This short report brings to light some interesting commits to scylla-cluster-tests.git master from the last week. Commits in the a75fe17d…76c762cb range are covered.
There were 59 non-merge commits from 10 authors in that period. Some notable commits:
The Azure backend can now provision several network interfaces per VM, with the NIC layout derived from scylla_network_config. Each interface can also carry an IPv6 address, reachable from outside the VNet when the test config marks it public.
A new Azure multi-DC schema and topology changes 12h pipeline runs with every Scylla address on IPv6. It is the first job to exercise the Azure multi-NIC support end to end.
The instance_provision option accepts a new auto value, resolved at config load from the effective test duration. It picks spot at or below spot_max_test_duration, 12 hours by default, and on_demand above it.
AWS spot provisioning orders candidate zones and regions by spot placement scores, and GCE ranks them through the Capacity Advisor API. GCE zone fallback also extends to multi-node batches and to on-demand.
The vnode-to-tablet migration longevity test gained two rollback scenarios. One rolls a single node back mid-flight and re-upgrades it before finalization, and the other rolls the whole cluster back to vnodes.
Master rolling upgrades run weekly against minicloud through a new minicloud build parameter on the existing jobs. The master-only twins fire every Friday, and the real-cloud alternation keeps ami-arm and azure-image.
The cql-stress image was moved from 0.3.2 to 0.5.0. The releases add leader-aware routing for strongly consistent Raft-per-tablet tables, default keyspace creation to NetworkTopologyStrategy, and add shardAwarePortRange and tcpReuseAddress to the -mode option.
Latte was bumped to 0.53.0. The release adds the shard-aware-port-range and tcp-reuse-address connection options, and fixes the load command running zero cycles.
See you in the next issue of last week in scylla-cluster-tests.git master!