Now that ores-legacy handles all the traffic for ores.wikimedia.org, we can start decomming it.
The starting point is https://wikitech.wikimedia.org/wiki/Server_Lifecycle#Remove_from_production
High level idea:
- Remove old icinga alerts etc.. Set LVS checks to not page if all nodes are down.
- Remove ores nodes from pybal, plus its config (VIP, etc..). This requires coordination with Traffic since some pybal restarts are probably needed (to update LVS etc..).
- Shutdown all ores nodes + downtime.
- Clean up DNS records.
- Clean up puppet from ores classes and configs (including POSIX groups etc..) (the ores-admin group was only ever applied to hosts which now get fully decommissioned, as such the group entry can be removed entirely and the GID added to the reclaim list in L71 of data.yaml)
- Cleanup TLS certificates for ores.discovery.wmnet
- Follow up with Service Ops to remove the Redis instances dedicated to ORES on rdb nodes (see https://wikitech.wikimedia.org/wiki/Redis#Cluster_redis_misc)
- Check in Horizon for old VMs etc.. We should probably shutdown everything, and close projects that we don't need (so we release capacity etc..).
- Decommission https://ores-beta.wmflabs.org/, all hosted in deployment-prep's settings (we can access to those via Horizon).
- Update the Wikitech/Mediawiki documentation
- Delete old grafana dashboards
- Archive ores-related repositories - tracked in T349632