1.0.17
Breaking Changes📦 chroma
⚠ 1 breaking✨ 12 features🐛 5 fixes🔧 4 symbols
Summary
Version 1.0.17 focuses on enhancing observability with new metrics and tracing capabilities, alongside performance improvements like reduced peak memory usage in the compactor. A key change involves removing the 'id' field from the GetCollectionByResourceNameRequest structure.
⚠️ Breaking Changes
- The structure of the request for getting a collection by resource name has changed: the 'id' field has been removed from 'GetCollectionByResourceNameRequest'. If you were manually constructing this request, ensure you remove the 'id' field.
Migration Steps
- If you were using the 'GetCollectionByResourceNameRequest' API, ensure you remove any 'id' field you were previously sending in that request structure.
✨ New Features
- Added metrics for span and metric exporters (though this was later reverted, see bug fixes/reverts).
- Added a metric for the number of aborted tasks in the dispatcher.
- The 'take_result_channel()' method now returns an Option type.
- Implemented tracing for errors when collections are soft-deleted.
- Created the 'update_tenant' API.
- Added support for scorecarding fork collections by collection or tenant.
- Added an index on collections using 'created_at' as the sort key.
- Added a flag to the Dockerfile to enable the use of AVX512 instructions.
- Added support for WordPress (AI Engine Pro) integration.
- Made the number of concurrent block flushes configurable.
- Implemented span uploading.
- Added documentation for collection forking.
🐛 Bug Fixes
- Removed the 'id' field from 'GetCollectionByResourceNameRequest' to fix an issue.
- Reverted the addition of metrics for span & metric exporters (PR #5236).
- Removed two spans identified as spammy from RLS (Row-Level Security).
- Ensured errors are not traced if 'should_trace_error' is false.
- Reduced the recall threshold in 'test_add' to 95 for stability.
🔧 Affected Symbols
GetCollectionByResourceNameRequesttake_result_channel()update_tenant APItest_add