Worker count math for 4 vCPU API boxes
Operational notes on benchmarks: what to pin, what to smoke, what to ignore.
What changed
Worker count math for 4 vCPU API boxes is the short version. The parts that actually break production are below.
Practical notes
- Pin the dependency in lockfiles before rolling fleet-wide.
- Re-run your smoke against
/docsand one authenticated route. - Watch p99 for 15 minutes after deploy; mean latency lies.
from fastapi import FastAPI
app = FastAPI()
@app.get("/health")
def health():
return {"ok": True, "tag": "benchmarks"}
Caveats
Do not treat changelog marketing as a migration guide. Diff the release against your current pin and only upgrade when the smoke list is green.