OpenWhisk on Kubernetes meeting notes
Date: Friday, September 15th
Attendees: Dan Lavine (IBM), Dave Grove (IBM), Jim Crossley (RedHat)
Start:
9:10:
Jim - RedHat is working with their own deployment of for Kubernetes + OpenWhisk for OpenShift. The main difference between RedHat's and the current incubator-openwhisk-deploy-kube repo is that OpenShift's have elevated privileges in the Docker Container. This is so the
Jim - There will be a PR for those changes and most of them are minor.
Jim – There is a larger change for how we build the CouchDB image. Some of the logic for the init scripts we need to be built into the Docker Image. There is also an issue where we need specific versions for OpenShift some times and so we built in the ability to provide environment variables for checking out specific versions.
Dan – That should be fine. We just need to make sure the images work for all the environments.
Dan – For the CouchDB Docker image, I would like to use an offical CouchDB image that OpenWhisk is hosting. Initially when Kube + OW was kicked off, we disscussed that CouchDB would probably only be used for dev on Kube. This is because having a hosted CouchDB makes doing backups, monitoring, etc easier to do. However, for dev it would be nice to build a basic snapshot image in CI where CouchDB is set up and publish that image.
Jim – the only concern we would have with an image like that is that it runs on OpenShift.
Dan – that would be totally fine to change the permissions/setup and figure out how we host it officially.
Jim – I will submit a PR next week then with the changes
9:15:
David – More feedback on the controller saving unique UUIDs for Invokers from Ben Browning. Will need to save the state of the controller externally so when the controller restarts, it does not have to start from scratch again. Currently we are able to use it for Kube DaemonSets, just need to do fix saving the state.
Dan – I thought we are able to do stuff like this with Reddis when deploying multiple controllers.
Jim – Right, there was the same problem with the load balancer, and so there should be some previous work already out there to use as a reference.
9:17:
Dan – From my end, it is unfortunate but the OpenWhisk team is spread fairly thin and so I just need to keep pushing on getting some new CI images. I would like to work with someone though since I have no idea how the CI currently works.
Dan – I also added the last few components to the deployment, Reddis and ApiGateway.
Dan – There is one more Ansible script for route management needs to be ported over so we can deploy it through Nginx.
David – If you are really blocked on CI stuff, there is some more pressure to get Kuberentes + OpenWhisk running.
End