have the feeling that a better solution would be to able to > > deploy OpenWhisk using just virtual machines. VIrtual Machine are not going > > away and be replaced by Kubernetes in the foreseeable future and AWS is > > still more a "virtual machine" than a container based Cloud. > > > > So I have this idea of creating an installer, based on docker-machine. I > > try to call it "wsk-machine". The idea is that wsk-machine would create a > > cluster and deploy it easily a multi node, production ready cluster on > > everything that docker-machine supports. > > > > How crazy (and wrong) sonds this idea? Please tell me it is completely > > wrong (and why). > > > > -- > > Michele Sciabarra > > [email protected]
