This short report brings to light some interesting commits to scylla-cluster-tests.git master from the last week. Commits in the 4f94cd24…47ed698f range are covered.
There were 67 non-merge commits from 15 authors in that period. Some notable commits:
The nemesis system received a substantial refactoring. The disrupt_method_wrapper was removed entirely, with all logic consolidated into execute_nemesis after a series of extraction, simplification, and target pool handling improvements. A new nemesis hooks mechanism for ClusterTester enables running code before and after each nemesis without hardcoding it in nemesis code. Comprehensive unit tests and a nemesis developer guide with a writing-nemesis skill were added alongside the rework.
Cluster health checks can now run in parallel using a ThreadPoolExecutor, with the number of workers configurable via the new cluster_health_check_parallel_workers parameter (default: 5).
Performance regression testing is being migrated to ARM64. New i8g.4xlarge Jenkins jobs were created for both vnodes and tablets, consolidating all four subtests into a single job.
A new error injection REST API client adds support for enabling, disabling, and inspecting error injections on dev builds — useful for testing failure scenarios that require controlled fault injection.
Jenkins CI gained a structured test failure summary that parses JUnit XML results and pre-commit logs, then posts a formatted comment on the PR.
Manager tests now support configurable backup methods, and the majority of jobs on Scylla 2025.3+ were switched to native backup instead of rclone.
The Argus client was updated to 0.15.13, adding stress command timestamp support and POST retries.
The 5000-tables-latte scale tests were promoted from tier2 to tier1 for both tablets and vnodes pipelines.
See you in the next issue of last week in scylla-cluster-tests.git master!