- start with something that doesn’t require maps; say, a small fixed universe of hosts. This isn’t going to use incremental in an especially interesting way, but it’s a good place to start.
- Maybe even have an open universe of hosts, but have part of the query be to select which hosts you actually look at. That’s static enough to avoid maps. You’ll need an imperatively managed table of data on the outside, though.
- Various queries to push the model more, in particular:
- find set of checks of a certain kind that failed (filter_map)
- find set of stale nodes (clock + filter_map’)
- Rank queries? (nth-worst host?, median host?)
- Merging examples? Drop rates on hosts with most highly loaded switches?
- Benchmarks! Make sure that we have all-at-once computations for each of these.
- Talk to Dlo and Perl about useful queries to implement.