●Change8

1.5.9

📦 chromaView on GitHub →
✨ 25 features🐛 6 fixes🔧 5 symbols

Summary

Version 1.5.9 introduces significant enhancements around sharded collection rebuilds, maxscore indexing capabilities, and the scaffolding of a new foundation CLI project. This release also includes several bug fixes related to data integrity during deletion and garbage collection.

Migration Steps

  1. If relying on specific internal testing configurations, note that Tilt fault injection CLI was added.
  2. If using sharded collections, note that Group by support is now available.

✨ New Features

  • Enabled rebuilds for sharded collections.
  • Added support for Group by functionality with sharding.
  • Added workflow to build and publish service container images.
  • Updated language around Chroma Cloud to be more representative.
  • Added change stream to collection compaction cursors.
  • Named and sized all worker threads.
  • Added SparsePostingBlock.
  • Added basic maxscore writer/reader.
  • Added maxscore lazy cursor.
  • Added SIMD optimizations for maxscore calculations.
  • Added benchmarking for maxscore.
  • Added maxscore option in schema.
  • Added maxscore index to metadata segment.
  • Wired maxscore reader into search functionality.
  • Added MCMR support for log Garbage Collection (GC).
  • Parameterized multi-region configuration testing.
  • Refactored offset_id to be a mutable AtomicU32 in record.
  • Added spanner-cli wrapper binary.
  • Deferred Spanner initialization to first use in the log service.
  • Added read-only backend failover support in the rust-client.
  • Implemented GC for empty MCMR collections.
  • Scaffolded foundation CLI project.
  • Added login, logout, and whoami commands to the foundation CLI.
  • Added client header to Gemini embedding functions.
  • Added documentation for file-upload sync API, AWS credential reuse, and EU region.

🐛 Bug Fixes

  • Fixed issue where database name was incorrectly renamed on soft delete.
  • Fixed flaky indexing progress test by simplifying it.
  • Used correct version file per collection in ancestor walk for garbage collection.
  • Preserved legacy HNSW metadata keys in sysdb.
  • Rejected NaN/Infinity values in base64-encoded embeddings.
  • Verified file path for compaction operations.

Affected Symbols

collection_compaction_cursorshnswoffset_idAtomicU32Gemini embedding functions