I can’t find much in the way of documentation on suspend / auto suspend. The most recent thing I see is this discussion.
I’m interested to know if the limits are documented anywhere or have been updated since the initial announcement post?
I’m curious to try it out but our machines are 8gb memory and 2-4cpu, which exceeds the limits initially documented.
Additionally, i’m curious about some other details:
Are there limits on how many machines can be suspended?
How long does a suspended machine live until it’s reclaimed (assume it’s not around forever if not used)?
We are dynamically spinning up a sandbox machine for our users. Given the nature of this, we don’t know if/when they’ll come back. Ideally, it boots instantly when they do come back. I’m not sure if there would be problems with keeping many machines around in a suspended state.
The post you linked to contains the latest information on auto-suspend; for your use case, with an 8GB machine it won’t work as the limit is still 2GB.
There is no actual limit on how many machines may be suspended, and in principle they could stay around forever in that state. Note however that the post you linked to does mention that resume-from-suspend behavior is not guaranteed, meaning that you must be prepared for suspended machines to not resume and do a cold-boot instead, as there are multiple reasons and scenarios in which the system won’t be able to resume a suspended machine and will do a normal start instead.
In general, suspend is a good fit for something like web service machines which might constantly be stopping and starting according to load and for whom it’s beneficial to start more quickly (read: resume from suspend instead of cold-booting) but that can still work acceptably if there’s a cold boot. It’s probably not going to be a great fit for “keep the machine stopped for long periods of time and wake up instantly when needed” for the reasons I covered above.
So the prudent thing to do is to consider suspend a best-effort optimization, but not make it a core behavior of your offering. Make sure that your experience with a cold boot is still reasonable for your users’ sandbox machines.