Last week in scylla-cluster-tests.git master (issue #30; 2023-11-18)

This short report brings to light some interesting commits to scylla-cluster-tests.git master from the last week. Commits in the 225293b9…80160f38 range are covered.

During this period, we had 47 non-merge commits from 10 authors. Here are some of the noteworthy commits:

As we gather kallsyms, which often provide essential information for existing issues, we have developed a tool to decode kernel callstacks. Although it’s not yet ready to be integrated with an automatic decoder, the content is prepared for use as a standalone script for now.

Due to InsufficientInstanceCapacity errors, certain tests will now be conducted in a different availability zone. For now, two tests have been relocated: large partition 200k pk and alternator short longevity. When adding new tests, please consider using an availability zone other than ‘a’ to minimize the likelihood of failure due to insufficient capacity.

SCT now supports multiple K8S clusters. Scylla nodes can use dc_idx to map to the appropriate k8s_cluster . This has enabled multi K8S cluster creation in EKS. While there’s more work needed for full multi-dc longevity tests on K8s, we’re significantly closer to achieving this.

Previously, installing Scylla from a repository would always select the latest version. With this change, you can select a specific version by adding the version after the colon in the scylla_repo URL. This will aid us in testing specific versions of Scylla in customer cases and reproducers.

When adding new nodes, we can now specify the instance_type. This means that whoever is calling the .add_nodes() can define a different instance type than what is configured in the test YAML file.

During node setup, if the package installation failed, it would cause the test to fail. We have now added retries for this process, enhancing stability. Furthermore, in most cases, package installation has been moved to the install_package node method to reduce code duplication and improve stability.

The large collection test now verifies the appropriate log messages and system.large_cells content.

Rolling upgrades have begun to use docker-based loaders, so we use a newer version of c-s and java-drivers.

We have removed support for deprecated distros from the code (Ubuntu 16, Ubuntu 18, Debian 8, Debian 9).

See you in the next issue of last week in scylla-cluster-tests.git master!