This page permanently redirects to gemini://rainywhile.net/gemlog/2023-03-24-re-oracle-cloud-always-free.gmi.

Oracle Cloud Always Free "Idle" Compute Instance Reclamation

=> I previously wrote about my use of Oracle Cloud's seemingly "too good to be true" Always Free tier here

This Tuesday I received an email from Oracle about their new policy of automatically stopping instances in the Always Free tier that have been idle for more than 7 days. The email stated that my (one and only) compute instance fell under their definition of idle.

Here is that definition from Oracle's documentation:

Idle Always Free compute instances may be reclaimed by Oracle. Oracle will deem virtual machine and bare metal compute instances as idle if, during a 7-day period, the following are true:

• CPU utilization for the 95th percentile is less than 10%

• Network utilization is less than 10%

• Memory utilization is less than 10% (applies to A1 shapes only)

=> 2023-09-13 Update: these limits have now changed to 20%

All I'm running 99% of the time on this server is a handful of (very) low-traffic gemini capsules and websites, which means I don't hit any of these thresholds. If you're just running a gemini capsule on one of their Arm A1 VMs then you might be at risk too (unless your capsule is much more popular than mine, or your server is very inefficient!)

The documentation isn't clear about whether you need to meet all three points to be considered non-idle - for now I'm hoping that just one is enough. To that end, I've written a simple C program that runs on startup, allocates (and writes to) a 2GB byte array and then sleeps forever (waking up every hour to write a byte to a random array index; just to make sure the memory stays 'in use' according to any monitoring Oracle might do. I'm pretty sure this step isn't necessary). This is just enough to push my baseline RAM usage to about 11%.

Because I only noticed the email today, I've temporarily set a few more instances of this program running to ensure that the 7-day average hits 10% by next Tuesday. Fingers crossed that I don't actually need any of that wasted RAM in the meantime!

If this capsule goes offline in the next few days then the above clearly wasn't enough. Stopped instances should be able to be restarted, so it should just mean a bit of downtime, followed by 7 days to figure out a better strategy.

It's also worth noting this point from Oracle's email, which might be my only option (although I would need to read a bit more on the configuration options to make sure there's no risk of a surprise massive bill):

You can keep idle compute instances from being stopped by converting your account to Pay As You Go (PAYG). With PAYG, you will not be charged as long as your usage for all OCI resources remains within the Always Free limits.

I can't complain too much about this change - beggars can't be choosers after all - but thought it was worth mentioning in case it affects any other Gemini capsule owners, as well as providing advance notice of potential downtime on this capsule.