Cezar Moise

icon location Bucharest, Romania

Results 98 comments of Cezar Moise

Add testcase for doubling cluster with smaller instances while keeping 90% utilization

Updated results with run using racks.

Add testcase that adds smaller instance while having 90% storage utilization

# Results https://argus.scylladb.com/tests/scylla-cluster-tests/984afbe7-1e5e-4f94-8343-d519526b93f0 Run date: 2025-03-02 Scylla version: 2025.2.0~dev-20250301.0343235aa269 Initial cluster: 3 racks with one i4i.4xlarge per rack Write to 90%, RF=3, initial tablets=1024 Run mixed workload at around 50%...

Add testcase that adds smaller instance while having 90% storage utilization

> What flavours did you use? i4i.xlarge initial then add i4i.large

Add testcase that adds smaller instance while having 90% storage utilization

> 1. I think something is wrong with the description - `Create 3 node cluster with rf=4.` - how can we set RF=4 if we have just 3 nodes? Swapped...

Add testcase that adds smaller instance while having 90% storage utilization

Running into out of space issues when decommissioning: https://github.com/scylladb/scylladb/issues/21783

Add testcase that adds smaller instance while having 90% storage utilization

> > Running into out of space issues when decommissioning: [scylladb/scylladb#21783](https://github.com/scylladb/scylladb/issues/21783) > > This testcase shouldnt do any decommissioning, right? Yes, that was using the existing GrowShrink Nemesis. Wrote a...

Add testcase that adds smaller instance while having 90% storage utilization

Updated results with 50% load

Add testcase that adds smaller instance while having 90% storage utilization

Updated resuls with 50% load, adding nodes in parallel.

Add testcase that adds smaller instance while having 90% storage utilization

Added new run with 2025.1. Looks like a performance regression, but it may be related to other issues I encountered **Adding New Nodes** | Cycle | Scylla Build | Latency...

Add testcase that adds smaller instance while having 90% storage utilization

Updated with a new run. All ok now. For this I reduced the `threads` of the stress commands that fill the cluster from 200 to 150 each. ![Image](https://github.com/user-attachments/assets/c46fcd19-f799-4f3a-a7ae-bb59af619fa2)