+1 In my previous stint, I had worked with Spot (https://spot.io/) as one of our vendors. Absolutely great product, amazing customer support and ability to take feature requests, or otherwise address our pain points quickly and effectively. - Source: Hacker News / about 1 month ago

FWIW, I am also a big spot.io fan for our workload. During the holidays I run 30-50% spot instances and run 100% spot most of the year. Source: over 1 year ago

Also, you definitely should look into Reservations, and (sale pitch coming) Spot can help you manage those. Source: over 1 year ago

All of this is on spot-instances. We used spot.io (I believe the product is called "Ocean") and they basically took care of all the backend logic to make spot-instances available for the ECS cluster. Source: about 2 years ago

Does cloud provider matter? I would say/think so. Not just cloud provider, but further more, how you set it up, which begets cloud provider. Are you setting it up with only the aws cli? Or did you terraform it? Maybe you chose a particular terraform module or maybe you used eksctl. Maybe you used kops or kubeadm. All these things matter when you get to cluster autoscaling, tainting particular node types to... Source: almost 3 years ago

There are tools CastAI (https://cast.ai/) and KubeCost (https://www.kubecost.com/) which helps you get these values. I haven't tried it personally, but they are promising. There are other tools as well. One of the approach OP suggested, monitor the values over a period of time to determine the right requests value, is really good one. I would modify it a bit. Generally take the p95 value for requests and 1.5-2x... Source: 11 months ago

Not sire if this helps but someone just showed me this free tool that looks at cost for Kube https://cast.ai/. Source: about 1 year ago

Curious about what about cast.ai sets it apart for you? I went with spot because it is owned by a big company and knew it wasn't going to disappear. I think cast was still in invite only mode, as well. Source: over 1 year ago

I found that cast.ai seems to have this functionality but am wondering if there is a free option. Also pursuing gMaestro but they're not available on arm64 yet. Source: over 1 year ago

If you're using Kubernetes, CAST AI is the fastest way to significantly reduce your compute bill and keep it there. It manages compute capacity automatically and has dedicated support to get you started even faster. The best part - Kubernetes cost monitoring and security insights are free. Source: over 1 year ago