Release Notes
These docs cover SDK 3.0.10, tested with Milvus 3.0.6. Check Migration & Compatibility for server requirements.
3.0.10
Section titled “3.0.10”count()returns the original server error status withdata: 0when a query fails, instead of throwing aTypeError. Checkstatus.error_codebefore using the count. Fixes #668.flushSync()handles streaming flush checkpoints when segment IDs are not yet available, preserves database and timeout options while polling, and returns server failures.HttpClientinvokes the built-infetchwith the global receiver required by edge runtimes.- Updated and verified documentation examples, search metadata, Markdown mirrors, and LLM documentation bundles.
The target server remains Milvus 3.0.6. These fixes introduce no new server version requirements.
- Function chains extend gRPC search, hybrid search, and query. REST search and hybrid search support
functionChains. - Search aggregation is available through gRPC and HttpClient. REST also supports search by ID, grouping, ordering, and expression parameters for query and search.
- Partial upserts can replace individual field paths through gRPC and REST.
- Portable snapshot workflows use
exportSnapshot,getExportSnapshotState, andrestoreExternalSnapshot. Both clients support these methods. MilvusClientadds replication configuration and checkpoint APIs, duplexcreateReplicateStream, and server-streamingdumpMessagesfor WAL messages.MilvusClientaddscomputePhraseMatchSlop,batchUpdateManifest, and proxy and telemetry administration RPCs.- HttpClient adds collection schema changes and truncation, external refresh, snapshot jobs, manual import commit/abort, resource and privilege groups, role management, analyzers, file resources, segment metadata, and quota metrics for Milvus 3.0.6. The REST reference covers all 113 server routes, including compatibility aliases and the two legacy standalone function routes.
- StructArray schemas support Array sub-fields. Partial upserts preserve the StructArray’s type and schema.
- The SDK pins the Milvus 3.0.6 protobuf definitions. The consolidated SDK test suite runs on pull requests.
- Fixed VarChar primary-key escaping, count query options, typed vector insertion with dynamic fields, and score formatting for exponential notation.
- Fixed sparse vector serialization, gRPC pool shutdown, and several dependency issues. CI tests against Milvus 3.0.6.
- Aligned the documented Bloom and Roaring expressions with
membership_match. - Updated YAML dependencies.
- Bloom and Roaring filters: build binary membership parameters locally with
buildBloomFilter,BloomFilterBuilder,buildRoaringBitmapandRoaringBitmapBuilder. - Client telemetry: configure collection and heartbeat reporting, inspect effective settings, and handle server commands. Telemetry is enabled by default.
- Index enums:
HNSW_SQ,HNSW_PQ,HNSW_PRQ,AISAQ,NGRAMandFMINDEXnow haveIndexTypeconstants. This exposes names already accepted as strings; availability depends on the server. - Dependency updates: Parquet writer dependency updated to 1.8.9; js-yaml updated. Text embedding tests switched provider; this does not change the public SDK API.
3.0.4 additions covered here
Section titled “3.0.4 additions covered here”- Text fields:
DataType.Text, without VarChar’smax_lengthsetting; requires StorageV3. - Function-field backfill: adding a BM25 or MinHash output field starts asynchronous server backfill; monitor schema and index readiness.
- External collection field additions preserve
external_fieldmappings.