This message was deleted.

I use

--unsafe-no-fsync=true

on my test clusters, but that’s definitely not recommended for production use. I have also seen people tinker with the heartbeat interval, but I don’t have any good guidance on that. SSD and NVME is so cheap these days I’m not sure why anyone wouldn’t just do that.

Because we have hardware already deployed at customer sites.

@bright-lifeguard-9803 did you find anything to tune that down?

@icy-secretary-33916 we moved to a ramdisk for etcd. We might back peddle on that because of the complexity it adds. We're testing this config in the lab now...

cat /etc/rancher/rke2/config.yaml.d/99-etcd.yaml
etcd-arg:
    - unsafe-no-fsync=true

@bright-lifeguard-9803 i changed to ssfd siscs. did you perhaps had any isue ith high disk aktivity or the etcd processes writing about 4GB/s per day ?