Autosuspend doesn't work with multiregion scaling?

xHomu · October 2, 2024, 8:51pm

We have been using autosuspend since beta, and it was among our favorite Fly.io feature since! The ability to add redudancy without incurring significant performance/cost penalty is huge!

Everything works as promised with very little fiddling scaling to a single region:

fly scale count 2 --region sjc

The secondary spins down and only comes online during traffic spikes.

We encountered an sjc outage recently, so I’m experimenting with multi-region scaling:

fly scale count 5 --region sjc,lax

However, autosuspend seem to no longer work as well, with the machines up almost all the time, despite similar traffic levels:

Have I mistaken how multiregion works, or perhaps a misconfig fly.toml?

github.com

manawiki/mana/blob/main/fly.toml

app = ""
primary_region = "sjc"
kill_signal = "SIGINT"
kill_timeout = "5s"

[build]
  dockerfile = "Dockerfile"

[env]
  PORT = "3000"

[http_service]
  internal_port = 3000
  force_https = true
  auto_stop_machines = "suspend"
  auto_start_machines = true
  min_machines_running = 0
  processes = ["app"]
  [http_service.concurrency]
    soft_limit = 1600

This file has been truncated. show original

ignoramous · October 2, 2024, 10:57pm

Might not be auto-suspend. You can switch back to auto-stop and see if this issue persists?

Our multi-region app, which to be fair gets a LOT of traffic (est. 300 to 500 reqs/s), has faced the issue of machines spinning up (apparently needlessly) then never spinning down, for quite a while now: Overriding staged secrets for Machines - #7 by ignoramous

If you’re on the Launch plan (or higher), consider writing to Fly’s support. They’re pretty responsive.

xHomu · October 2, 2024, 11:10pm

I’ll give support a shot and see if they can identify anything for us.

Will report back!