(see also T125485 for related more-complicated situation in esams)
The worst/oldest machines remaining in the eqiad clusters are:
cp104[34] - Last 2x of the oldest set still in use, currently testing cache_maps - to be decommed when maps gets real hardware soon
cp1056, cp1057, cp1069, cp1060 - The only non-SSD still use in eqiad - former bits cluster, currently misc cluster
The upload cluster is over-provisioned (as with esams, we were being conservative with how TLS rollout would go on CPU) and could easily lose 2x of its older (but still decent) spec SSD machines. The parsoid cluster is due for termination in the next few weeks and also has 2x decent spec SSD machines to give up. We can send these 4 machines to the misc cluster and decom the non-SSD ones there.
Steps:
[ ] 1. Decom cache_parsoid, freeing cp1045 + cp1058
[ ] 2. Remove cp10[56]1 from cache_upload (excess capacity)
[ ] 3. Add the 4x machines above into cache_misc
[ ] 4. Remove the current 4x non-SSD from cache_misc for decom/reclaim/spare