Now that we have moved a majority of jobs to k8s, we can take some hardware back from the jobrunner cluster. Based on the graphs it seems we can reclaim at least 50% of the jobrunner hardware while refreshLinks and cirrusSearchLinksUpdate remain on metal, but we should be careful to account for spikes and avoid letting the remaining hosts be overwhelmed, so a good start is 25%.
As of time of writing, there are 30 jobrunners in codfw, and (minus parse1* hosts), 23 in eqiad. Take care not to repurpose videoscaler-only hosts.
Initial candidates for ~25% migration with rack placement:
[*] mw1460 (A8)
[*] mw1461 (A8)
[*] mw1469 (B6)
[*] mw1439 (D8)
[*] mw1486 (C5)
[*] mw1495 (F)
[*] mw2395 (A3)
[ ] mw2427 (A6)
[*] mw2267 (B3)
[ ] mw2430 (B6)
[*] mw2357 (C6)
[ ] mw2282 (D4) (blocked T355333)
[ ] mw2446 (D5)