auto-start and auto-stop response times

[S3bb1]

Continuing the discussion from Auto-scaling and uptime monitors:

@pepicrft

We are using Fly’s auto-scaling feature to auto-start and auto-stop machines based on the traffic, but we noticed our uptime monitors trigger because some requests take more than 30 seconds to respond. We wonder if it has something to do with auto-scaling taking longer than expected. What would be the best way to gather that information? What would you recommend to do in this case?

We currently also observe this pattern mentioned in the previous thread. Since we use the auto-start and auto-stop feature our uptime monitor shows a lot outliers which take around 30 secs.

On the fly metrics dashboard none of them show up and everything seems “normal”.

Does anyone else has seen this before?

[khuezy]

I haven’t noticed anything strange w/ stopped => start times. It’s been inline with what I’m experiencing on Lambda (3-4s.) It might be region specific?

[Moreno]

you can use suspend instead of “off”, it’s much faster

[khuezy]

Do note that suspend is buggy and not production ready.