Last week in scylla-cluster-tests.git master (issue #36; 2024-01-27)

This brief report highlights some significant commits to the scylla-cluster-tests.git master from the past week, covering commits in the d311c6c1…cf79ff75 range.

During this period, 24 non-merge commits were made by 5 software engineers in test and 1 software engineer, with some notable changes:

We have resolved an issue regarding the restoration of monitoring when ports were occupied. Now, in such cases, a random port is chosen.

The upgrade of scylla-bench with gocql support for tablets was reverted due to a scylla-bench issue.

Support for RackAwareRoundRobinPolicy, introduced in scylladb/java-driver#227, was added to the cassandra-stress thread. This is automatically used when we set up multiple racks, allowing the loader to spread across multiple AZs. A new test was added to fully utilize this feature.

The network interface configuration has been refined to better test Scylla networking with multiple NIC/IP combinations. Now, it’s easier to define which Scylla endpoint is mapped to which interface and IP.

The subject of the cloud resources usage email has been reverted back to static. Please ensure you correctly assign the “important” label when your name is listed, so you don’t overlook any lingering resources.

Metrics from the sct-runner revealed an issue with high CPU usage that could lead to a ‘Bad packet length’ problem. To combat this, we updated the sct-runner instance and began to trim db log lines longer than 5K to prevent searches and regexes from taking too much time.

Finally, the new tablets_initial_scale_factor scylla.yaml option is now supported and can be used in the append_scylla_yaml test configuration.

Stay tuned for the next issue of last week in scylla-cluster-tests.git master!