thorium.eqiad.wmnet hosts the following sites:
- https://turnilo.wikimedia.org (UI for Druid data, will replace Pivot)
- https://superset.wikimedia.org (UI for creating analytics dashboards from internal data)
- https://hue.wikimedia.org (Hadoop User Experience UI)
- https://yarn.wikimedia.org (Internal YARN ResouceManager UI)
- https://stats.wikimedia.org (Wikistats)
- https://analytics.wikimedia.org (Analytics dashboards and datasets)
- https://datasets.wikimedia.org (deprecated, redirects to analytics.wikimedia.org/datasets/archive)
- https://metrics.wikimedia.org (redirect to https://metrics.wmflabs.org/ (Wikimetrics))
Of these, all of the internal sites can be hosted from anywhere, and are prime candidates to move to Ganeti instances. The public sites (with the exception of the 2 redirects) require large amount of disk space (around 1.5TB total).
We'd like to reimage thorium to Debian Stretch, but before we do, it might be easier to move the internal sites off of thorium onto one or multiple Ganeti instances. This will also allow us to upgrade turnilo and superset (T201430) to newer versions for Debian Stretch before we take down the existing instances.
So! Please allocate one or multiple Ganeti instances that we can use for this purpose. With the exception of yarn.wikimedia.org, the sites do a fare amount of work to query for and build graphs and dashboards, so a decently speced node would be good. Perhaps 3 instances would be best, one for turnilo, one for superset, and another for both hue and yarn. 4G RAM each, 4 core each?