Nicholas Peshek
Nicholas Peshek
OK, I worked through this: Since a node's out, we're seeing sustained CPU usage across the cluster and shard 0 cannot always respond to node operations in a timely manner....
I was on Scylladb 4.6.2. I have since upgraded the cluster to a custom compiled version with the timeout extended and I can get replaces to start finally. The other...
Finally able to get back to this ticket. We experience the same issue with the long settle on every restart. For my cluster we’re usually seeing between 1500 and 3000...
@geobeau: Yep. That dropped our startup from 3000+ seconds to about 40.
I believe this is a bug in hwloc-libs. https://github.com/open-mpi/hwloc/commit/ff102fdfa95d911a4a1eac33c6cd80cdfe30445d To test if this is the problem: ``` HWLOC_COMPONENTS=-x86 htop ``` Worked well for me.