1.14.1

Change log

Improvements

• https://github.com/qdrant/qdrant/pull/6458 - Speed up WAL-delta shard transfer significantly via batching and more careful synchronization
• https://github.com/qdrant/qdrant/pull/6462 - Improve GPU indexing speed for payload-related HNSW links, reuse GPU resources
• https://github.com/qdrant/qdrant/pull/6444, https://github.com/qdrant/qdrant/pull/6495, https://github.com/qdrant/qdrant/pull/6503, https://github.com/qdrant/qdrant/pull/6587 - Greatly improve payload index load time by replacing RocksDB with mmaps as persistence layer
• https://github.com/qdrant/qdrant/pull/6446 - Make GridStore flush slightly faster, update gaps in batch
• https://github.com/qdrant/qdrant/pull/6510 - Optimize GridStore reads, tell the kernel we use random reads
• https://github.com/qdrant/qdrant/pull/6502 - Better IO/CPU resource scheduling for optimizers
• https://github.com/qdrant/qdrant/pull/6487 - Batch IO when merging segments in optimizer, improve indexing speed
• https://github.com/qdrant/qdrant/pull/6513 - Buffer IO while reading files
• https://github.com/qdrant/qdrant/pull/6501 - Speed up HNSW construction by improving heuristics computation
• https://github.com/qdrant/qdrant/pull/6504 - Improve performance of isEmpty and !isNull conditions with a specialized index
• https://github.com/qdrant/qdrant/pull/6536 - Limit S3 upload parallelism, prevent network errors on high CPU machines
• https://github.com/qdrant/qdrant/pull/6548 - Add config option to limit number of collections

Bug fixes

• https://github.com/qdrant/qdrant/pull/6403 - Fix potential inconsistency during index creation, make it atomic
• https://github.com/qdrant/qdrant/pull/6478 - Address performance degradation for large batch queries with a lot of segments
• https://github.com/qdrant/qdrant/pull/6445 - Fix not freeing all GridStore blocks if there are many updates
• https://github.com/qdrant/qdrant/pull/6363 - Fix strict mode unindexed_filtering_retrieve not being enforced in group_by queries
• https://github.com/qdrant/qdrant/pull/6498 - Fix strict mode search_allow_exact preventing full-scan in segments without main HNSW graph
• https://github.com/qdrant/qdrant/pull/6505 - Abort shard transfers related to peer being force removed
• https://github.com/qdrant/qdrant/pull/6533 - Always overwrite existing payload on upsert, fix REST vs gRPC discrepancy
• https://github.com/qdrant/qdrant/pull/6549 - Correctly cancel on optimization failure, keep collection consistent on red status error
• https://github.com/qdrant/qdrant/pull/6560 - Use HTTP 400 instead of 403 for strict mode errors, prevent web-ui from constantly asking new API key
• https://github.com/qdrant/qdrant/pull/6582 - Register payload indices in consensus on snapshot recovery