The Scylla team is pleased to announce the release of Scylla 1.1.3, a bugfix release of the Scylla 1.1 stable branch. Release 1.1.3, like all past and future 1.x.y releases, is backward compatible and supports rolling upgrade. All the fixes below are also included in the stable 1.2.1 release. It is recommended to migrate to the 1.2 release as soon as possible.
The following bugs have been fixed:
- When migrating SSTables from Cassandra or a Scylla deployment with a different number of shards (cores), some SSTables are shared between more than one shard. This proved to be wasteful #1314. Scylla now splits SSTables between shards after startup.
- A node running out of disk space did not gracefully shutdown and did not notify the cluster it is not available #1353, #1311
- nodetool refresh did not invalidate the cache, causing some reads to ignore the new loaded data #1212
- Reads concurrent with table drop/truncate or ring movements might bring back stale data until node is restarted #1291
Asias He (3): streaming: Fix indention in do_send_mutations streaming: Switch log level to warn instead of error repair: Switch log level to warn instead of error Avi Kivity (1): Be more conservative when deciding when to shut down due to disk errors Nadav Har'El (3): Rewrite shared sstables soon after startup Rewrite shared sstables only after entire CF is read Rewriting shared sstables only after all shards loaded sstables Pekka Enberg (4): utils/exceptions: Whitelist EEXIST and ENOENT in should_stop_on_system_error() release: prepare for 1.1.3.rc1 service/storage_service: Make do_isolate_on_error() more robust release: prepare for 1.1.3 Raphael S. Carvalho (1): db: fix read consistency after refresh Tomasz Grabiec (2): row_cache: Implement clear() using invalidate() row_cache: Make stronger guarantees in clear/invalidate