Kacper Rzetelski

> @rzetelskik @vponomaryov @slivne, here's a dtest that reproduces it with blocking gossip traffic TO the node being > decommissioned > > https://gist.github.com/fruch/7487febf59afd60a7972a08e4ef44daf @fruch does the decommissioning actually succeed? Because...

@fruch to further specify my comment above - I don't think it's enough to introduce such iptables rule to reproduce this exact issue. I don't have enough expertise to say...

> I don't understand what do you mean by nodes ops commands ? you are talking about repair-based-operations ? see https://github.com/scylladb/scylladb/blob/afa7960926f15566029d5e0ed20a00a15da70b28/idl/partition_checksum.idl.hh#L74 > The decommission did succeed, and I've blocked the...

> I was blocking only packet targeting the gossip port on the node being decommissioned, meaning all other communication from that node to the other node is working. so nodes...

> I'm blocking node1, node2 --> 7000 node3 > ... > so the node_ops command can go from node3 to node1 and node2, Yes, node_ops can go from node3 to...

@fruch you're right. And that also leads me to believe my initial hypothesis ("[...] the node ops, used to broadcast the information about decommissioning of node 2 propagate both ways...

https://github.com/scylladb/scylladb/issues/11302 has now been closed as completed in https://github.com/scylladb/scylladb/commit/58c65954b8cedc498678e97027885aea1cad24f3

@tnozicka why would we want to initiate a rolling restart whenever a new image is published using an existing tag? I've consulted this behaviour with @zimnx and he expressed concern...

Now awaiting https://github.com/scylladb/scylla-operator-images/pull/1 to land

1. Since [scylladb/k8s-local-volume-provisioner](https://github.com/scylladb/k8s-local-volume-provisioner) is on the way, I suggest we wait for it instead of using the static volume provisioner. 2. Minikube's `storage-provisioner` wasn't setting `VolumeNodeAffinity` and the logic in...