At this point the cluster is just about ready to be deployed.
The steps needed after some additional documentation about what we did and how it works are:
- Ensure that it is clear how to deploy PAWS while the deploy-hook is disabled for all parties.
- Change the old cluster to using sqlite and the new cluster to toolsdb
- Switch over the paws.wmflabs.org DNS entry and update all URLs on-wiki to paws.wmcloud.org.
- Do a final rsync of the NFS.
- Clean up the old cluster
I propose that I spend until next Friday (2020-07-31) cleaning up the documentation and getting the PR ready (including tagging "newbuild" with "latest" so that the "latest" tag can be a dupe of whichever tag we actually want deployed). This should include touching base with everyone on the project as is possible about how deployments should work in the meantime.
On that day, we can announce the new cluster as a "public beta" for a week (??) for people to try in case there are bugs we haven't found yet. Since it will be on sqlite, it won't perform as well as it will be able to after switchover, but it should be sufficient. The PR might cause some quirks with the deploy-hook on the old cluster so don't do the PR until we are about to switch.
On 2020-08-07, we cut over to the new cluster following the steps above up to the last one. We do not want to clean up the old cluster until the new one has seen a bit of load and hasn't exploded. After that, we can proceed with working on things like the CI fixups and deploy-hook updates. It will also make a nice minikube instruction-set a clear ticket, etc.