With platforms like Scaleway that has an API and provide not only macOS infrastructure but a more generic one. We can add builtin functionality that will scale Orchard workers up and down based on the demand.
There are couple things to consider:
Orchard Auto-Scaler can configure workers automatically via SSH. Our Ansible roles are not that hard to replicate in scripts.
Auto-scaler should be aware of the 24 hours minimum limit of that we can rent hosts for.
I think the initial algorithm version should be super simple. There should be just couple configurations:
maximumWorkers/minimumWorkers - range to scale workers within.
upscaleSpeed - percentage to which upscale in one iteration. For example, if there are 10 unschedulable VMs in the queue and speed in 50% then the auto-scaler will create 3 nodes to fit 5 VMs.
upscaleInterval - duration to wait before the next upscale attempt.
Similar downscaleSpeed / downscaleInterval
Something else?
In case of auto-scaling we should also consider changing scheduling algorithm for maximum utilization -- prefer to schedule as many VMs on the same host as possible before going to the next one.
With platforms like Scaleway that has an API and provide not only macOS infrastructure but a more generic one. We can add builtin functionality that will scale Orchard workers up and down based on the demand.
There are couple things to consider:
I think the initial algorithm version should be super simple. There should be just couple configurations:
maximumWorkers/minimumWorkers- range to scale workers within.upscaleSpeed- percentage to which upscale in one iteration. For example, if there are 10 unschedulable VMs in the queue and speed in 50% then the auto-scaler will create 3 nodes to fit 5 VMs.upscaleInterval- duration to wait before the next upscale attempt.downscaleSpeed/downscaleIntervalIn case of auto-scaling we should also consider changing scheduling algorithm for maximum utilization -- prefer to schedule as many VMs on the same host as possible before going to the next one.