We have a small Google Cloud Run API that gets sporadic requests throughout the day. Most requests are quick, but the first request after a quiet stretch can be noticeably slower, so we’re considering setting a minimum instance count.

For a service with uneven traffic, how do you weigh Cloud Run minimum instances against the extra cost of keeping capacity warm? Do you set a minimum only for latency-sensitive routes, or accept occasional cold starts for a lightly used service?