@yuvipanda: I successfully created a scheduled job by copying the example kubectl run hello --schedule="*/1 * * * *" --restart=OnFailure --image=busybox -- /bin/sh -c "date; echo Hello from the Kubernetes cluster", looked at it with kubectl get jobs and deleted it with kubectl delete scheduledjobs hello. Is there anything else that needs to be done to resolve this task?
Shouldn't image be something like docker-registry.tools.wmflabs.org/toollabs-python-web:latest?
To resolve this task I think we need documentation somewhere on wikitech mentioning the feature exists and what the current level of support is.
I think the feature exists, and there's no official levels of support from us. You can refer to upstream docs to use it right now, and there might be community documentation.
I think in some ways using kubectl is similar to using qsub directly - there's upstream docs, we have it installed and turned on, and you can do more powerful things with it than with jsub.
This might change in the future if we have more resources.
@Legoktm I think 'we' can create a page on wikitech that collects documentation about 'advanced' Kubernetes usage. There are some things I can contribute from the Stashbot config about setting up and running a continuous job using a custom deployment file. The page needs to be very clear however that using these kinds of features is for intrepid adventurers only, could break at any moment, and not officially supported in any way by the Tool Labs/Cloud Services team.
I really, really want to make using Kubernetes beyond a webservice backend common, but we have a list of more pressing issues for the Cloud Services team to work on that is going to keep it from being a goal for the foreseeable future. There was a hope at one point that we could 'transparently' migrate people from OGE to Kubernetes, but we have done enough work towards that at this point to have proven that it will not be possible. Building a lot of custom scripting for Kubernetes like we did for OGE would create even more tech debt for us to cleanup eventually. The blocker to moving forward properly is paying down enough of our existing tech debt in the OpenStack deployment to free up resources needed to move forward with net new projects like selecting and deploying an easy to use command and control system for running tools on Kubernetes.
(This task was about allowing running cron jobs on Kubernetes, and that has been resolved; https://wikitech.wikimedia.org/wiki/Help:Tool_Labs/Web/Kubernetes does not even mention kubectl or explain what pods are, so requiring documenting how to use advanced features for this task to be closed would IMHO be too much effort. Stashbot's configuration looks interesting for a blog post to make people aware and curious.)