Skip to content

v1.7.0

Compare
Choose a tag to compare
@timvisee timvisee released this 08 Dec 11:07
· 827 commits to master since this release
dbab7f8

Change log

Features 馃

  • #2796, #3124 - Add support for sparse vectors, also supporting SPLADE and BM25 datasets (docs)
  • #2790 - Add discovery search, explore and discover vectors in a defined context (docs)
  • #2807 - Add user defined sharding, partition data with a custom sharding distribution (docs)
  • #2432 - Add shard snapshot transfer, fast transfer of shards using snapshots including their index and quantized data (docs)
  • #3079 - Add Manhattan distance metric
  • #2979 - Add support for read-only API key (docs)

Improvements 馃寣

  • #2869 - Improve quality of HNSW index building on high CPU systems
  • #2768 - Add index for geo map payloads to improve search performance
  • #3013, #3026, #2942, #3103, #3054 - Improve stability of consensus on big high load clusters
  • #2931 - Improve search tail latencies with many parallel searches on high CPU systems
  • #2748, #2771 - Allow to configure the timeout for searches
  • #3067 - Improve group by performance by fetching vectors only once
  • #3076 - Improve group by performance by handling with_payloads more efficiently
  • #2803 - Improve performance for empty match-any filters
  • #3088 - Randomize shard placement to roughly get an even spread across all peers
  • #3155 - Make indexed_only more lenient
  • #3084 - Improve /readyz health check, only mark as ready when started node has caught up
  • #2855 - Add support for using API key with HTTP Bearer authentication
  • #2848 - Expose update rate limiter parameter in configuration
  • #3169 - Show actual bind address instead of localhost for Web UI in greeting message
  • #2846 - Clean up recovered snapshots from an URL instead of leaving them behind
  • #3036 - Clean up old RocksDB files more often to save disk space
  • #3054 - Improve WAL performance by implementing a faster checksum algorithm

Bug fixes 馃

  • #2933, #3039 - Fix potential divide by zero panic in numeric and geo index
  • #2888 - Fix panic if gRPC request has multiple validation errors
  • #3034 - Fix overvalidation on vector names in internal requests
  • #2904 - Fix best_score formula with just negative examples
  • #2927 - Fix duplicate numeric index entry for empty payload
  • #2969 - Fix potentially syncing local consensus state when there are still pending commits
  • #2895 - Fix custom TLS certificate not working when recovering snapshots from URL
  • #3093 - Fix panic when loading from RocksDB with corrupted tail
  • #3105 - Fix hard error for missing column family not found when flushing data
  • #3156 - Fix shard transfers potentially starting too early
  • #2960 - Fix potential memory overallocation in vector storage
  • #3079 - Fix binary quantization for euclidean distance
  • #2864 - Fix bad recall performance on payload indexes
  • #3174 - Fix response codes being too strict, breaking snapshot creation in Python client
  • #3073 - Fix NO_COLOR not having an effect on output logs
  • #3086 - Fix migration from local to distributed on systems with 1 or 2 CPU cores
  • #2921 - Fix storage and snapshot directory permission errors with Qdrant Docker image

Deprecation 馃

  • #2785 - Mark point, vector and index counts as approximate, we may eventually remove these

Dashboard 馃嵄


Thanks to @kaancfidan @keremnc @ava57r @Apmats @julianbraha @golangboy @ilkecan @ologbonowiwi @russcam @dzhao @joein @ffuugoo @IvanPleshkov @agourlay @coszio @timvisee for their contributions!