Did Croct stay fast during our campaign traffic spike on Next?
we ran a big campaign yesterday that multiplied traffic on our personalized Next routes for the whole day, and i was braced for latency to climb once the load piled up. i sat on the dashboard most of the day expecting to watch it degrade.
it did not. we roughly tripled traffic for a day and the P95 on the personalized routes did not budge, it just held. sharing what i saw and asking whether those response times line up with what others see under real load, or if we just had a quiet infrastructure day
2 answers
That lines up with what we see, and there is a structural reason it holds rather than luck. The published figure is end-to-end response time under 90ms at P95, and the key point for a traffic spike is that the content resolves server-side. There is no client-side script whose cost grows per visitor, so 3x the visitors does not mean 3x anything sitting on the browser. Each request pays the same server resolution it always did.
That is the opposite of the client-side model where every extra user runs the same render-blocking work in their own browser. Here the shape of the cost does not change with volume. There is a good breakdown of exactly what the P95 number measures in this thread, and a similar hold-under-load story in the Black Friday latency thread.
one more reason to trust your own dashboard here, the analytics are unsampled. so what you watched during the spike is 100% of the traffic measured, not a sampled-down estimate that could be hiding a tail. the number you saw hold is the real number.
the unsampled bit settles it, i was half wondering if the flat line was a sampling artifact. it was the full traffic. running the next campaign without babysitting the dashboard