[ceejayoz]

> That being said, do they dump your instance if they reboot the physical server? I was under the impression that "Oh we want to upgrade they hypervisor kernel, your instance is toast", was fairgame/default on Amazon.

Nope. In most cases, you get two weeks warning before a physical host is decommissioned. I've gotten as little as 24 hours (guessing it was a more significant than usual hardware issue).

When that happens, you can just stop the instance, then start it again. That moves it onto a new physical host. If you don't do it in the 14 days, Amazon does it automatically. No loss of data, and just a few minutes of downtime if your startup scripts are OK.

[jcampbell1]

You mean they copy the ephemeral disk over? Really?

Amazon should really advertise this, if that is the case. I was under the impression that Amazon couldn't be relied on as a VPS.

[aluskuiuc]

If you get a notification that your instance is going to get terminated - then yes, you can stop/start, but then you will lose ephemeral disks (but not EBS volumes). If you get a notification that your instance is just going to get rebooted, then nothing's going to get lost.

As I said above - the terminology can get hairy pretty fast. It's valuable to take stock of what you really need persisted, and what use can be made of 'scratch disks'. Local ephemerals are fast and cheap (included in the price of any instance). They come with the operational overhead of needing to rsync data off of them yourself if you want to retire that instance however.

[ceejayoz]

No, not ephemeral disks. They're typically used for temp files, caches, etc., or servers you can afford to lose. EBS volumes are persistent and work just fine between stop/starts.