At one point @millions-microphone-3535 you asked if the nodes were under pressure, they def aren't. 9 nodes each with 256 cores (arm) and a Ti of Ram. There's only 3 VMs in the entire cluster (it's new). I think I have another cluster with older Intel Procs I might be able to reproduce this on as well. It's just more difficult as it doesn't trigger every time (only 1 in 6 this last time) but let me know if that's something your team wants.

yeah, i was curious about the "killing the pod manually triggered the drain for the other two nodes" scenario

Best I could figure is that sometimes there's duplicate engines running so I maybe it was blocking the engines somehow because the stuck pdb also had the mount ? idk.

let me take a look at the support bundle