We had an incident yesterday where NameRes RENCI Dev slowed down but without visibly struggling on CPU or memory, so we can't use those to control horizontal scaling in the same way we do NodeNorm web frontends. Some possibilities:
- We could reduce CPU/memory, which might make that strain more obvious.
- The actual limiting factor is probably the number of simultaneous connections -- is there some way we could track that, maybe through the
/status endpoint?
- We might also be able to detect a slowdown in NameRes rates.