Weaviate v1.38.12 introduces fixes for async replication and multi-node aggregation

The release of Weaviate v1.38.12, as detailed in the project's GitHub release notes, focuses on performance improvements and bug fixes without introducing breaking changes or new features. Key updates include fixes to runtime configuration, HNSW cache reads, and LSMKV write-ahead logging. Performance enhancements target async replication, shard routing, and zstd encoder concurrency. Additionally, multi-node aggregation issues were resolved, addressing JSON round-trip errors that caused panics for mean, median, and mode calculations. Backup and restore processes were refined, including routing RBAC role restoration through RAFT and backing up index.db from inactive dynamic tenants.

Operators should carefully review the fixes related to shard handling and replication performance. The atomic shard status check-and-set mechanism and changes to async replication batching could alter how resources are scanned and routed in multi-node setups. Systems relying on lazy shard loading or dynamic tenant backups may experience behavioral differences. Testing these scenarios in a staging environment before upgrading is advisable to ensure compatibility with existing configurations.