Autoscaling with HPA is not working

[josephburnett]

Even with a slow rampup (10 parallel/100 total, 20/100, 30/100) the server falls over without being scaled up. Defaults are 80% CPU, min 2 and max 10.

[evankanderson]

80% CPU load is probably way too high, particularly if the function isn't purely compute bound. I'd expect you'd need to either drive off a custom metric like desired parallelism, or set the CPU threshold closer to 50%.

…

On Thu, Feb 8, 2018 at 10:32 AM, Joseph Burnett ***@***.***> wrote: Even with a slow rampup (10 parallel/100 total, 20/100, 30/100) the server falls over without being scaled up. Defaults are 80% CPU, min 2 and max 10. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#107>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHlyNxCOwWJQX7ATpocnNRZ75tDfJRyPks5tSz2qgaJpZM4R-0nB> .

-- Evan Anderson <[email protected]>

[josephburnett]

I was able to the HPA to scale up on CPU. Created an app with a 100,000 for loop. Ran 250 QPS. Saw an increase from 1 to 3 pods. But it was very slow to happen (~2-3 minutes).

I'm actively working on setting up QPS autoscaling (first #112 then wiring HPA into Prometheus). In the meantime, I agree 50% would be a better threshold.

NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
hpa/p-64e3cf1d-34f7-4efa-8bd2-f638d15960a0--autoscaler Deployment/p-64e3cf1d-34f7-4efa-8bd2-f638d15960a0--ela-deployment 166% / 80% 1 10 3 11m

NAME READY STATUS RESTARTS AGE
po/p-64e3cf1d-34f7-4efa-8bd2-f638d15960a0--ela-deployment-59dfffd7 3/3 Running 0 11m
po/p-64e3cf1d-34f7-4efa-8bd2-f638d15960a0--ela-deployment-59dhknq5 3/3 Running 0 5m
po/p-64e3cf1d-34f7-4efa-8bd2-f638d15960a0--ela-deployment-59dl646h 3/3 Running 0 1m

[YasminBZ]

Hello @josephburnett please how did you setup the HPA autoscaling for example if i want to autoscale my function using minmemory 128 maxmemory 250 mincpu, maxcpu and a target of 50% ??

Thank you