How to add a talos worker that runs completely in RAM?

[nickjanssen]

Hello, I have nodes with a lot of RAM (256GB) and I would like to run worker nodes completely in RAM without any attached storage.

I'm running a lot of scraping tasks which are constantly throwing NodeDiskIOSaturation errors in Grafana because they are writing a lot of temporary files to disk. If I could somehow have the whole worker node, including the pods, run in RAM, it should become blazing fast and get rid of the errors.

Apparently there is a kernel option called CONFIG_BLK_DEV_RAM which would allow one to mount a RAM device right before booting into talos, but the talos kernel does not appear to enable it.
Would it be feasible for me to enable this option, rebuild the kernel, boot from PXE using the custom kernel and params, and try installing to a RAM device?

[smira]

We are looking towards supporting this flow, probably in 1.11 via #8367

[blackliner]

This would be an amazing feature! I have 512GB nodes with 2 NVMe drives, and I would love to

1. PXE boot talos (providing config through kernel args)
2. skip installation, just come up, configure, join the cluster
3. format the NVMe drives to raid0 and use as ephemeral/local storage for POD scratch space and container images
4. at the end of the day, kill the node and join it another workload/cluster/whatever

This way we would get rid of all state on these worker(!) nodes, unlocking full cattle mode!