This short report brings to light some interesting commits to scylla-cluster-tests.git master from the last week. Commits in the 52b00b03…7c0bca29 range are covered.
There were 61 non-merge commits from 13 authors in that period. Some notable commits:
libssh2 became the default SSH transport across all SCT components fixing the OCI create-runner flow.
Required-params validation got stricter: blank and whitespace-only strings are now treated as missing, not just None.
Scylla Manager testing moved its default to version 3.11 across the framework, and restore gained a version-aware keyspace filter that excludes the CDC cdc_generation_descriptions_v2 table on Scylla >= 2026.2, where it no longer exists but may still live in older backup snapshots.
Two new implementation plans landed. One reworks the pipeline labeling design to embed a pydantic TestMetadata model directly in test-case YAML with Argus integration and CI enforcement. The other proposes a one-time nemesis precheck() hook that evaluates static skip conditions once at disruption-list build time.
Reporting gained a perf-weekly-status-report skill that generates Gmail-compatible HTML summaries of ScyllaDB performance results from the Argus CLI.
OCI backend coverage expanded with OEL10 artifacts tests and an OCI analog of the tier2 multi-DC test.
Scylla-bench was bumped from 1.1.1 to 1.3.0 mostly bringing performance improvements.
latte advanced to 0.49.0 with custom workload metrics, parallel-scan support on the Alternator API, and a scylla-rust-driver bump to 1.7.0.
The Python driver was moved to 3.29.11 with a long-awaited fix for Bad file descriptor error.
Nemesis robustness improved on several fronts: AddRemoveDcNemesis stopped overriding rpc_address so multi-NIC configs bind CQL on the correct interface instead of timing out in wait_for_init, and _remove_node_add_node now always re-adds a node even if the best-effort repair fails, keeping the cluster at its original size.
See you in the next issue of last week in scylla-cluster-tests.git master!