This short report brings to light some interesting commits to scylla-cluster-tests.git master from the 2 last weeks. Commits in the 1d62cf45…496bb812 range are covered.
There were 60 non-merge commits from 13 authors in that period. Some notable commits:
Stress commands can now be written as Jinja templates, a strict-mode replacement for manual string substitution in test configs. Longevity tests gained an effective_disk_size_bytes variable, so a command can scale its data set to the disk size of the target instance, and configs can define shared variables reused across commands. The elasticity-90-percent test is the first case converted to the templated form.
SCT gained NVMe device diagnostics for DB nodes parses SMART logs, error logs and self-test results, and filters cloud-attached volumes out of the device list.
GCE nodes can now run with multiple network interfaces, so tests that put client and cluster traffic on separate interfaces work on the -gce pipelines without config edits.
Extraction of nemeses into sdcm/nemesis/monkey/ picked up the node isolation and MV/SI disruptions.
Four LWT gradual-throughput performance tests compare tablets against vnodes for a heavy (~1.3KB rows) and a light workload profile.
Tier 1 jobs now run with ScyllaDB automatic repair on a one-hour period through the new configurations/auto-repair-1h.yaml. ScyllaDB 2026.1 schedules and runs repairs on its own, without Scylla Manager and without a nemesis driving them, and nothing in SCT exercised that so far.
Tier1 and Sanity longevity tests dropped their explicit STCS override. A dedicated longevity-100gb-4h-stcs.yaml keeps STCS coverage.
Minicloud jobs can run on a local KVM Jenkins agent with no sct-runner at all. An OS rolling-upgrade job on stock Ubuntu 24.04 was ported to minicloud.
The default Scylla Manager version moved to 3.12 across the framework, which brings the event-based hold retention-lock mode for backups and new metrics for holds and versioned files.
A scylla-doctor bump to 1.13 brings a warning for unused NVMe devices, Chrony time-sync collectors and analyzers, and collection of Scylla Manager Agent logs.
Docs updated: Kernel panic detection and The Docker backend’s SSL certificate handling and tar-stream transfer.
See you in the next issue of last week in scylla-cluster-tests.git master!